US curbs Nvidia AI chip exports to Chinese firms via loopholes

This move is another clear sign that advanced AI compute is now treated as strategic infrastructure on par with oil or rare earths. By closing the overseas-subsidiary loophole, Washington is targeting not just Chinese firms inside China but their global corporate networks, making it harder for Chinese AI labs to quietly amass top-tier Nvidia Rubin, Blackwell and AMD MI350x accelerators through third countries.

For the race to AGI, that matters because access to frontier GPUs is increasingly the bottleneck. US labs and cloud providers already enjoy scale advantages; constraining Chinese access further tilts the playing field toward American and allied ecosystems. At the same time, large Chinese buyers will redouble efforts to source domestic accelerators and to optimize for lower-tier or gray-market hardware, which could accelerate indigenous chip and model-efficiency innovation.

The bigger picture is a world where compute, not just algorithms or data, becomes the main lever of AI power. Export controls like this fragment the global market, push more AI R&D into national silos, and harden the link between AI capability and geopolitical blocs. That doesn’t stop progress toward AGI, but it channels where that progress happens and who controls it.