AI & Algorithm Governance

China's AI policy is still promotion-first, and the newest documents show that tilt continuing across compute infrastructure, education and diplomacy, while the hard governance track remains flat. The anchor is the national compute build-out: Guizhou's installed capacity reached 176 EFLOPS, up more than 200-fold from the 0.84 EFLOPS recorded at the start of the 14th Five-Year Plan, with smart computing making up over 90% of the total ¹. Shanghai is now styling itself an "innovation hub for AI development and governance" ², and Hong Kong is being positioned as a bridge for intelligent computing ³—yet there are still no new national algorithm or large-model rules. For outside readers, the important point is that China is normalizing AI as infrastructure, curriculum and export diplomacy while keeping control mechanisms at the pilot stage.

Guizhou's numbers are the most concrete public baseline for the promotion machine. The province has 50 key data centers in operation or under construction, up from seven a decade ago, and it plans to push capacity to 190 EFLOPS in 2026 while targeting roughly 300 billion yuan in digital-intelligence industry output and a computing industry that could top 100 billion yuan ^1,4. Provincial officials are selling the province on cost: Huawei Cloud estimates deployment in Guiyang is about 30% cheaper than coastal cities, latency to Chengdu-Chongqing is 6 ms and to the Greater Bay Area 10 ms, and over 90% of customers come from outside Guizhou ⁴. This is no longer a pilot; it is a marketed national facility.

The new political signal comes from Shanghai. At a municipal Party committee study session, Party Secretary Chen Jining, after hearing from Shanghai AI Laboratory director Zhou Bowen, called for seizing the "new foundation, new paradigm, new industry, new governance" window and building Shanghai into an innovation hub for AI development and governance, citing Xi Jinping's keynote at the 2026 World AI Conference ². Nothing in that session amounts to a new rule, but it folds "governance" into the promotion narrative instead of treating it as a separate constraint. Hong Kong is being attached to the same infrastructure story: the innovation and technology bureau's permanent secretary told an industry forum that local university and research computing power was only 60 PFLOPS in 2023 and is expanding through the Cyberport AI supercomputing center, casting Hong Kong as the connector between national AI development and global technology cooperation ³.

The human-capital layer remains visible. Hong Kong's Education Bureau ran "AI Leadership" training for about 1,400 principals and middle managers, framed by the recently published "Digital Education Blueprint" and an "AI Literacy Learning Framework" that teaches students to learn, use, and responsibly use AI ⁵. This is the state deliberately building the social and educational base for AI adoption, not just an industrial policy add-on.

Diplomatically, Beijing is leaning on robotics and trade fairs. At the World Humanoid Robot Games, the Tiangong Ultra humanoid robot ran 100 metres in 8.86 seconds, under Usain Bolt's 9.58 seconds ⁶. The foreign ministry immediately converted the feat into an open-cooperation pitch—"technology interconnection, standards co-construction, market sharing"—and invited global cooperation on AI and embodied intelligence ⁷. Guangdong used the Guangdong-Singapore Cooperation Council to announce AI projects with Singapore ⁸, and the Xiamen investment fair has registered delegations from 123 countries and regions, with AI and green-tech firms showcased and multinationals now prioritizing compute power as a new production factor ⁹. This is the "new three samples" export narrative continuing.

The control track still has not moved. No new national algorithm or large-model rule appears in this batch; the tightest existing mechanism remains the National Healthcare Security Administration's staged pathway for medical AI, from concept to insurance listing ¹⁰. Shanghai's phrase "safe and controllable" and its ambition to be a "development and governance innovation hub" could indicate that Beijing will package tighter oversight as innovation in governance, but on the evidence so far the balance is unchanged.

For outside readers, the metric to watch is not rulemaking but compute milestones. Guizhou's 190 EFLOPS and 300-billion-yuan targets by end-2026 are the clearest test of whether the promotion machine can deliver. The larger risk is not a sudden policy pivot but a "governance innovation" squeeze as deployment scales—watch for the staged-review model to be extended from medical AI to autonomous driving or embodied intelligence.