Jensen Huang used his first X post to argue open-weight AI is a safety feature

NVIDIA CEO Jensen Huang broke a years-long X silence this month to argue that open-weight AI is a safety and cybersecurity feature, not a risk. "Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty," Huang wrote in his first-ever X post. "The world needs both frontier closed models and frontier open models."

Open-weight AI means systems whose trained parameters — the numerical weights that determine how a model produces its outputs — are released publicly. That sets them apart from OpenAI's or Anthropic's flagship models, which stay behind an API. A user can download an open-weight model, run it on their own hardware, and modify it. The trade-off is that once the weights are out, anyone can fine-tune them or strip the safety guardrails the original lab put in place.

The same week as the X post, NVIDIA published a position paper titled "Open Weights and American AI Leadership" and launched the Open Secure AI Alliance, an industry coalition built around the same argument. The intervention lands as regulators in Washington and Brussels write the rules that will decide whether open-weight models face the same export controls, safety disclosure requirements, and pre-deployment review as closed frontier systems. Which definition of "safe openness" wins will shape which companies can compete, which models can ship, and which countries can build sovereign AI stacks.

Anthropic has taken the other side. The company has published a position on open-weight models arguing that the risks of open-weight frontier models — including the inability to recall once distributed — require targeted controls such as restricting chips to authoritarian jurisdictions, cracking down on model distillation, and requiring safety testing for all frontier-class systems. Anthropic has said it has never advocated for a blanket ban on open-weights models.

A separate letter signed by tech leaders and aimed at training "Uncle Sam" on the value of open-weight AI has also circulated in Washington. The Register reports that Huang did not sign on, even though NVIDIA is publicly aligned with the letter's argument.

NVIDIA sits in the training chokepoint for nearly every frontier model, open or closed. If regulators decide that "open-weight" is policy-acceptable for frontier-class systems, the ruling expands the total market for training compute and standardizes on the chip stack NVIDIA already dominates. If regulators decide the opposite, the closed-frontier labs — also NVIDIA customers — get a defensive moat, and Huang's other large customers get a regulatory gift.

Huang's first tweet in years was a public entry into a regulatory fight where his company sits at the training chokepoint for every frontier model. The comment period that resolves which definition of "safe openness" gets written into the rules is open now.