Nvidia's Huang launches an open-weights policy letter that doubled to 50 signers in a day

Jensen Huang posted on X for the first time on July 24 — not a chip announcement, a policy letter. Huang shared 'Open Weights and American AI Leadership,' a document initially signed by 25 companies including Microsoft, Meta, Palantir, Hugging Face, Mistral and IBM, arguing that Washington should not restrict open-weight AI models in the name of national security. Huang framed the stakes in explicitly historical terms, comparing today's push to restrict open weights to licensing fights in the 1980s that, in his telling, nearly strangled open-source software before it could take hold. The post had drawn more than 11 million views within a day.

The letter lays out five arguments for keeping model weights downloadable rather than locked behind an API: lower costs for startups and researchers, preserved competition across model developers, clouds, chips and applications, customer control over infrastructure, and a security case — open models let outside researchers 'audit code and patch vulnerabilities rather than relying on a small number of closed single points of failure.' It closes by drawing a line between model distillation, which it calls 'a legitimate, longstanding research technique,' and unlawful extraction of a competitor's outputs, arguing the two should be handled through targeted legal and commercial measures rather than a blanket restriction on openly released weights. Microsoft CEO Satya Nadella backed the letter the same day, writing that the signers 'are outlining a path for open-weight models to strengthen American competitiveness and expand economic opportunity, while protecting national security.'

The letter landed three days after Treasury Secretary Scott Bessent gave the administration's clearest signal yet that it might move the other way. On Fox Business on July 21, Bessent said the U.S. has found 'watermarks' of American models inside Chinese ones and warned overseas developers could face sanctions for 'stealing from our great companies,' with a decision possible within days or weeks; he separately proposed requiring U.S. companies to disclose when they're running a Chinese model at all. Huang, in an Axios interview the next day, pushed back directly: 'These Chinese models are excellent. Open-source models that are excellent should be used,' he said, calling the odds of a U.S. ban 'zero possibility' and describing distillation — the technique Bessent frames as theft — as 'fundamental to intelligence.'

Who's still not on it

The list moved fast. By July 25 — about a day after Huang's original post — the roster had roughly doubled to 50, picking up OpenAI, Google, AMD, Cisco, Cloudflare, GitHub, Block and Ollama, among others. What didn't move: Anthropic and Amazon are still off the list. That absence reads as coordinated rather than incidental — Amazon is Anthropic's largest outside investor, has committed up to $25 billion to it, and Anthropic is the closed frontier lab most directly exposed if an open-weight surge erodes what it can charge for Claude. Google's own signature, despite backing Anthropic too, makes Amazon and Anthropic's shared holdout more conspicuous, not less.

The letter's timing lines up with a stretch in which the fastest-moving open-weight releases have come from Chinese labs, not American ones. Moonshot AI's Kimi K3, Alibaba's Qwen3.8-Max-Preview and DeepSeek's V4 have all shipped or previewed within the same few weeks, each pitched as competitive with, or a fraction of the cost of, closed Western flagships. That's the backdrop the letter argues against restricting: if U.S. policy makes it harder to release open weights domestically while Chinese labs keep shipping them anyway, the signatories' case is that America cedes the open lane entirely rather than protecting it. Whether that's the right read of the security tradeoff is exactly what Bessent's camp disputes — but it's the argument that pulled 25 companies onto a letter within hours of each other, and another 25 to add their names the next day.

The underlying split explains the lineup better than any single company's statement does. Nvidia's business gets bigger the more players train and run models on its chips, whether the weights are open or closed, so an ecosystem with dozens of open competitors is a win for the company that sells the shovels — the same logic pulled in cloud and infrastructure names like Cloudflare and Cisco once the letter had visible momentum. A closed frontier lab has the opposite incentive: restrictions that slow well-resourced, fast-moving open competitors protect the premium a closed model can charge. Huang's own framing — that the world needs 'both frontier closed models and frontier open models' — is also a hedge specific to Nvidia: it sells chips to both sides of that split and loses if either is regulated out of existence.

Nothing here is settled. The letter is industry advocacy aimed at Washington, not a change in policy, and neither the Commerce Department nor Bessent's Treasury had issued a public response as of publication. Bessent's own sanctions review — the one citing watermark evidence — is still pending on his stated 'days or weeks' timeline. The coalition is betting that a public letter which doubled its signer count in a day can shape that decision before it's made; whether it does is the actual test of this week's post, not the post itself.