A coalition of 25 technology companies published an open letter on July 24, 2026 urging the United States government to avoid what it calls premature restrictions on open-weight AI models. Titled "Open Weights and American AI Leadership," the letter was signed by Nvidia, Microsoft, Meta, IBM, Dell Technologies, Palantir, Mistral, Hugging Face, Mozilla, The Linux Foundation, Andreessen Horowitz and Y Combinator, among others. OpenAI, Anthropic and Google did not sign.
Nvidia chief executive Jensen Huang amplified the letter in the first post he has ever made on X. "Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty," he wrote, adding that "the world needs both frontier closed models and frontier open models." The post drew millions of views within hours.
Key Highlights
- 25 signatories published the letter on July 24, 2026, hosted on Nvidia's and Microsoft's corporate sites
- OpenAI, Anthropic and Google are absent — the three labs most associated with closed frontier models
- The letter explicitly defends distillation as a legitimate research technique, asking for targeted legal frameworks instead of sweeping bans
- It arrives as the Trump administration weighs restricting US access to Chinese open-weight models, days after Moonshot AI released Kimi K3 on July 16, 2026
- Huang's amplification marked his debut post on X, an unusually direct political intervention for the Nvidia CEO
What the Letter Argues
The coalition's central claim is that American AI leadership should not be measured by a single frontier model. As the letter puts it, US leadership "will be judged not by one frontier model, but by whether America builds a strong, open ecosystem."
Four arguments run through the document. First, access: open weights let companies, universities and public institutions build on existing models instead of training from scratch, and let them match a cheap specialized model to a cheap task rather than routing every request through a frontier API. Second, competition: a broad base of open models sustains rivalry across chips, clouds and applications, and reduces vendor lock-in. Third, sovereignty: organizations and countries that can download and run weights locally keep control of their own data and deployment decisions. Fourth, security: the letter argues that "openness may be one of the most important paths to AI safety and security," on the grounds that closed systems concentrate points of failure while open models can be audited by a far wider community of researchers.
The letter also leans on a historical analogy. Software developed by the open-source community now underpins most of the internet, the signatories note, despite early institutional skepticism in the 1980s. Their warning is that AI faces the same fork in the road and could get the answer wrong.
The Distillation Clause
The most policy-sensitive passage concerns distillation — training a model on the outputs of another model. The letter draws a line between lawful distillation, which it treats as a foundational research practice rooted in open-source tradition, and unlawful extraction or misappropriation. It asks Washington for targeted legal frameworks aimed at the latter rather than sweeping restrictions that would criminalize a technique used across the field.
That framing is not abstract. White House adviser Michael Kratsios had accused Moonshot AI of using distillation to replicate a US model, an allegation that put the technique itself at the center of the policy debate.
Why Now
The timing tracks a specific sequence of events. Moonshot AI released Kimi K3, an open-weight model that independent testing placed near the frontier, on July 16, 2026. Senior US officials responded by raising concerns about Chinese open-weight models, and the administration reportedly began weighing restrictions on their use inside the United States.
The competitive backdrop is stark. Chinese open models reached roughly 41 percent of downloads on Hugging Face in spring 2026, and Alibaba's Qwen family passed one billion cumulative downloads by March 2026, overtaking Meta's Llama as the platform's most-downloaded open model. An Andreessen Horowitz partner has estimated that around 80 percent of US AI startups now build on Chinese base models.
Read against those numbers, the letter is as much a competitiveness argument as a philosophical one: if American open models are restricted or never funded, the default open model for the world's developers is likely to be Chinese.
The Split in the Industry
The absence of OpenAI, Anthropic and Google is the letter's loudest silence. All three have built their businesses around closed frontier models, and both OpenAI and Anthropic have warned Washington about the risks posed by capable Chinese open models. Anthropic in particular has publicly argued that broad open-weighting of frontier capabilities is a dangerous path.
Commercial interests are visible on the open side too. Microsoft's advocacy aligns with Azure Foundry Managed Compute, which hosts open-weight models inside Azure's perimeter with Microsoft curating which models qualify. Nvidia benefits directly whenever more organizations train and serve their own models on its hardware. Hugging Face, Mistral, Mozilla and The Linux Foundation have more straightforwardly ideological stakes.
Impact
For developers and businesses outside the US — including across the MENA region — the outcome of this debate shapes practical questions: which model weights can legally be downloaded and self-hosted, whether sovereign deployments on local infrastructure remain viable, and how much of the AI stack can be run without a dependency on a handful of American API providers. Organizations that have standardized on open weights for data-residency or cost reasons have a direct stake in where Washington lands.
What's Next
The letter is an input to an ongoing policy process, not a decision. Washington has not published a rule on Chinese open-weight models, and the distillation question remains open. The signatories are asking for restraint and targeted enforcement; the closed labs are asking for guardrails. Whichever way the administration moves, it will set the terms for how open-weight models are built, shared and deployed well beyond 2026.