Chinese AI Labs Lead Global Open-Source Model Adoption, Challenging US Dominance
Chinese AI laboratories are increasingly embracing an open-source approach, providing models as downloadable "open-weight" packages for developers, eliminating the need for commercial agreements with American companies. This movement gained momentum after DeepSeek released its R1 model in January 2025, which demonstrated performance on par with US systems but at reduced costs. Major Chinese players such as Z.ai, Moonshot, Alibaba's Qwen, and MiniMax are also joining this trend, narrowing the competitive gap with their US counterparts. Research indicates that by August 2025, Chinese open-weight models represented 17.1% of global AI downloads, outpacing the US's 15.86%. However, these models also reflect China's content moderation policies. In February, Anthropic accused Chinese labs of unlawfully extracting features from Claude. The Global South is increasingly turning to Chinese models to assert AI sovereignty.
Key facts
- Chinese AI labs release models as downloadable open-weight packages
- DeepSeek open-sourced its R1 reasoning model in January 2025
- Chinese models matched US system performance at lower costs
- Chinese open-weight models accounted for 17.1% of global downloads August 2024-2025
- Alibaba's models have more user-generated variants than Google and Meta combined
- Anthropic accused Chinese labs of illicit capability extraction in February 2026
- Singapore chose Alibaba's Qwen over Meta's Llama for regional model
- Malaysia announced sovereign AI ecosystem would run on DeepSeek
Entities
Institutions
- DeepSeek
- Z.ai
- Zhipu
- Moonshot
- Alibaba
- MiniMax
- MIT
- Hugging Face
- Meta
- Anthropic
- OpenAI
- AI Singapore
- Linux
- Android
Locations
- China
- United States
- Singapore
- Malaysia
- Nairobi
- São Paulo
- San Francisco
- Silicon Valley
- Global South