When the chief executive of the world's most valuable chipmaker becomes the loudest evangelist for giving AI away, it is worth asking who benefits. Over the past month Jensen Huang has posted on X for the first time to defend open-source AI, published an open letter praising open-weight models, helped launch an AI safety consortium built around them, and this week shipped one. Nvidia's Nemotron 3.5 Lightning is small enough to run on a single graphics card in a home PC, and free for anyone to download, modify, and use without asking permission or paying a cent.
The stated case is high-minded. Open models, Huang wrote, "strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty." They let companies control their own future, spur competition, and push prices down. It is a genuinely reasonable argument, and it happens to be extremely good for Nvidia. Every open model, no matter who builds it or how cheap it becomes, still has to run on something. "Free AI should be great for hardware," Huang told Axios. "Free AI should be great for chips."
This is the quiet logic beneath the recent wave of open-source manifestos. Proprietary models from OpenAI and Anthropic keep the value, and the margin, at the model layer. Open models push that value out to whoever owns the compute. Cheaper models mean more usage, and more usage means more GPUs. For a company that sells the shovels, commoditizing the gold is a feature, not a threat.
The politics gave the movement a sharper edge. The Washington debate caught fire after Kimi K3, a model from China's Moonshot AI, narrowed the gap with the best American systems. Some policymakers worried it was a national-security problem, or the product of intellectual-property theft through distillation, the trick of training a smaller model on a bigger one's answers, and floated whether Chinese model makers could be sanctioned the way chip exports are. The industry's answer, voiced by Huang and by Meta's Mark Zuckerberg days earlier, reframes the panic: the fix is not to restrict open models but to make sure the best open models are American. As Zuckerberg put it, "Our goal should be for American open source models to be the best globally."
It is a tidy alignment of interest and principle. Restricting Chinese open models is hard and possibly futile; racing to out-open them is easier, and it sells more hardware. Nvidia says early users of Nemotron include the security firm CrowdStrike and the legal-AI company Harvey, and it paired the release with a cybersecurity consortium alongside Microsoft, wrapping the giveaway in a safety story.
The open question is whether any of this contradiction matters. A movement can be self-interested and still produce a public good, and cheaper, more controllable, more widely available models are plausibly better for almost everyone except the labs trying to sell access to closed ones. But it is worth remembering, each time the case for open AI is made by someone who profits from it, that the argument and the incentive point in exactly the same direction. When the shovel-seller tells you the gold should be free, he is not necessarily wrong. He is just not disinterested.