Chinese AI Policy

Chinese AI policy is the national approach of the People's Republic of China to governing and promoting artificial intelligence. It couples two goals that the US market-led approach and the EU's rights-based EU AI Act (Regulation 2024/1689) treat separately: becoming the world leader in AI, an explicit state objective since the 2017 New Generation AI Development Plan, and maintaining party-state control over information and social stability. Policy is therefore simultaneously promotional (industrial subsidies, state investment, talent programs, compute buildout) and restrictive (registration, content rules, algorithm filing).

Promotional pillar: industrial policy

China treats AI as strategic national infrastructure. The 2017 New Generation Artificial Intelligence Development Plan set the target of global AI leadership by 2030, and successive five-year plans, local-government guidance funds, and state-owned-enterprise procurement have channelled large-scale capital into models, chips, and data centers. Usage has scaled with the buildout: China's National Data Administration reported daily national token consumption of 140 trillion in March 2026, up from 100 trillion in December 2025 and 100 billion in early 2024 (Source: scmp.com).

Under US Export Controls (AI), this pillar increasingly emphasizes compute self-sufficiency — domestic accelerators (Huawei — Ascend AI Accelerators, SMIC fabrication) intended to substitute for restricted Nvidia hardware. Capital remains a constraint on the private side of the buildout: the Wall Street Journal reported on July 21, 2026 that Chinese AI companies were rushing to raise funds, with researchers and executives at those companies saying shortages of investment and chips are holding them back against leading American rivals (Source: wsj.com); the same day, Ant Group's overseas arm raised $1.2 billion to accelerate its global expansion (Source: theinformation.com). A further strategic choice has been Chinese labs' embrace of open-weight model release (DeepSeek, Alibaba's Qwen, and others), which spreads Chinese models globally, builds ecosystems, and partially routes around the compute gap by maximizing the value extracted from each trained model.

Top-leadership signaling has tracked the technology's rising priority. On July 13, 2026, China's Ministry of Foreign Affairs said President Xi Jinping would make his first appearance at the World AI Conference, delivering the keynote at the opening of the three-day Shanghai summit starting Friday, July 17, 2026 — a signal, per Bloomberg, of the technology's growing importance to Beijing as rivalry with the U.S. intensifies (Source: bloomberg.com). Xi delivered the keynote on July 17, 2026 — the first by a Chinese head of state at the conference — and the Ministry of Foreign Affairs' official English text is tracked as Joining Hands to Build a Just and Equitable System For Global AI Governance (Xi Jinping, WAIC keynote, July 2026). It organizes China's position into four principles: encouraging "open source, openness, collaboration and sharing" as AI moves "from the digital world into the physical world"; ensuring AI is "always under human control" through laws, technological monitoring, and early-warning systems while jointly opposing "overstretching the national security concept in the field of AI and placing one country's security over that of others"; protecting civilizational diversity and shaping "the values of AI with humanity's common values"; and practicing "true multilateralism" recognizing "the important role of the United Nations" toward "a consensus-based global governance framework." It confirmed that the World AI Cooperation Organization had "come into being in Shanghai" under an agreement signed by 29 countries, and announced three five-year commitments: 5,000 AI training and seminar opportunities for developing countries, international AI application cooperation centers with ASEAN, the League of Arab States, the African Union, CELAC, the SCO, and BRICS, and access for 30 countries to the AI-powered meteorological warning system MAZU. Contemporaneous coverage read the address as also criticizing U.S. dominance in remarks courting developing countries (Source: wsj.com; theinformation.com; english.news.cn; reuters.com; english.scio.gov.cn). Stratechery's Ben Thompson read the speech as preceding Alibaba's July 19 return to open-weight release at the flagship tier with the Qwen3.8-Max preview (Who's Afraid of Chinese Models? (Ben Thompson, Stratechery, July 2026)). The openness message coexists with a reported restrictive track: Reuters reported on July 7, 2026 that Beijing is considering curbing overseas access to China's top AI models (Source: reuters.com). The deliberations advanced by July 21, 2026, when China's consultations with leading domestic AI companies on tighter export controls — potentially barring foreign nationals from downloading open weights of frontier Chinese models — became public (Source: ft.com). The Ministry of Commerce consulted Alibaba, ByteDance, Huawei, Qualcomm, and TSMC on controls covering AI model weights and overseas fabrication of Chinese-designed chips (Source: techieray.substack.com). The same day, US Treasury Secretary Scott Bessent said the US supports open-source AI but "we have the ability to sanction" overseas models that steal from US companies via distillation (Source: cnbc.com); see Distillation.

Control pillar: the regulatory regime

China was an early and aggressive AI regulator, but with a control-oriented rather than rights-oriented logic. The administering body is the Cyberspace Administration of China (CAC), working with MIIT and other ministries. Its key instruments are:

• Generative-AI registration (da moxing bei'an / 大模型备案) — under the 2023 Interim Measures and the revised Cybersecurity Law (effective Jan 1, 2026), any firm offering generative-AI services to the Chinese public must complete formal registration with the CAC. The registration regime, detailed at China Generative AI Registration (da moxing bei'an / 大模型备案), functions as a market-access survival condition, not a privilege tier.
• Algorithm filing and recommendation rules — earlier regulations (2021–2022) require registration of recommendation algorithms and deep-synthesis (deepfake) services.
• Content and "core socialist values" requirements — generative outputs must align with state content rules, which shapes training-data curation and model behavior in ways with no direct Western analogue.

Comparative positioning

Chinese AI governance is commonly contrasted with the US and EU models along several dimensions:

This positioning underlies the recurring US–China AI competition debate (AI and National Security; House Select Committee on the Strategic Competition between the United States and the Chinese Communist Party; SCALE Act — Rep. John Moolenaar (April 2026)). The export-control question — whether US restrictions preserve a US lead or merely accelerate Chinese self-sufficiency — turns directly on how effective China's promotional pillar proves to be.

Debates and positions

Analysts disagree on China's trajectory. One view holds that state direction plus scale lets China close the frontier gap quickly, especially via open weights and compute self-sufficiency. Another holds that control-pillar constraints (content rules, data restrictions, capital misallocation through subsidy-chasing) impose a structural drag, and that export controls bite harder than Beijing admits. Current expert testimony to US bodies frames it as a genuinely open race (Source: brookings.edu).

Relationships

• depends-on: China Generative AI Registration (da moxing bei'an / 大模型备案)
• related: Export Controls (AI)
• related: AI and National Security
• related: House Select Committee on the Strategic Competition between the United States and the Chinese Communist Party
• related: SCALE Act — Rep. John Moolenaar (April 2026)
• related: Huawei — Ascend AI Accelerators