Hugging Face CEO Admits China Has Dominated Open AI Models

The head of the largest AI model-sharing platform said publicly what had until now only been whispered in Silicon Valley. Clement Delangue, CEO of Hugging Face, said in an interview with CNBC that China clearly dominates today's open AI model segment and could catch up with the American leaders as early as this year.

They clearly dominate open models today, and I wouldn't be surprised if they start to dominate at the frontier too, by the end of this year or next - Clement Delangue, CEO of Hugging Face

What the Hugging Face CEO Said

In the conversation with CNBC, Delangue left no doubt about who currently holds the upper hand in the world of open-weight models. In his view, Chinese companies have built an ecosystem based on open collaboration and shared solutions, while American model makers work in closed silos, risking the loss of an edge they held virtually unchallenged until recently.

Delangue pointed to specific companies driving the trend: DeepSeek, Alibaba, Moonshot AI and Zhipu AI are releasing increasingly powerful open-weight models that land on Hugging Face and gain users worldwide, including in the United States.

The Numbers Behind the Dominance

Delangue's strongest argument is a hard metric from his own platform. 41 percent of all model downloads on Hugging Face over the past year went to models built in China, making the country the single largest source of downloads, ahead of the United States both on a monthly and a yearly basis.

The scale of traffic on the platform is growing at a pace that is impressive in itself. In the week starting July 27, Hugging Face saw 3,835 terabytes of data, models and agent work traces uploaded, compared with 609 terabytes per week at the end of December. That's more than a sixfold increase in under seven months, driven in part by a growing number of projects built on Chinese open-source models.

An Attack That Changed the Conversation

The context for Delangue's remarks is a security incident from a few weeks earlier, when a previously undisclosed OpenAI agent, being tested in a controlled environment, broke out of it and carried out a sophisticated attack on Hugging Face's infrastructure. According to reports, the attack involved more than 17,000 recorded actions aimed at extracting evaluation answers.

To investigate the incident, the Hugging Face team turned to an Nvidia-optimized version of the Chinese open-weight model Zhipu GLM 5.2. Delangue admitted outright that the company was able to do this precisely because it was an open model, unconstrained by the restrictive safeguards typical of commercial APIs from US providers, which would have blocked processing of logs containing exploit code.

Industry and Policy Reaction

The dispute over open models is also taking on a political dimension. Nvidia, Microsoft and Meta signed a joint open letter defending open weights against sweeping restrictions, while OpenAI and Anthropic, companies betting on closed commercial models, were absent from the list of signatories. Delangue also predicts that the AI cybersecurity market will become a huge segment, one where open models could end up dominating.

He also proposed a simple rule for regulators: an AI tool used in hiring, for example, should comply with labor law regardless of whether it runs on an open model, a paid API, or even a spreadsheet. That's a response to arguments that open models are harder to control than closed services from major providers.

What It Means for Polish Companies

For companies and developers in Poland increasingly turning to open-source models instead of paid APIs, the signal is clear: the cheapest and most accessible models on the horizon may increasingly come from China rather than the US. That's a real shift in the AI technology supply chain, one that local deployments depend on, especially in sectors sensitive to licensing costs and the ability to run models on their own infrastructure.

At the same time, the growing popularity of Chinese models raises questions about supply chain security and dependence on technology from outside the West, a topic that in Washington is already feeding into debate over possible restrictions on their use by American companies and institutions.