← Briefing archive

weekly · 2026-11-09 — 2026-11-15

Higher foundry activity meets equipment constraints as chip financing advances

SMIC reports higher aggregate utilization while warning of equipment delays. SMEE describes dry-lithography progress, and Alibaba discloses T-Head's proposed listing. These developments strengthen the evidence for industrial activity without establishing advanced-node self-sufficiency or frontier workload parity. No dated Observatory forecast changes.

Published 2026-11-16 · Evidence through 2026-11-15 · Completed period

More foundry activity does not reveal advanced-AI-chip output

SMIC's November 12 unaudited quarterly filing reports utilization of 97.3%, up from 93.7% in Q2. Quarter-end monthly capacity was 1,116,750 eight-inch-equivalent standard-logic wafers, with quarterly sales of 3,033,683 on the same equivalent basis. These figures describe the quarter ended September 30, not November production. They support higher aggregate foundry activity but do not isolate advanced-node capacity, good large AI dies, process yield or domestic-tool share. The sovereignty question therefore remains about qualified output and replenishable inputs, rather than utilization alone.

SMIC / HKEX · SMIC reports 97.3% Q3 utilization and higher wafer capacity (source)

Equipment delivery is a constraint on expansion timing

At the November 13 earnings call, co-CEO Zhao Haijun reportedly cited equipment delivery times of 16–24 months and possible disruption to capital-spending execution, while saying the year's expansion plan remained intact. He expected tight wafer supply in 2027 and AI-memory supply-demand balance around mid-to-late 2028, with compute and supporting chips still constrained. These are management statements and forecasts, not measured future capacity. They highlight delivery schedules and customer qualification as tests of expansion plans; the report does not identify the delayed tools or quantify their effect on advanced-node output.

STAR Market Daily (Guo Hui), republished by Eastmoney · SMIC earnings call — equipment delivery delays and AI supply outlook (source)

Dry-lithography claims need an operating denominator

November 9 CIIE reporting describes SMEE exhibiting SSC800 and SSB800 platform models, covering KrF and I-line dry lithography. It attributes to SMEE claims of over 90% domestic content in key components, more than 10 million cumulative wafer passes supported and an effective operating rate above 92%. These are vendor claims conveyed by reporting, without identified customers, installed-tool count, measurement period, uptime methodology or the domestic-content denominator. Repeated wafer processing is not a count of unique good wafers. The evidence concerns dry platforms and cannot establish advanced immersion or EUV qualification. Independently documented customer acceptance and sustained throughput remain the next milestones.

Guancha (Sina republication) · SMEE dry-lithography models and attributed operating claims at CIIE (source)

T-Head opens a financing route, not a funded capacity commitment

Alibaba's November 10 disclosure says T-Head confidentially submitted its Hong Kong listing application on September 11 after exchange clearance to proceed with the proposed spin-off. Publication and application dates are distinct. Offering size, structure and Alibaba's eventual stake were not finalized; Alibaba expected T-Head to remain a subsidiary following completion. A separate listing could provide access to capital and clearer standalone valuation, but the application does not establish completed financing, proceeds, manufacturing expansion or a change in fabrication and memory dependence. Approval, offering terms and use of proceeds are the consequential next checks.

Alibaba / SEC · Alibaba discloses confidential T-Head listing application (source)

Forecasts and benchmark judgments remain unchanged

No dated Observatory forecast changes. The evidence does not resolve the manufacturing, memory and operating-performance uncertainties behind the existing judgments. Watch for accepted lithography equipment, disclosed advanced-node yields and good-die output, actual financing and reproducible training time-to-quality or inference throughput/latency measurements. The week’s recorded reviews did not verify a new matched contemporary Chinese frontier end-to-end benchmark. This is a verification limit, not evidence that none exists. Post-period context: November 16 repository-head checks found unchanged MLPerf Training v6.0 and Inference v6.1 heads, without an exhaustive artifact audit. The retrieved InferenceX page still carried an unofficial-deployment warning, so it was not accepted as authoritative new evidence. These publication-day checks do not reconstruct the exact period-end leaderboard.

SemiAnalysis · Continuous inference benchmark dashboard (source)MLCommons · MLPerf Inference v6.1 closed submissions (source)MLCommons · MLPerf Training v6.0 submissions (source)