Storia
luglio 28, 2026
Moonshot’s Kimi fight turns AI rivalry into a Washington loyalty test
The White House and Treasury Department have accused Chinese AI firm Moonshot of using restricted Nvidia chips and improperly 'distilling' Anthropic's Fable AI model to create its new Kimi K3 model. U.S. officials are investigating the potential intellectual property theft and have threatened sanctions, escalating a debate over the national security risks posed by powerful open-weight AI models from China.
Moonshot AI’s Kimi K3 has quickly become more than a buzzy Chinese model release. In Washington and Silicon Valley, it is now a stress test of what matters more: preserving U.S. technological advantage through competition, or trying to contain a foreign rival through controls.
The immediate spark was Kimi K3’s strong debut. Reports across the tech press described the model as unusually capable, cheaper than many U.S. rivals, and central to China’s increasingly aggressive open-weight strategy.1 Moonshot said demand surged so sharply that it had to pause new subscriptions after GPU capacity was strained.2
That commercial success collided with a harder line from the Trump administration. Treasury Secretary Scott Bessent said the U.S. supports open models in principle, but not when they are allegedly built on stolen intellectual property: “There’s a very technical AI word for it called distillation, but you and I would call it theft.”3 White House technology adviser Michael Kratsios went further, accusing Moonshot of “large-scale, covert industrial distillation” of Anthropic’s Fable and of accessing restricted Nvidia GB300 systems through Thailand.4
But that case is far from settled. TechCrunch reported that several experts doubt distillation alone explains Kimi K3’s performance, especially given how recently Fable became public. “You can’t distill that much data, train a model, and release it in two weeks,” researcher Braden Hancock said.5
The backlash from open-model advocates has been fierce. Critics argue Washington risks protecting a handful of U.S. frontier labs more than U.S. competitiveness itself. Axios reported fears of a de facto ban on Chinese open models, while others warned that “the weaponization of regulatory uncertainty as a competitive tool should be completely unacceptable.”6 On X, Yann LeCun amplified the view that “highly performant open weights frontier models such as Kimi are a competitive threat to OpenAI & Anthropic, but probably for everyone else these are a win.”
7
Even some U.S. figures are trying to lower the temperature. David Sacks wrote that the “Kimi Panic needs to stop” and argued American frontier labs still lead.
8 That leaves the core dispute unresolved: whether Kimi is mainly evidence of Chinese theft, or proof that open competition is eroding the closed-model advantage the U.S. industry has relied on.