microsoft.com web signal

Microsoft, NVIDIA, Meta lead open-weights push, OpenAI joins

TL;DR

  • A coalition of about 25 U.S. tech companies published a July 24, 2026 letter arguing open-weight AI models are essential to American AI leadership.
  • Signatories include Microsoft, NVIDIA, Meta, IBM, Dell, Palantir, Andreessen Horowitz, Hugging Face, Y Combinator, Mozilla, Mistral and the Linux Foundation.
  • Anthropic and Google did not sign; OpenAI was absent at launch and reportedly joined the following day, taking the roster to around 35.

A coalition of American tech companies put its name to a joint policy letter on July 24, arguing that U.S. AI leadership will not be judged by any single frontier model, but by whether the country builds "a broad, open ecosystem that diffuses into every sector of the economy." The framing is the interesting part: the pitch is not about who owns the best model, it is about how widely capability spreads.

The substance is a mix of economics and safety. Open-weight models, the ones anyone can download, inspect, and run on their own infrastructure, let organisations "build on advanced models without training one from scratch or paying frontier-model prices for every task," the letter argues. It also claims "open weights create rivalry not only among model developers but across cloud chips, applications, and services," and that customers who use them "will not become locked into a single provider." The safety case is the more contested one, phrased as "openness may be one of the most important paths to AI safety and security," on the theory that a broad community can inspect models for vulnerabilities.

The signatory list is where the politics show. According to The New Stack, the launch roster of about 25 included Microsoft, NVIDIA, Meta, IBM, Dell, Palantir, Andreessen Horowitz, Hugging Face, Y Combinator, Mozilla, Mistral and the Linux Foundation. Anthropic and Google were not on it. OpenAI was also absent at launch, then reportedly joined the following day alongside Cisco, Cohere and GitHub, taking the count to around 35, per FourWeekMBA.

The honest caveat is that this is a manifesto, not a settled argument. The safety claim in particular cuts both ways: weights, once released, cannot be recalled, and the closed-frontier labs' position is that this makes open models harder to secure over time. The reporting also does not give you the coalition's policy asks in enforceable form, what capability thresholds would trigger review, who runs the evaluation, or how post-release incidents get coordinated.

The forward-looking read is who this arms. If open weights stay unrestricted in the U.S., the beneficiaries sit a layer above the frontier labs: startups on Hugging Face, enterprises on Dell and IBM hardware, and the model builders like Meta and Mistral who already ship weights. It is a coalition drawn along exactly that line.

Shared on Bluesky by 2 AI experts