Is the closed-source AI era ending? Chinese models now dominate 41% of Hugging Face downloads
chinese open-weight models captured 41% of hugging face downloads this spring, surpassing us competitors. the shift signals enterprises are increasingly rejecting expensive closed apis in favor of cheaper, customizable alternatives for production workloads.
chinese open-weight models secured 41% of hugging face downloads this spring, pushing us models out of the top spots on platforms like openrouter, where deepseek and tencent currently rank higher than anthropic's claude. vercel reported that open models handled nearly a third of requests in june, highlighting a shift toward infrastructure-heavy, low-cost inference. hugging face ceo clem delangue noted that companies increasingly prefer owning their models over renting closed apis to avoid vendor lock-in, with half of the fortune 500 now deploying private or open models. this dynamic suggests that while frontier models remain critical for specialized tasks, the bulk of enterprise ai volume may migrate to customizable, open-weight architectures.