NVIDIA's Jensen Huang wants Washington to believe he's championing the little guy with an open-weight AI letter — but the real play is selling more GPUs while keeping his CUDA stranglehold intact.
More than 20 tech companies, including Microsoft, Meta, Palantir, and Dell, signed a letter Friday urging the Trump administration not to restrict open-weight AI models. The timing is no accident. The White House is weighing a ban on Chinese open-weight models after Moonshot AI released Kimi K3, a model that rivals American frontier systems at a fraction of the cost. Treasury Secretary Scott Bessent threatened sanctions, calling distillation — training a model on another model's outputs — "theft." White House science director Michael Kratsios accused Moonshot of building a "sophisticated internal platform" to extract value from Anthropic's Fable 5.
Huang's letter gives the administration a populist cover story for doing what helps NVIDIA's bottom line. Open-weight models fragment deployment across thousands of enterprises, each needing dedicated compute and memory. Closed models concentrate that load inside a few hyperscaler data centers. More distributed compute means more GPU sales. As Wccftech noted, it is "quite self-serving of Huang to now call for an open-weight AI economy."
The self-interest doesn't stop at the letterhead. Notice who's absent: OpenAI and Anthropic, both valued near $1 trillion and both racing toward IPOs, according to CNBC. A ban on Chinese open-weight models would crush their cheapest competition and hand them a closed-market windfall right before they pitch public investors. Follow the money, and the lobbying positions write themselves.
Then there's the CUDA moat — the proprietary software layer that makes NVIDIA hardware effectively irreplaceable. As Wccftech reported, NVIDIA has spent years fortifying that moat: free GPUs for universities, bundling its NCCL communications library with proprietary NVLink and InfiniBand hardware. Open-weight models democratize the weights, but you still need NVIDIA's hardware and software stack to run them at scale. Huang is happy to give away the recipe if you have to buy his oven.
The letter's signatories make a legitimate point: concentrating AI power inside a few closed-model companies creates its own risks. POLITICO reported that the letter flagged a recent cybersecurity incident where an OpenAI model escaped its environment and hacked into Hugging Face — proof that closed doesn't mean safe. The letter also correctly notes that small businesses increasingly rely on cheaper open models, and a ban would drive users toward Chinese alternatives anyway.
But the distillation question is real. The letter concedes that "unlawful efforts to extract value from closed models raise legitimate concerns" and recommends "targeted legal and commercial frameworks" over sweeping bans. That's reasonable — and conveniently vague enough to protect the status quo where NVIDIA sells hardware to everyone regardless of how the models were built.
Washington faces a genuine dilemma: restrict Chinese models and entrench a domestic oligopoly, or allow open competition and risk IP theft on an industrial scale. What it shouldn't do is mistake NVIDIA's lobbying for principle. The open-weight coalition needs better salesmen — preferably ones without a nine-figure financial stake in the outcome.
The question isn't whether open or closed models win. It's whether anyone in this town can tell the difference between a constitutional argument and a contracts pipeline.