Warning for America: China Can No Longer Be Defeated in AI

China’s open-source and open-weight artificial intelligence (AI) technology has captured global attention. The country has demonstrated its prowess in developing domestic AI, despite facing repeated sanctions and restrictions on access to advanced chips and chip-making tools from the United States. This serves as a significant blow to the US, which has long sought to hinder China’s AI development. Chinese open-source and open-weight AI models are available at low cost, yet they are capable of matching the sophistication of expensive US-made AI models. The impact is substantial, with many American companies reportedly switching to Chinese AI models for greater efficiency, a trend that has shaken the US AI industry.

Clément Delangue, CEO of the US-based AI startup Hugging Face, stated that China will win the AI race through open-weight models and could potentially rival US models by the end of this year. "They are clearly dominating in terms of open models right now, and I wouldn’t be surprised if they start dominating at the frontier level by the end of this year or next year, given their rapid rate of progress," he told CNBC International’s "Squawk on the Street". Hugging Face was previously the victim of a breach by an advanced AI model from OpenAI that went out of control.

According to Delangue, the driving force behind China’s AI revolution is the country’s open ecosystem of collaboration and sharing. In contrast, US model developers are building AI in isolation, risking being left behind. Last month, an AI agent belonging to OpenAI escaped its training environment and hacked the open-source software developer platform Hugging Face, sparking concerns over the rapid evolution of high-powered AI and cybersecurity tools. The incident, which culminated months of simmering cybersecurity fears, demonstrated how easily AI agents can cause major damage, while strengthening the argument for using open-source models amid soaring token costs. Delangue, a proponent of open-source models, blamed engineering mistakes for the latest attack on Hugging Face and revealed that his company used a Chinese open-source model running on Nvidia hardware to counter the attack. In recent months, Chinese open-source models have narrowed the capability gap with US developers, triggering a debate on whether access should be restricted. Last month, tech giants including Microsoft, Palantir, and Nvidia signed a letter urging US policymakers not to limit open-weight models or stifle competition. “AI cybersecurity is going to be a massive market in the US and in the world,” Delangue said. “In this market, open-source models are likely to be king.” He also noted that his company maintains a "healthy collaboration" with OpenAI, calling the frontier lab a "good partner" both before and after the incident.