Meta, Microsoft, Nvidia and Two Dozen Others Ask Washington to Shield Open-Weight AI

Two dozen companies and institutions, spanning Big Tech, chipmakers, venture capital, and open-source nonprofits, have put their names behind a joint letter asking US policymakers not to restrict open-weight AI models. Titled "Open Weights and American AI Leadership," the letter lands at a moment when Washington is actively weighing export controls and safety rules that could touch how these models are built and distributed.

The signatory list is notable for its breadth. Meta, Microsoft, Nvidia, IBM, and Dell Technologies appear alongside cybersecurity firm CrowdStrike, data-analytics company Palantir, enterprise software maker ServiceNow, and AI players Hugging Face, Perplexity, and Mistral. Investors Andreessen Horowitz and Y Combinator signed too, as did the Linux Foundation and Mozilla, giving the effort a foothold in both commercial and nonprofit open-source circles.

Open-weight models differ from closed systems like those OpenAI and Anthropic offer strictly through paid API access. With open weights, the trained parameters themselves are published, letting anyone download, inspect, modify, and run the model on their own infrastructure rather than renting access to someone else's servers.

The letter also defends distillation, the technique of training a smaller model to mimic the outputs of a larger, more capable one, framing it as a legitimate and established engineering practice rather than a shortcut worth regulating. It draws a direct line to the decades-long open-source software movement to make that case.

On the security front, the letter's argument centers on parity: defenders trying to detect and simulate AI-driven attacks need models with capabilities comparable to what attackers might use, something closed, API-gated systems don't readily provide. Restricting open weights, the signatories contend, could leave defenders at a structural disadvantage.

Economically, the pitch is about market structure. The letter argues open weights lower the barrier to entry for startups and public-sector institutions, sharpen competition throughout the AI stack, and let enterprises sidestep vendor lock-in tied to a single closed provider.

Microsoft's involvement drew particular scrutiny given its own commercial position. CEO Satya Nadella addressed the balance on LinkedIn, writing that success comes down to using "the right model for each task" while managing costs, a framing that dovetails neatly with Azure's strategy of hosting both open and closed models for customers to mix and match.

"The right model for each task" while keeping costs down, Nadella wrote, tying the open-weight push to Microsoft's own multi-model cloud strategy.

That overlap between principle and business interest hasn't gone unnoticed. Critics have pointed out that a policy environment favoring open weights also happens to reinforce Azure's role as the infrastructure layer where those models, open or closed, ultimately run.