Nvidia, Microsoft and Meta warn against open-weight AI restrictions

Nvidia, Microsoft, Meta, Palantir and other technology companies urged policymakers on Friday to avoid open-weight AI restrictions that they said could weaken competition and push innovation outside the United States. The letter, signed by 25 companies, arrives as Washington and Silicon Valley debate how to respond to Chinese open-weight models that are challenging leading U.S. systems.

Open-weight models have become a focal point in the contest over artificial intelligence policy because they can be downloaded, adapted and run on users’ own computing infrastructure. The companies said in the letter that such models help spread the benefits of AI more broadly and reduce dependence on a small group of closed systems.

The statement comes after rising attention on Chinese developer Moonshot AI, whose Kimi K3 model has outperformed advanced U.S. offerings on some industry benchmarks, according to CNBC. The gains have intensified discussion among officials and executives over whether access to Chinese open-weight models should be limited in the U.S.

What are open-weight AI models?

Open-weight AI models are systems whose model weights are made available so users can download them, modify them and operate them on their own infrastructure. That differs from closed models, where the provider typically controls access through its own service and keeps the underlying weights private.

The signatories argued that policy should not treat closed systems as automatically safer. In the letter, they said closed models can still be compromised, misused or fail in ways that outside researchers cannot inspect, and that concentrating advanced AI in a small number of closed systems can add risk.

Nvidia Chief Executive Jensen Huang and Microsoft Chief Executive Satya Nadella each shared the letter on their personal social media accounts.

Why are Chinese AI models part of the debate?

Moonshot AI’s Kimi K3 release sharpened U.S. concerns about whether Chinese companies are using American technology without authorization. U.S. Treasury Secretary Scott Bessent told CNBC on Tuesday that the Trump administration would examine whether Chinese companies had stolen U.S. intellectual property, and said the government had the ability to sanction companies over such theft.

White House adviser Michael Kratsios wrote on X on Wednesday that Moonshot AI had built Kimi K3 by distilling Anthropic’s technology. Distillation is an AI training technique in which a smaller or less capable model is trained using outputs from a stronger existing model.

Kratsios said legitimate distillation has an important role in open AI development, while warning that large-scale covert distillation aimed at taking proprietary U.S. technology was unacceptable. The companies’ letter said concerns about unlawful distillation should be handled through targeted legal and commercial tools, rather than broad limits on methods that contribute to AI development.

Who did not sign the letter?

OpenAI and Anthropic were not among the signatories. CNBC reported that both companies are valued at nearly $1 trillion each and are preparing for potentially large initial public offerings that could come as soon as this year. Anthropic confidentially filed a prospectus with the Securities and Exchange Commission in June, followed days later by OpenAI, according to CNBC.

OpenAI President Greg Brockman told reporters in New York on Thursday that he had not taken part in discussions with the Trump administration about a potential U.S. ban on Chinese open-weight models. Brockman said OpenAI supports broad access to AI and views wider model availability and usage as positive.

The letter framed the issue as a question of U.S. technological leadership. The companies said American leadership in AI would depend on building an open ecosystem that reaches across sectors, rather than on the performance of a single frontier model.

This story draws on original reporting from CNBC.