AI & Algorithm Governance

China's AI policy remains promotion-first, but the new documents show that promotion now carries an explicit global economic and diplomatic edge. The 'new three samples' narrative — AI, robots, and innovative drugs — frames computing-hardware trade of 5.13 trillion yuan in H1 2026 (+56.6% y/y) as China's claim to 'global definitional power' (¹), while the foreign ministry's readout of the China–Indonesia strategic dialogue lists AI as a priority for 'new quality productive forces' cooperation (²). The control track is unchanged; the promotion track is converting domestic deployment into export and standards influence.

Domestic promotion continues through hard subnational targets. Chongqing's 15th Five-Year AI plan targets 250 billion yuan in core smart-economy industries, >90% penetration of smart terminals/agents, 50 AI future factories, and a 10,000-card compute cluster with >80% green power (³). Harbin's low-altitude-economy cluster is built around United Aircraft with 30+ suppliers within a 5-km radius (⁴); Hebei's 'one-person company' communities in Xiong'an (36 AI OPCs) and Langfang (2,000 sq.m shared space) spread AI entrepreneurship (⁵, ⁶). Shanghai's dual-10-gigabit network (72,000 5G-A base stations, 174 gigabit communities) and ByteDance's TRAE hackathon (37,000 developers) show infrastructure and skill pipelines advancing in parallel (⁷, ⁸).

The newest infrastructure signal is that compute is being treated as a strategic asset in semi-autonomous arenas too. Hong Kong's innovation chief admits local university/research compute was only 60 PFLOPS in 2023 and is being expanded through the Cyberport AI supercomputing centre (⁹). Macao's 2027 policy consultation explicitly calls for 'promoting the AI industry and building a Macao–Hengqin computing power platform' (¹⁰). Add Chongqing's 10,000-card cluster, and a decentralized, multi-jurisdiction compute build-out emerges — consistent with Beijing's 'six networks' infrastructure plan (⁷).

Governance is static but not neglected. The National Healthcare Security Administration is codifying a staged track for medical AI — from concept through insurance listing to payment (¹¹) — a controlled channel for innovation into the health system. Hong Kong's first 'AI against AI' cybersecurity challenge (290 teams) reflects a managed-security-by-design stance (¹²). No new national algorithm or large-model rules appear in this batch; the filing-and-review regime is unchanged.

The active thread is international positioning. The MFA statement describes AI as a new pillar of China–Indonesia cooperation under the 'new quality productive forces' umbrella, with a new five-year action plan and 'more ambitious bilateral trade targets' to be set (²). The 'new three samples' report argues China is shifting from an 'embedded' player to a 'system builder' in global value chains, citing surgery-robot exports up 3.3x and industrial-robot exports up 18.6% (¹). These are signals that Beijing wants to shape global AI governance and standards, not just domestic deployment.

The limits matter. The Indonesia agreement is a diplomatic framework, not a binding AI pact; the Macao compute platform is a think-tank recommendation awaiting the 2027 policy address; and the 5.13 trillion yuan figure covers hardware, not models or services. The control side — algorithm filing, content safety — remains the unchanged ceiling over all promotion. If large-model incidents mount, the state already has the staged-review machinery (as with medical AI) to tighten selectively.