Artificial Intelligence in China

Artificial intelligence in China refers to the development, deployment and governance of AI across the world's second-largest economy, an ecosystem shaped by three forces: a state industrial policy dating to 2017, a generation of efficient model developers that emerged partly in response to United States export controls, and the largest domestic user base for generative AI anywhere. Chinese open-weight models such as DeepSeek and Alibaba's Qwen now account for roughly 30% of global AI model downloads.^[1]

Policy and Background

China's national AI strategy dates to the New Generation Artificial Intelligence Development Plan of 2017, which set the goal of becoming a global AI leader by 2030.^[2] In August 2025 the State Council issued the AI Plus initiative, an adoption-focused programme intended to embed artificial intelligence across the economy, industry and society, with targets including a 70% penetration rate for intelligent terminals and AI agents in key sectors by 2027 and a ten-year horizon ending in an AI-integrated economy by 2035.^[2][3]

Companies and Models

The release of DeepSeek's V3 and R1 models in late 2024 and early 2025 marked China's arrival at the frontier. V3 was trained on 2,048 Nvidia H800 chips at a reported compute cost of about US$5.6 million, and R1 matched the leading American reasoning model of the day on mathematics and coding benchmarks at a fraction of the cost. The V4 family, released in April 2026, ran on Huawei, Cambricon and Hygon chips at launch, while the follow-up V4.1-Flash, released in September 2026, activates only about 3% of its parameters per word and sharply cut its memory requirements.^[4]

ByteDance's Doubao assistant led the domestic app market with 382 million monthly active users in June 2026, ahead of Qwen's 167 million and DeepSeek's 129 million, while AI-native apps overall reached roughly 499 million monthly users.^[5] By December 2025, 602 million Chinese people were using generative AI tools, an increase of 141.7% from a year earlier.^[6]

Compute and Export Controls

United States export controls have shaped the sector since 2022, when advanced Nvidia chips such as the A100 and H100 were restricted, followed by tightened rules in 2023 and license requirements for the H20 in April 2025. Nvidia's share of China's AI chip market fell from about 95% to effectively zero under the policy, and the company recorded a US$4.5 billion write-down on unsellable inventory.^[4] Policy has oscillated: in December 2025 Washington cleared exports of the more powerful H200 chip subject to revenue-sharing terms, though deliveries had not materialised by mid-2026.^[7]

Chinese firms adapted by optimising around scarcity — DeepSeek engineers rewrote routines at the PTX level to squeeze more from restricted hardware — and by adopting domestic silicon.^[8] Huawei's Ascend line is the anchor: roughly 750,000 of its Ascend 950PR chips are planned for 2026, and nine domestic AI processors were certified for government procurement. Analysts describe the result as induced innovation, where compute scarcity pushed Chinese labs toward efficiency gains that now influence global model design.^[4][7]

>See Also

• DeepSeek
• Qwen
• ByteDance AI
• AI Chip Export Controls
• Moonshot AI