White House Accuses Moonshot AI of Illegally Using Nvidia's Export-Banned Chips
Claire Weston
White House tech-policy chief Michael Kratsios accused Chinese AI firm Moonshot AI of training its Kimi K3 model on banned Nvidia GB300 chips accessed through servers in Thailand, while also alleging large-scale distillation of U.S. models — a sign the U.S.-China AI rivalry now spans both hardware controls and model IP.
What exactly is Moonshot AI accused of?
White House OSTP director Michael Kratsios posted on social media, accusing Moonshot AI of obtaining servers equipped with Nvidia GB300 chips in Thailand to train its latest model, Kimi K3.
The GB300 belongs to Nvidia's Blackwell series, which the U.S. explicitly bans from sale to Chinese companies. This means → even if the chips sit outside China, a Chinese firm using them still violates U.S. export rules.
Kratsios further alleged that Moonshot AI "built a sophisticated internal platform to distill U.S. models at scale."
What is "distillation," and why does the U.S. care so much?
Distillation — using a powerful "teacher" model's outputs to train a smaller "student" model, replicating some capabilities at lower cost — is a common AI technique. In plain terms = it is copying the homework: you never see the source code, but you train on the answers.
Anthropic has specifically accused Moonshot AI of distilling its models; Moonshot AI has not responded. OpenAI has also accused Chinese rivals of systematically using U.S. model outputs.
This reflects a core fear in U.S. tech: open-source does not mean competitors can use your outputs to train rival products. The IP boundary is being redrawn.
What is the U.S. government preparing to do?
Treasury Secretary Scott Bessent stated: "We support open-source models, but we do not support IP theft. If we find overseas models stealing our great companies' work, we have the ability to sanction them."
The Trump administration pledged in April to take further steps against "adversarial distillation," including holding foreign actors accountable.
Both chambers of Congress are advancing legislation to blacklist or sanction Chinese firms found to have engaged in unauthorized distillation. This means → from executive accusations to legislative tools, the U.S. is building a full anti-distillation enforcement chain.
Why did Kimi K3 rattle markets?
Moonshot AI released Kimi K3 last week, claiming it outperforms all competitors except Anthropic's Claude Fable 5 and OpenAI's GPT-5.6.
The release triggered a tech-stock sell-off, echoing the market reaction to DeepSeek last year. In plain terms = investors are not afraid of one model — they are afraid that "Chinese firms catching up at far lower cost" has been validated again.
This is also the direct trigger for the White House accusation — Kimi K3's performance sharpened the question of whether chip bans can actually constrain Chinese AI.
Where is the export-control loophole?
A May memo from the Commerce Department's Bureau of Industry and Security made clear: sales of advanced processors to Chinese companies' overseas subsidiaries are restricted.
The Trump administration is investigating whether firms have exploited loopholes to funnel advanced AI chips to Chinese subsidiaries abroad. This means → Moonshot AI's path to GB300 chips in Thailand is a textbook case of the very loophole under scrutiny.
Nvidia has said it "has always operated on the premise that no such loopholes exist," but did not respond to Kratsios's post.
What to watch next?
Whether Moonshot AI can maintain its overseas compute access under mounting regulatory pressure is the first key variable.
Whether the U.S. government escalates the accusations into formal sanctions will determine if this is a warning shot or an enforcement precedent.
In plain terms = if sanctions land, Moonshot AI's problem is no longer just where its chips come from — it is whether the company can continue to operate at all.
Content is for reference only, not financial advice.