The U.S. accuses China's Moonshot AI of using banned Nvidia chips to train its Kimi K3 model

A White House official has accused the Chinese artificial intelligence company Moonshot of having accessed advanced Nvidia chips despite the export restrictions that ban their sale to China. The accusation comes just days after Moonshot unveiled Kimi K3, a model that according to the article closes the performance gap with frontier models such as Anthropic's Fable 5 and OpenAI's GPT 5.6 Sol, and even surpasses them on some benchmarks. It is also the largest open-source model to date, with 2.8 trillion parameters.

Michael Kratsios, director of the White House Office of Science and Technology Policy, wrote on X on Wednesday that Moonshot "acquired servers equipped with GB300 and has accessed GB300 in Thailand, likely to train its AI models." Nvidia's GB300 chips are a generation behind its frontier Vera Rubin systems, but they are still considered cutting-edge technology. Nvidia, Moonshot AI, the White House and the British embassy of the People's Republic of China were contacted by CNBC for this article, though the text does not record substantive responses from any of the parties at the time of publication.

U.S. export restrictions ban shipping Nvidia's most advanced AI chips to China, although shipping less powerful semiconductors is permitted. There were already prior reports of Chinese companies accessing these chips through third countries, and this case fits that same dynamic: the use of U.S. computing infrastructure outside China's borders to circumvent the ban. In fact, the article recalls that ByteDance had reportedly been amassing computing capacity outside China, according to a March report by The Wall Street Journal.

The episode comes at a moment of intensifying race for AI supremacy between the United States and China. In addition to limiting access to critical hardware, U.S. lawmakers are weighing how to curb the growing adoption of Chinese AI models by Western companies. In parallel, Washington is seeking to close the legal loopholes that allow Chinese companies to access advanced chips by indirect routes: the Remote Access Security Act (RASA), a bipartisan legislative proposal, seeks to expand export controls to also include cloud-based remote access to critical hardware and software. The bill was already passed by the House of Representatives in January and now awaits action in the Senate.

Keegan McBride, director of science and technology at the think tank Tony Blair Institute for Global Change, quoted in the article, noted that "it is well known that China is using U.S. compute both inside the country and abroad" and that this "shows just how far ahead the United States is in compute," adding that "U.S. compute is the foundation of the current AI race." The text also recalls that, although the United States reportedly authorized ten Chinese companies in May to buy Nvidia H200 chips, a senior U.S. trade official noted this month that "very few" had been shipped to China and Hong Kong, which is reportedly pushing Chinese tech companies to increasingly seek their own development alternatives.

Beyond chip access, Kratsios added a second accusation: that Moonshot AI had "distilled" Anthropic's Fable model to develop Kimi K3, in line with prior complaints by Anthropic itself about "industrial-scale" campaigns to extract information from its model. Distillation is a technique by which a smaller model can imitate the performance of a larger, already-trained model, extracting knowledge from the latter. A spokesperson for the British embassy of the People's Republic of China previously responded to CNBC that the country "opposes unfounded accusations and malicious smears against its AI development."

In that same context, U.S. Treasury Secretary Scott Bessent threatened on Wednesday to sanction Chinese companies that carry out distillation attacks, stating on X: "When PRC companies carry out covert, industrial-scale distillation attacks that cross the line into intellectual property theft, sanctions and Entity List designations will be on the table."

Taken together, the case illustrates how control of AI chip supply chains has become the central battleground of the geopolitical dispute between Washington and Beijing, with mutual accusations of intellectual property theft, evasion of sanctions through third countries and legislative pressure to close off remote-access routes to critical computing, while Chinese companies keep unveiling models that, according to the figures provided in the article, come close to or surpass the products of leading U.S. firms.