How Chinese Models Took Over Open-Source AI | Irene Solaiman (Hugging Face)

Irene Solaiman (Hugging Face) on the data behind a shift in open-source AI: Chinese open-weight models now lead downloads, derivatives, and trends. Solaiman walks through Hugging Face usage data on open-source and open-weight AI. Downloads concentrate in the US, Western Europe, and China, and about a year ago the pattern turned sharply toward Chinese models: roughly 75% of downloads from the top 20 authors now come from models developed in China, and Chinese models have led platform trends since early 2025. Looking past downloads to derivatives, Qwen dominates what developers build on. She reads the shift through the lens of AI sovereignty, naming South Korea as a country to watch, and closes on the deeper stake: whose models spread globally, and whose cultural values travel with them. Chapters 0:00 How do we measure open-source AI usage? 0:24 Downloads by geography: US, Europe, China 0:51 ~75% of top-author downloads now from China 1:21 Chinese models top Hugging Face since 2025 1:56 Derivatives: why Qwen dominates 2:35 The explosion of Chinese open repositories 3:46 Hardware, diffusion, and AI sovereignty 4:39 Why building from scratch is so hard 5:14 Global influence and the values inside models More AI safety research: https://far.ai Alignment Workshop playlist: https://youtube.com/playlist?list=PLBY5kyt_LfFg&si=B9I56daxBmDAeRwQ FAR.AI is a research nonprofit working to ensure the safe development of advanced AI. We host the Alignment Workshop series and publish frontier alignment research.