Jensen Huang's Open-Weights Letter Could Lock In CUDA Dominance

The fight over open-weight AI is a fight over inference silicon, and the most aggressive recruiter for open weights is the company that sells the chips open weights run on.

NVIDIA's coalition of more than seventy signatories has endorsed open models as a pillar of American AI leadership. The top AI lab that refused, Anthropic, has the most to gain from breaking NVIDIA's stack.

Open-weight distribution fragments inference across many small deployments. While closed-weight labs race to design their own silicon, the open-weight ecosystem quietly trains the next decade of customers to read and write CUDA. Huang's letter is not an ideology. It is a moat.

Anthropic's refusal makes the pattern legible. Anthropic runs training on Google TPUs and is among the labs most exposed to a future where inference migrates off NVIDIA. NVIDIA's ask, in plain terms, asks the company with the most to gain from breaking its stack to commit publicly to keeping it. The Anthropic researcher's sarcastic reply to Huang, asking when CUDA and GPU drivers will be open-sourced, was the cleanest possible summary of the gap.

The lab that sees through Huang's letter is the lab with the most to lose from it.