US tightens AI chip exports to China via offshore units

This move is one of the sharpest tightening steps the US has taken since the original October 2022 and 2023 chip rules, because it explicitly targets corporate structures rather than geography. By pulling overseas subsidiaries of China-headquartered firms into the licensing net, Commerce is trying to shut what appears to have been a sizable back-channel for high-end Nvidia and AMD accelerators that are essential for state-of-the-art model training.

For the race to AGI, that means a more uneven global compute landscape. Chinese labs and cloud providers will face more friction and scrutiny when acquiring leading-edge GPUs, pushing them either to accelerate domestic accelerator efforts or lean harder on grey-market channels and model distillation. US- and allied-country labs, meanwhile, get a relative edge in scaling frontier models, but also inherit more responsibility for how those models are governed.

Long term, this kind of control nudges the world toward parallel AI hardware ecosystems: a US–allied stack and a China-centered stack, each with its own chips, toolchains, and safety norms. That fragmentation may slow some global collaboration but will not meaningfully dampen overall investment in powerful systems.