China’s rapid progress in open-source AI has fractured Silicon Valley in a way that few predicted. The release of Kimi K3 by Moonshot AI and the continued global spread of DeepSeek and Z.ai’s GLM models have forced a confrontation that the US tech industry has been quietly avoiding: what, if anything, should Washington do about Chinese open-source AI models that are freely available, widely used, and increasingly competitive with the best American systems? The answer depends entirely on who you ask, and the industry is now openly at war with itself over the question.
Anthropic, OpenAI, and the Distillation Accusation
Anthropic and OpenAI have lobbied regulators in Washington to restrict open-source AI models from China, according to five people close to the discussions cited by The New York Times. Their argument is that Chinese companies like Moonshot AI and Z.ai built their competitive models by “improperly harvesting data” from American AI systems, a practice known as distillation, and that allowing those models to circulate freely in the US market rewards intellectual property theft while undermining the commercial viability of American labs that invest billions in original research.
The argument has found sympathetic ears in the Trump administration. Treasury Secretary Scott Bessent warned that “open source is not open season on American IP,” while Michael Kratsios, Trump’s science and technology adviser, stated that “large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable.” Anthropic has previously accused Alibaba of running the largest AI distillation campaign yet, and the broader accusation of systematic distillation has become a central talking point in the lobbying effort. The specific claim against Moonshot AI is that Kimi K3 was trained in part by distilling outputs from Claude models — an allegation that Moonshot has denied.
Nvidia, Microsoft, Meta, and the Open-Source Coalition
The pushback has been swift and unusually public. Nvidia CEO Jensen Huang made his first-ever post to X, writing that “the world needs both frontier closed models and frontier open models.” Nine minutes later, Microsoft CEO Satya Nadella posted that open-source software was “essential to a healthy AI ecosystem.” Both signed a letter supporting open source, also signed by Meta’s Mark Zuckerberg, Palantir, and IBM executives. Elon Musk posted that “Jensen is right. This has my full support.” Sundar Pichai of Google noted that the company had “long benefited from open source.”
Nearly 200 Silicon Valley startups, calling themselves the Little Tech Association, signed letters urging the Trump administration not to limit access to Chinese open-source models. Their argument is economic: Chinese open-source models have dramatically reduced the cost of AI inference for American startups, enabling products and services that would be unaffordable if built exclusively on expensive US frontier APIs. Venture capitalist Bill Gurley summarized the divide bluntly: “You have two factions fighting over this issue. There’s the people who want OpenAI and Anthropic to own everything, and then there’s everybody else, including customers.”
The commercial stakes for Nvidia are particularly clear. If Chinese open-source models are restricted, the global AI ecosystem becomes more dependent on closed US frontier models, which run on US cloud infrastructure, not on Nvidia GPUs purchased by developers worldwide. Nvidia’s business model depends on developers buying its hardware to run models locally, and Chinese open-source models are a major driver of that demand.
Why China Benefits Regardless of Washington’s Decision
What makes this debate particularly significant is that China is positioned to benefit regardless of the outcome. If Washington restricts Chinese open-source models, it validates Beijing’s narrative that the US is using national security as a pretext to protect American commercial interests, and it accelerates the adoption of Chinese AI in markets outside the US, where no such restrictions apply. If Washington permits Chinese models to circulate freely, Chinese AI companies gain continued access to the world’s largest and most sophisticated developer ecosystem, accelerating their improvement through real-world feedback.
Xi Jinping’s WAIC speech, in which he cast China as a global champion of an open approach to AI, was a direct response to this dynamic. China’s open-source AI models are winning the Global South, and Washington is only now beginning to grapple with the implications. The US officials cited by the Times appear more likely to approach regulating Chinese open-source models individually as a national security issue rather than implementing a blanket ban, a targeted approach that may prove difficult to enforce in practice, given that model weights, once released, are effectively impossible to recall.
The fracture in Silicon Valley over Chinese AI is, in the end, a fracture over who benefits from the open-source model of AI development. China has already made its bet. The US has not yet decided.