China

• DeepSeek released V4.1-Flash (10 September), a slimmed-down, very low-cost model priced at a fraction of a cent per million tokens, using a new 'causal encoder-decoder' design with lower memory use. Bloomberg framed it as intensifying price pressure on rivals from OpenAI to Anthropic and domestic challenger Z.ai — reinforcing China's cost-led, open-weight competitive strategy.

  2026-09-10

  SourceBloomberg

• China's companion-AI / emotional-support-AI rules — the Interim Measures for AI Anthropomorphic (human-like) Interaction Services, effective 15 July 2026 — move into an early enforcement phase. In the first roughly three weeks, the CAC is reported to have issued around 12 fines totalling some 4.2M RMB against non-compliant companion-app and emotional-chatbot providers (for gaps in filing, security assessment, minor-protection and anti-addiction safeguards); the exact figures are thinly sourced and not yet officially confirmed. This complements the existing generative-AI content-labelling regime and the July Intelligent Agents framework.

  2026-08

  SourceRimonCaixin Global

• DeepSeek releases DeepSeek-V4-Flash-0731 (31 July), the official public build of its V4-Flash line — described as its April V4-Flash preview retrained on a substantially improved post-training pipeline for coding, agents and tool use. DeepSeek reports it beats its own larger V4-Pro preview across nine agentic and coding benchmarks despite activating only ~13B of 284B parameters (MoE, 1M-token context), at roughly $0.14/$0.28 per million input/output tokens; the upgrade is a silent same-endpoint swap for existing API users. CAISI had earlier assessed DeepSeek's V4 Pro as roughly eight months behind top US frontier models.

  2026-07-31

  SourceSimon WillisonMarkTechPost

• China's Global AI Governance Initiative reached a concrete institutional milestone: the launch of the World AI Cooperation Organization (WAICO), a new Shanghai-headquartered intergovernmental body with 29 founding member states, announced by Xi Jinping at WAIC with UN Secretary-General Guterres in attendance.

  Major

  2026-07-17

  SourceMinistry of Foreign AffairsCaixin Global

• Moonshot AI releases Kimi K3, a 2.8-trillion-parameter open-weight model that Artificial Analysis and other benchmarks placed just behind Anthropic and OpenAI's frontier systems, at roughly $15 per million output tokens; media widely described the reaction as a 'second DeepSeek shock,' with same-day selloffs in TSMC (-7%), SoftBank (-9%) and rival Chinese lab Z.ai (-30% in Hong Kong trading), plus a brief dip for Nvidia. Full model weights were scheduled for release on July 27, 2026. Reports differ on the exact release date (July 16 vs. 17, likely a timezone artifact of US vs. Chinese coverage).

  Major

  2026-07-16

  SourceFortuneThe Neuron

• Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents take effect — China's first dedicated regulatory framework for AI agents.

  Major

  2026-07-15

  SourceMRKT3.0

• The Cyberspace Administration of China reported first-phase results of its 'Qinglang' campaign to rectify improper AI applications, launched in April 2026: more than 14,000 non-compliant AI products (websites, apps and agents) removed, over 6 million pieces of illegal content cleared, some 26,000 accounts suspended, 1,300+ product listings taken down and nine illegal open-source datasets removed — targeting unregistered models, weak safety review and content filtering, data poisoning and inadequate AI-content labelling. A second phase, running through the second half of 2026, focuses on AI-generated disinformation, impersonation, harms to minors and online astroturfing, with heavier penalties signalled.

  Major

  2026-07-06

  SourceGlobal Times

• TC260's Ethics-Safety Guidelines take effect (1 July), just two weeks before the Intelligent Agents framework — ByteDance and Alibaba have already begun restricting AI-companion features ahead of the deadlines.

  2026-07-01

  SourceBird & Bird

• Interim Measures for the Administration of AI Anthropomorphic Interaction Services take effect, jointly issued by CAC, NDRC, MIIT, the Ministry of Public Security and SAMR, covering AI companion apps and emotionally-responsive chatbots.

  2026-04-10

  SourceRimon

• DeepSeek's continued open-weight model releases remain a central reference point in international debate over China's frontier AI capability.

  2026

  SourceChinaTalk

• Implementation of the AI+ Initiative — a State Council opinion originally issued in August 2025 — continues to expand state-directed AI integration across manufacturing and public services through 2026.

  2026

  SourceState Council

China's AI research runs through a mix of state academies (Chinese Academy of Sciences), elite universities (Tsinghua, which incubated Zhipu AI), and standards bodies (TC260) that feed directly into binding regulation — a relationship some observers describe as an unusually tight loop between research output and regulatory text compared to many other jurisdictions.

