AI & Algorithm Governance

China's AI policy remains promotion-first, and the new documents push the promotion story deeper into the real economy and the labor market. The strongest new signals are entirely on the demand side: Chongqing's automotive leaders are redefining cars as 'iterable electronic products' with intelligent driving as the core value (L2 penetration 70.5%), AI bootcamps are placing graduates who were locked out of the AI boom into entry-level AI-training jobs, and Zhongji Innolight's record optical-module earnings confirm that compute-hardware demand is still soaring. The control track and the global-standards push are unchanged; what is new is that the domestic deployment story now has visible social consequences and a market-side caution flag.

Industrial promotion continues to be concrete and subnational. At Chongqing's 'automotive new quality productive forces' forum, Chang'an and Seres presented intelligent driving, smart cabins and software as the new axis of competition, with L2-level assisted-driving penetration across domestic models at 70.5% ¹. This shows AI moving from dedicated AI zones into the flagship manufacturing sector, consistent with Harbin's drone cluster around United Aircraft, where more than 30 suppliers sit within a 5-km radius ². The policy machine is not just building AI islands; it is wiring AI into export-oriented industries.

The labor market is adapting in ways that cut both ways. Sixth Tone reports that AI training bootcamps are absorbing recent graduates, with one young woman learning Python in a three-month program to become an 'AI Junior Trainer' — a growing entry-level role ³. This is the human-capital counterpart to ByteDance's TRAE hackathon (37,000 developers) and the Xiong'an/Langfang one-person-company communities ⁴, ⁵, ⁶. But it also signals a two-tier market: the AI boom is creating many low-level support jobs while the high-skill engineering roles remain hard to enter, which could feed social friction if the promotion track stumbles.

Compute remains the strategic bottleneck, and it is being addressed across jurisdictions. Hong Kong's innovation secretary acknowledged that local university and research compute was only 60 PFLOPS in 2023 and is now being expanded through the Cyberport AI supercomputing centre ⁷; Macao's 2027 policy consultation already includes a Macao–Hengqin computing power platform ⁸; Shanghai has deployed 72,000 5G-A base stations and 174 10-gigabit communities under the national 'six networks' plan ⁹. This decentralized build-out positions semi-autonomous SARs as nodes in a national compute grid rather than passive observers.

The export narrative is real but now has a market-side counter-signal. The 'new three samples' story — AI, robots, innovative drugs — frames H1 2026 computing-hardware trade of 5.13 trillion yuan (+56.6%) and surgery-robot exports up 3.3x as China's move from 'embedded' to 'system builder' ¹⁰. Yet Zhongji Innolight, a bellwether optical-module maker, delivered H1 revenue of 41.78 billion yuan (+182%) and net profit of 13.65 billion (+242%), then saw its stock fall 7.72% the day after a 164-institution conference call, with main funds net-selling 39.3 billion yuan, because investors questioned cash flow and customer concentration ¹¹. The global AI hardware boom is generating real revenue, but markets are beginning to price in concentration and sustainability risks — a caution for the 'global definitional power' argument.

Governance remains static. There are still no new national algorithm or large-model rules in this batch; the National Healthcare Security Administration's staged pathway for medical AI — from concept to insurance listing to payment ¹² — remains the clearest example of how the state channels innovation through managed review. Diplomatically, China-Indonesia cooperation on AI under 'new quality productive forces' is a framework, not a binding pact ¹³. The overall balance is unchanged: promotion is the engine, control is the ceiling, and international positioning is the amplifier.

For outside readers, the practical implication is that China's AI economy is broadening — into autos, labor markets, and export supply chains — while the governance architecture stays flat. The new counter-signal from Zhongji Innolight and the bootcamp labor story are refinements, not reversals; the standing assessment of promotion-first holds. If large-model incidents or export volatility mount, the state has both the staged-review machinery and the policy levers to tighten selectively.