Semiconductors & Chips

China's semiconductor self-sufficiency campaign is steady but stuck below the leading-edge ceiling: central ministries are issuing loftier goal statements, while provinces and companies continue to carry the actual industrial policy. The new documents in this period add no turning point—no new Big Fund, no export-control response, no revised technology roadmap—but they reaffirm the pattern of localized capital assembly and mature-node scale-up, with design localization progressing faster than manufacturing independence.

The two most prominent central-level signals this week are both goal-setting rather than programmatic. On Aug 26, MIIT planning chief Yao Jun promised "super-conventional measures" (超常规举措) for integrated circuits and machine tools to reach "decisive breakthroughs" ¹; two days later, the NDRC said it is accelerating major State-backed projects and building a 2027 pipeline under a "two priorities" initiative aimed at major national strategies and security capacity ². Neither names a funding vehicle, a procurement mandate, or a roadmap. The NDRC framing is broad enough to channel state investment into strategic tech, but without a chip-specific annex it is best read as a general infrastructure stimulus, not a semiconductor policy package.

The operative policy layer remains provincial and local. Guizhou reports 81.2% of its intelligent-computing capacity runs on domestic chips, with total compute above 176 EFLOPS and AI-type computing more than 90% ³. Hubei party secretary Guan Zhi'ou again urged Dinglong Semiconductor Materials to build a complete R&D-production-supply-service chain ⁴. Liaoning held a "patient capital" matchmaking event on Aug 27, with semiconductors listed among emerging sectors ⁵, and Tuojing Technology topped out its Shenyang equipment base—an ~18-billion-yuan project planned for 1,000+ thin-film deposition tools a year ⁶. Even Chongqing's central-enterprise cooperation meeting, with 193 projects worth over 750 billion yuan, is framed around "national strategies" and local manufacturing clusters rather than a dedicated chip program ⁷. The center articulates goals; provinces and cities assemble capital.

Foundry results confirm the mature-node story. SMIC's Q2 sales crossed $3 billion, up 36.1% YoY, with China-region revenue at 90.2% and net profit +261.7%; Hua Hong set a record $717.5 million at 102.8% utilization ⁸. Both raised Q3 guidance, and consumer electronics remains the largest end-market for both—meaning AI demand is inflating mature fabs as much as it is advancing leading-edge logic. These figures are consistent with state-backed demand soaking up capacity at 28nm and above, not at the cutting edge.

Design localization is real but manufacturing dependence persists. Xiaomi's Aug 24 launch of three Xuanjie chips—the O3 SoC, O100 AI accelerator, and D100 AD chip—shows cadence: the O3 starts shipping in September with the Xiaomi 18 Fold and is the first flagship with domestic CXMT LPDDR6 ⁹. But the logic die's leading-edge process still relies on external foundry capacity, so the design win does not change the lithography constraint. MiniMax's first-half results, with B2B revenue up 703% to $73.9 million, show Chinese AI monetization is accelerating ¹⁰—which will keep demand for compute chips high, but that demand is increasingly served by domestic mature nodes and Nvidia imports elsewhere.

Equipment and materials deepen incrementally in the mature segment. Tuojing's products are already in about 100 domestic chip lines ⁶, and Hubei's Dinglong push underscores consumables localization ⁴. But Tianfu Communication's quarterly report shows the bottleneck persisting one level up: Q2 profit recovered while upstream optical chips remain constrained ¹¹. The cost environment cuts both ways: storage vendors' price increases are squeezing handset margins, and global equity markets are jittery—memory stocks fell on Aug 24, the Philadelphia Semiconductor Index dropped 2.70%, and Nvidia logged a seventh straight losing session ¹². A new $2.4 billion debt financing, led by Blue Owl, for Iren Ltd. to buy Nvidia Blackwell Ultra GPUs for a Canadian data center ¹³ illustrates that global capital is still flowing to AI compute even as China builds its own supply chain.

None of this changes the standing judgment. MIIT's "super-conventional measures" and the NDRC's acceleration plan are the closest things to central signals in this batch, but they lack implementation detail. Watch whether they translate into Big Fund-style money or binding procurement, SMIC's Q3 earnings and capex, CXMT's LPDDR6 ramp beyond Xiaomi, and whether provincial IC plans become enforceable. Until then, China's chip story is design-plus-foundry-plus-equipment-plus-materials, moving at different speeds, all below an untouched ceiling.