The Nvidia open weights letter published Friday asks U.S. policymakers to avoid early restrictions on downloadable AI models, putting Nvidia, Microsoft, Meta and 22 other organizations on record as Washington weighs a tougher line on Chinese AI systems.
Nvidia CEO Jensen Huang promoted the three-page letter in his first post on X, writing that the world needs both closed frontier models and open frontier models. The timing is pointed: the letter came days after reports that the Trump administration was renewing interest in a ban on Chinese AI models. The letter does not name China, Moonshot AI or DeepSeek.
The signers include Nvidia, Dell Technologies, Microsoft, IBM, Box, ServiceNow, CrowdStrike, Palantir, Telnyx, Replit, Perplexity, Andreessen Horowitz, Y Combinator and Emergence Capital. Meta, Mistral, Black Forest Labs, Arcee AI and Reflection are also on the list, and all already release model weights. The Linux Foundation signed as well.
Three of the most prominent U.S. AI labs are absent: OpenAI, Anthropic and Google.
What is the Nvidia open weights letter asking for?
The letter asks Washington not to impose what it calls premature limits on AI models whose weights can be downloaded. In practical terms, open weights are the model parameters that let users run or adapt an AI system outside the company that built it, usually on their own infrastructure or through another cloud provider.
The companies argue that open models support safety work, cybersecurity testing, competition and national AI capability. They also ask policymakers to expand access to compute for startups and researchers, and to fund shared datasets and evaluation systems.
That request tracks Nvidia’s commercial interests, though the company is hardly alone here. Huang said at Nvidia’s CES 2026 press Q&A that open models now account for one in every four generated tokens. Those models can run on enterprise clusters, regional clouds and on-premises racks, all places where buyers may be purchasing GPUs instead of relying only on a few large API providers.
The Linux Foundation’s presence also connects the letter to Nvidia’s own model work. The foundation oversees the OpenMDW-1.1 license that Nvidia used in June for Nemotron 3 Ultra, a 550-billion-parameter model. Artificial Analysis rated that model at 47.7 on its intelligence index, compared with 53.9 for Moonshot’s Kimi K2.6.
Why is distillation part of the fight?
The letter also addresses distillation, the technique of training one model using outputs from another. The signers argue that policymakers should not treat distillation itself as theft, while saying unlawful extraction from closed systems should be handled with targeted legal rules.
That passage lands in the middle of a broader dispute over Chinese AI models. Treasury Secretary Scott Bessent said on Fox Business earlier this week that the administration would examine Chinese open-source models for intellectual property theft and could sanction companies behind them. Bessent said officials had found watermarks from U.S. large language models in Chinese systems.
Huang took a different line in an Axios interview two days later, saying U.S. companies should be allowed to use Chinese models and calling claims about Chinese backdoors a misconception. The letter’s distillation language is the one part that moves beyond open weights, which makes its inclusion hard to miss.
This story draws on original reporting from Tom's Hardware.
