Update: NVIDIA, Meta, Microsoft sign 25-firm open-weights letter
After 200 startups, Big Tech joins the fight against a US open-weight ban — and defends distillation, the trick that keeps challenger models cheap.
Copy markdown
Now the incumbents are on record
On July 24, 25 US companies — NVIDIA, Microsoft, Meta, Palantir, IBM, Dell, a16z, Mistral, Hugging Face, and Y Combinator — signed "Open Weights and American AI Leadership," warning Washington against "premature restrictions." Yesterday it was 200 startups pushing back; now the chipmaker and the hyperscalers are too.
The distillation clause is the real fight
Buried in the letter is a defense of distillation — training a small model on a bigger one's outputs, the cheapest way challenger labs and open projects close the gap without billion-dollar pretraining. A broad ban would hand the closed-frontier labs a moat. If you fine-tune or distill from open weights, this is the clause to watch.
Why it hits your model shelf
The letter answers real moves: export controls have already blocked Anthropic's most capable models, OpenAI is under pressure to delay a top release for security review, and a ban on Chinese open weights is under active consideration. Qwen, DeepSeek, GLM, Kimi, and Llama are what's actually in play.
The split is now explicit
Jensen Huang posted that "open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty" — flipping the security argument. Anthropic, notably absent, holds the opposite: once weights are public you can't un-publish them. Chipmakers, hyperscalers and open shops on one side; the closed-frontier safety camp on the other.