• National AI safety institute

  China has no government-run AI safety institute comparable to the UK AISI or US CAISI. The closest analogue, the China AI Safety and Development Association (CnAISDA, hub at Tsinghua University), is a coalition-style body; as of mid-2025 reporting it lacked dedicated staff, its own budget, or a Chinese-language website, though some 2026 coverage describes it with named leadership (Xue Lan, Yi Zeng, Andrew Yao) and roughly a dozen member organizations, suggesting it may be more developed than earlier reporting implied. Sources also disagree on its founding date (some cite February 2025, others June 2025), and its current operational status could not be freshly confirmed.

• Cloud market share

  Cloud provider market-share figures vary meaningfully by source, quarter, and whether 'overall cloud' or 'AI-cloud-specific' revenue is measured. The dominance figures shown here reflect one dataset — Omdia's H1 2025 estimates for China's AI-cloud market segment specifically, as reported by SCMP — and should be read as indicative of relative standing at that point rather than a precise or durable breakdown; broader estimates of the overall, non-AI-specific cloud market more often place Huawei Cloud ahead of Volcano Engine, unlike the AI-cloud-specific figures shown.

• Foreign hyperscaler presence

  AWS, Microsoft Azure and Google Cloud do not operate normal sovereign cloud regions in mainland China. AWS and Azure are only available there through legally separate, locally-operated entities (Sinnet/NWCD for AWS, 21Vianet for Azure), and Google Cloud has no mainland China region at all — a structural feature of China's data-sovereignty and foreign-ownership restrictions on critical cloud infrastructure.

Deliberately sourced beyond government publications. The bars below show each category’s actual share of this page’s citations — an internal editorial guideline rather than a rigorously derived target, so treat it as a rough steer, not a precise benchmark.

1. Implementation Opinions on Intelligent AgentsResearchMRKT3.02026-07
2. Interim Measures for AI Anthropomorphic Interaction ServicesThink tanksRimon2026-04
3. Opinion on Deepening the Implementation of the Artificial Intelligence+ Initiative (State Council Document [2025] No. 11)GovernmentState Council2025-08
4. Global AI Governance InitiativeGovernmentMinistry of Foreign Affairs2023-10
5. DeepSeek model release coverageExpert commentaryChinaTalk2026
6. Interim Measures for Generative AI ServicesGovernmentCyberspace Administration of China2023-07
7. Algorithmic Recommendation Management ProvisionsGovernmentCyberspace Administration of China2021-12
8. BAAI research publicationsResearchBeijing Academy of Artificial Intelligence2026
9. Chip export control and self-sufficiency analysisThink tanksCAICT2026
10. South China Morning Post AI industry coverageNewsSouth China Morning Post2026
11. Tsinghua University AI governance researchResearchTsinghua University2026
12. Alibaba Cloud Qwen model researchIndustryAlibaba2026
13. China's new regulations on AI anthropomorphic interactive servicesThink tanksBird & Bird2026-07
14. China launches World AI Cooperation Organization (WAICO) at WAIC 2026NewsCaixin Global2026-07-17
15. Official coverage of the Intelligent Agents implementation guidelinesGovernmentgov.cn (English)2026-05
16. TC260's renaming from National Information Security Standardization Technical Committee to National Cybersecurity Standardization Technical CommitteeGovernmentTC2602026
17. China AI-cloud market share data (Omdia, H1 2025)NewsSouth China Morning Post2025
18. DeepSeek's 2026 funding round and reporting on Liang Wenfeng's public absence since February 2025NewsSouth China Morning Post2026-07
19. NDRC leadership (Zheng Shanjie, Chairman)GovernmentNational Development and Reform Commission2026
20. MIIT leadership (Li Lecheng, Minister)GovernmentMinistry of Industry and Information Technology2026
21. Markets experience new DeepSeek shock after MoonShot AI releases Kimi K3NewsFortune2026-07-17
22. Kimi K3 goes openNewsThe Neuron2026-07-17
23. Moonshot nears $30 billion valuation after Kimi K3 releaseNewsYahoo Finance2026-07
24. deepseek-ai/DeepSeek-V4-Flash-0731Expert commentarySimon Willison2026-07-31
25. DeepSeek Upgrades DeepSeek-V4-Flash-0731 with Major Agentic and Coding GainsNewsMarkTechPost2026-07-31
26. China's First AI Companion Rules to Curb Addiction, Protect MinorsNewsCaixin Global2026-07-16
27. Chinese internet regulator announces progress in AI rectification campaign, with over 14,000 irregular AI products removed from cyberspaceNewsGlobal Times2026-07-06
28. DeepSeek's New Low-Cost Model Deals a Fresh Blow to OpenAI, Z.aiNewsBloomberg2026-09-10