1 hour ago
21
The scoreboard in the global AI race just got an update, and it’s not the one Silicon Valley was hoping for. Chinese-developed open-weight AI models now account for 41% of all downloads on Hugging Face, the world’s largest repository for AI models, compared to 36.5% from US-developed models during the period from February 2025 to February 2026.
Hugging Face CEO Clément Delangue has been vocal about the trend, pointing to China’s growing dominance in the open model ecosystem. In a prior measurement period ending August 2025, Chinese models held a narrower edge at 17% versus 15.8% for US models.
The download numbers tell a clear story
Chinese AI labs like Alibaba, with its Qwen series, and Zhipu AI, with its GLM family, have been releasing models that compete with or exceed American counterparts on industry benchmarks. Chinese labs have been attaching permissive licenses to their models, making them easier to deploy commercially without the legal overhead that sometimes accompanies US alternatives.
When US models couldn’t respond, a Chinese one did
Delangue offered a striking anecdote that underscores why this shift matters beyond download charts. In July 2026, Hugging Face itself faced a cyberattack. The company turned to US-developed AI models first to help with the response. They didn’t work well enough, hampered by safety guardrails that limited their utility in the high-pressure, adversarial scenario.
So Hugging Face pivoted to Zhipu AI’s GLM 5.2 model, which handled the task effectively.
What this means for the competitive landscape
Delangue’s argument centers on three advantages of open models: cost, accessibility, and deployment efficiency. Open-weight models let companies run AI locally, customize it for specific use cases, and avoid the per-query pricing of closed API services like those offered by OpenAI or Anthropic.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·