Key Takeaways
- Chinese-developed models account for 41% of large language model downloads on Hugging Face, compared to 36.5% from the United States.
- Alibaba’s Qwen model family has generated over 113,000 derivative models and captured more than 50% of global open-source model downloads.
- The surge was triggered by DeepSeek’s R1 release in January 2025, which sparked a wave of open-source releases from Chinese firms.
- China now holds 60% of the world’s AI patents by volume, according to World Intellectual Property Organization data.
- A US government commission has warned that China’s open-source dominance poses “the most serious long-term challenge to US AI leadership.”
The numbers arrived quietly but carry enormous weight. Hugging Face’s Spring 2026 report on the global open-source AI ecosystem revealed that cumulative downloads of Chinese-developed models have crossed 10 billion a milestone that Chinese state media was quick to amplify and that analysts say marks the clearest sign yet of a tectonic shift in the geography of artificial intelligence.
Between February 2025 and February 2026, Chinese-developed models accounted for 41 percent of large language model downloads on Hugging Face, the dominant platform for open-source AI distribution. That figure puts China ahead of the United States, which claimed 36.5 percent of downloads over the same period. The inversion, while not entirely surprising given the pace of releases from Chinese labs over the past year, represents the first time China has held the plurality of that market.
At the centre of the surge is Alibaba’s Qwen family of models. The series has spawned more than 113,000 derivative models on Hugging Face more than Google and Meta combined and by March had accumulated nearly 1 billion downloads of its own, capturing over half of global open-source model downloads according to an analysis by Interconnects AI. The Qwen family’s dominance reflects a deliberate strategy by Alibaba to flood the ecosystem with capable, permissively licensed models that developers around the world can fine-tune and deploy.
The inflection point for Chinese open-source AI came in January 2025 with DeepSeek’s release of its R1 reasoning model, which stunned the industry with performance that rivalled closed systems from OpenAI and Google at a fraction of the training cost. The release triggered a cascade of open-source activity from Chinese firms. Baidu went from zero model releases on Hugging Face in 2024 to over 100 in 2025. ByteDance and Tencent each increased their output by as much as nine times in the same period.
The U.S.-China Economic and Security Review Commission took notice. In a March report titled “Two Loops,” the commission warned that China’s open-source approach is creating a “self-reinforcing competitive advantage,” enabling Chinese labs to innovate near the frontier even as American chip export controls limit their access to the most advanced hardware. The more widely Chinese models are adopted, the faster they improve; the faster they improve, the more they are adopted a loop the commission described as the most serious long-term threat to American leadership in AI.
China has also become the world’s largest holder of AI patents, with applications accounting for 60 percent of the global total according to data from the World Intellectual Property Organization. The scale of that intellectual property position suggests the downloads headline is not merely a consumer metric but reflects deep structural investment in AI as a national competitive priority.
The momentum shows no sign of slowing. On April 25, DeepSeek released a preview of its V4 model, which it intends to release as open-source and which it claims performs competitively with leading closed models from Anthropic, OpenAI, and Google. Independent developers not major corporations now account for 39 percent of all downloads on Hugging Face, according to the same report, suggesting that the open-source AI ecosystem is becoming more decentralised and harder for any single actor, American or Chinese, to control.
What is clear is that the competition has moved well beyond who trains the most powerful model. As one analyst platform put it in a widely circulated assessment: “Everyone is shifting the focus of competition from ‘who trains the strongest model’ to ‘who controls deployment and usage.’” On that front, China is no longer catching up in many respects, it has taken the lead.
Your First 10 AI Skills
10 practical AI skills, copy-paste prompts and a 7-day plan to start using AI with confidence.
Download the guide →