Analysis
Nvidia is simultaneously deepening hardware support for Chinese open-weight AI models like DeepSeek and Qwen and warning investors that a potential White House crackdown on those same models could hurt its business, CNBC reported. The dual posture reflects how tangled Nvidia's incentives have become as both the dominant hardware supplier to U.S. AI labs and a company whose chips run the open Chinese models increasingly competing with those same labs.
Nvidia isn't alone in this position. Microsoft, Meta, Palantir and more than 20 other companies co-signed a July statement urging policymakers to avoid "premature restrictions" on open-weight models -- a category that includes both U.S. open releases (Meta's Llama family) and Chinese ones (DeepSeek, Qwen), and the companies pushed a country-agnostic argument: restrict the category and everyone building on open weights loses, not just Chinese labs.
Huang went further personally, posting on X for the first time on July 24 specifically to argue for the importance of open models in expanding AI access globally. That's a notable departure from Huang's typical avoidance of overtly policy-flavored public statements, and it signals how seriously Nvidia views the threat of restrictions that could cut off a meaningful slice of the open-model ecosystem its chips serve.
“Huang went further personally, posting on X for the first time on July 24 specifically to argue for the importance of open models in expanding AI access globally.”
The current White House security review framework, as reported, exempts open-weight models from scrutiny entirely, instead focusing on closed, proprietary systems from OpenAI, Anthropic and Google's DeepMind. That's the opposite of what China hawks inside the administration have been pushing for -- restricting Chinese-origin open models specifically -- and creates a live policy fight playing out inside the same administration Nvidia has to navigate for its own export licenses.
David Sacks, an AI adviser to Trump, has publicly argued that restricting Chinese open models would actually undermine U.S. competitiveness, since American companies increasingly build on top of DeepSeek and Qwen rather than avoiding them -- a genuinely contrarian position inside a Republican administration otherwise focused on China decoupling, and one that puts him closer to Nvidia's position than to the export-control hawks.
The stakes for Nvidia are concrete: it needs Chinese developers using its hardware to run whichever models -- American or Chinese -- dominate open-source adoption, because chip sales don't care which country's model runs on top. A crackdown that pushes Chinese labs and developers toward domestic chip alternatives, even inferior ones, is a longer-term threat to Nvidia's China revenue than the current DeepSeek-Qwen dynamic, which at least keeps developers on Nvidia silicon.
What to watch is whether the administration's open-model exemption survives the internal fight Sacks and the restriction-minded faction are having, particularly as it intersects with Nvidia's own China export-license negotiations playing out on a separate track.