Chinese open models surpass US downloads on Hugging Face

For several weeks this summer, the AI industry focused on Anthropic’s latest frontier models (the most capable, resource-intensive AI systems) and Washington’s fight over who could access them. But developers quietly moved elsewhere: Chinese open-weight models (AI models whose trained parameters are released publicly for anyone to customize) accounted for 41% of downloads on Hugging Face this spring, surpassing U.S. models. On OpenRouter, a service providing unified access to AI models, the top six most popular models are all open models from Chinese firms including Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai, while Anthropic — developer of frontier models like Claude — trails behind them. Data from Vercel, a cloud platform for frontend developers, shows open-weight models absorbing much of the volume-heavy infrastructure of AI apps while closed models operate as the higher-cost, premium layer: open models handled nearly a third of AI requests on the platform in June.

Hugging Face CEO Clem Delangue predicts, “Maybe in a few years, the frontier models will be for experimenting and [for] 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.” He says the pattern reflects a broader preference for owning AI infrastructure rather than renting it: companies don’t want to outsource core capabilities to a third-party API they can’t control, inspect, or own. That preference shows up in Hugging Face’s own activity — a new repository is created every seven seconds on the platform, which hosts almost three million public models and one million public datasets, according to Delangue. Half of all Fortune 500 firms are using Hugging Face to deploy private and open source models, he says.

The rise of open models coincides with a steady stream of open-weight releases from Chinese AI labs, including Z.ai, a Beijing-based company that released an open-weight model called GLM-5.2. Microsoft CEO Satya Nadella has also warned enterprises against locking into a single model provider, arguing that if learning flows only one way, economic value shifts toward the owners of that infrastructure rather than the creators of the underlying knowledge — which is why he says it’s important to distribute learning infrastructure so each firm can control its own loop.

The open-model surge has intensified debate over how much access to powerful AI should be restricted. Anthropic CEO Dario Amodei has argued that scaling powerful open model weights could become dangerous because once they are released, they become difficult to control; other critics argue open models are easier to access by bad actors who could use them to spread disinformation or enact cyber or biological warfare. Delangue counters that keeping powerful models closed simply concentrates the technology in the hands of a few companies and reduces transparency into how they work. “The biggest risk in AI is concentration of power. The way you make the world safer, in my opinion, is by leveling up the playing fields and creating transparency on these models,” he said. “You don’t really make it safe by keeping it behind closed doors for just a few players. You make it more dangerous because you create asymmetry of power and asymmetry of capabilities.”

Why it matters

The growing popularity of open-weight models, particularly from Chinese firms, is challenging the dominance of expensive, closed frontier models from U.S. providers as enterprises seek cheaper, customizable alternatives they can fully control.