Open models now handle 41% of AI downloads, edging out U.S. giants

While the industry obsessed over Anthropic’s frontier models and Washington’s export controls this summer, the real action was elsewhere. Chinese open-weight models quietly captured 41% of downloads on Hugging Face this spring, surpassing their U.S. counterparts. On OpenRouter, the top six most popular models all come from Chinese firms—Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai—leaving Anthropic’s Claude Opus 4.7 trailing in seventh place.

That’s not a fluke. Data from Vercel shows open models absorbing nearly a third of AI requests on the platform in June, doing the unglamorous, high-volume grunt work while closed models operate as a premium layer. Hugging Face CEO Clem Delangue put it bluntly on a recent Equity podcast: “Maybe in a few years, the frontier models will be for experimenting and some really high-value tasks, and most of the production workloads will actually be powered either by private models within companies or by open source models.”

The calculus is shifting from raw intelligence to ownership and economics. Delangue argues companies are waking up to the bill that comes with scaling closed frontier models and recoiling at the idea of outsourcing core capabilities to a black-box API they don’t control. The numbers back that up. A new repository is created on Hugging Face every seven seconds, and half of all Fortune 500 firms now use the platform to deploy their own private or open source models. That’s a far cry from the “one model to rule them all” narrative.

Microsoft CEO Satya Nadella added fuel to this argument recently, warning against single-provider lock-in and calling out the irony of model providers restricting distillation rights while reserving the right to learn from customer data. Meanwhile, Beijing-based Z.ai just dropped GLM-5.2, an open-weight model that excels at agentic coding and competes with Anthropic’s latest on security vulnerability detection. Every few months, the economics of proprietary AI take another hit from a cheaper, more customizable Chinese release. The debate over whether that’s dangerous—Anthropic’s Dario Amodei has warned open model weights become hard to control once released—is getting harder to ignore, but the market is voting with its downloads.