AI & Algorithm Governance

China's AI policy is still built for speed: the promotion track dominates, and governance is being pushed into infrastructure, market-integrity, and judicial channels rather than new content rules. The August 28-31 documents—the 2026 Big Data Expo wrap-up, Guizhou's 'green renewal' compute-center push, MIIT's computing-power node list, and Shanghai's space-computing forum—all confirm that picture. The one new wrinkle is a Bloomberg report ¹ that Beijing has rebuked Anthropic and set terms for a key US-China AI dialogue, adding a bilateral, company-specific edge to China's otherwise conference-driven global AI governance narrative.

The infrastructure mechanism remains the centerpiece. MIIT's August 28 notice formally named 17 provinces as national regional nodes for computing-power interconnection under a '1+M+N' architecture with 'unified identification, unified standards, unified rules' ². Chongqing's inclusion ties into its 15th Five-Year AI plan and the '疆算入渝' data highway. Guizhou's numbers fill in the same story: 176 EFLOPS total, 169.94 in use (96%), 38万 racks, 40万+ servers, 50 key data centers, a 400G backbone reaching 44 cities, and over 90% of users outside the province ^3,4. The Digital Expo's own wrap-up put 2025 data-industry output at 6.78 trillion yuan (+15.7%) and the number of data firms at 48.2万 ⁵.

Governance is not retreating—it is moving to the compute market and the courts. The sharpest signal remains the Guizhou High Court's 16-measure opinion supporting 'Digital Smart Guizhou' (数智贵州) of August 27, which embeds AI model training, algorithm governance, and generative-output ownership questions into adjudication rules and explicitly names false computing-power advertising and resource overselling as abuses to crack down on ⁶. No new central algorithm-registration or large-model content rules appeared in this batch. Instead, the legal sector is shaping de facto precedents that can later be nationalized.

A new twist in this batch is Guizhou's 'green renewal' (绿色焕新) track: under the 2024 State Council equipment-renewal plan (《推动大规模设备更新和消费品以旧换新行动方案》) and the Data Center Green Low-Carbon Development Action Plan (《数据中心绿色低碳发展专项行动计划》), Guizhou is now pairing its compute build-out with hardware recycling, refurbishing, and circular use of retired data-center equipment ⁴. This adds a second de facto filter on top of domestic-content ambitions: state-linked compute may increasingly have to meet green credentials as well as sovereignty credentials.

Provincial 'AI+' promotion continues unabated. Hubei's executive vice governor Zhang Wenbing chaired an August 27 meeting prioritizing embodied intelligence and explicitly calling for support for AI companies to list and for a listing-reserve pipeline ⁷. Jiangsu's '双创计划' review drew 2,000+ global participants, with an explicit AI track ⁸. At the factory level, Xiaomi's Beijing plant now runs ~700 industrial robots at 100% automation, and its humanoid robot '铁大' is 'working' in the plant, with self-tapping nut success reportedly up to 98% ⁹—a concrete benchmark of embodied AI crossing from demo to line. Two cross-border education signals reinforce the promotion track: Hong Kong's HKUST held its first Shanghai million-yuan entrepreneurship competition, launching 'HKUST-InnoBay@Shanghai' and linking embodied-intelligence talent with Shanghai's capital and application scenarios ¹⁰; and the University of Macau opened its 'Aoqin Forum' at the new Hengqin campus with an AI-focused lecture on AGI frontiers ¹¹.

Shanghai's first space-computing industry forum, held in Songjiang, adds a new promotional frontier: it deliberately brought together satellite, semiconductor, cloud-computing, and AI firms, with spaceborne AI inference and even large models on the agenda ¹². The framing—putting a computer in orbit and running inference on satellites—shows the compute-first logic extending beyond data centers. This is promotion, not control; no regulatory framework for spaceborne AI appeared.

Internationally, the pattern has one new element. The Bloomberg report ¹ that China rebuked Anthropic and set terms for a US-China AI dialogue suggests that Beijing is willing to use its market and regulatory weight against individual US frontier labs—a sharper instrument than the WAIC-style 'principles to action' framing ¹³ and the Digital Expo's 21-country data-economy cooperation ⁵. The watch items remain: whether the SPC nationalizes the Guizhou court rules, whether the 81.2% domestic intelligent-computing share hardens into procurement language, whether the node system's 'unified rules' become binding exchange standards, whether green standards become a second filter, and now what operational terms Beijing set for the US-China AI dialogue.