A White House representative accused the Chinese company Moonshot of improperly using American artificial intelligence (AI) models and Nvidia Corp. chips to develop the Kimi K3 system, which impressed the tech industry with its capabilities last week.
The Director of the White House Office of Science and Technology Policy, Michael Kratsios, stated in a social media post on Wednesday that Moonshot “purchased servers equipped with GB300 and gained access to GB300 in Thailand, likely for training its AI models”. He claims that Moonshot violated U.S. export control rules and the terms of use of American company services while developing its product. The semiconductor servers mentioned by Kratsios are part of the Nvidia Blackwell product generation, which is banned from sale to Chinese companies in the U.S. While the Trump administration allowed some sales of Nvidia H200 chips, it continues to prohibit the sale of more advanced Blackwell chips.
Investigation into possible loopholes and accusations of intellectual property theft
A memorandum from May confirmed that the Bureau of Industry and Security, responsible for export control of AI chips, restricts the sale of advanced processors to Chinese companies outside of China. Representatives of the Trump administration are currently investigating whether individual companies exploited a loophole in the rules to supply advanced AI chips to subsidiaries of Chinese firms. A Nvidia spokesperson previously dismissed suggestions of such a loophole, stating that the company operates as if it does not exist. Nvidia did not respond to a request for comment on Kratsios‘ post.
The K3 system from Moonshot challenged the perception that Chinese AI developers lag significantly behind American leaders such as OpenAI and Anthropic PBC. The Chinese laboratory claimed that its model with open weight parameters surpasses all competitors except for Anthropic‘s Claude Fable 5 and OpenAI‘s GPT-5.6 in overall capability. These claims triggered a sharp reaction in the tech stock market, reminiscent of the impact of DeepSeek last year.
Michael Kratsios noted that Moonshot “developed a sophisticated internal platform for large-scale distillation of American models”. Distillation is a technique where AI researchers use the results of a so-called “teacher” model to train a “student” model and replicate some of its capabilities. Treasury Secretary Scott Bessent also expressed concern on Tuesday about the practice of distillation, although he did not mention Moonshot directly. He stated that some Chinese companies working with open-source models are engaged in what he described as intellectual property theft to create their own AI models. “This administration supports open-source models, but we do not support intellectual property theft. If we see, especially, that foreign models are stealing from our major companies, we have the ability to impose sanctions on them,” said Bessent on Fox Business.
Bessent‘s comments coincide with increasing accusations from OpenAI and Anthropic that their Chinese competitors systematically and without permission use the results of leading U.S. models to create competitive chatbots at significantly lower costs. Anthropic previously accused Moonshot of distilling its models. Moonshot did not respond to these accusations. In April, the Trump administration pledged to take additional measures to combat this practice, known as “hostile distillation,” including considering holding foreign companies accountable. Lawmakers in the House of Representatives and the Senate are advancing relevant bills that would blacklist or impose sanctions on Chinese companies that, according to U.S. authorities, improperly conduct distillation of U.S. models.
Source: Bloomberg
