The US Commerce Department’s interim final rule curbing China’s remote access to advanced AI chips was “written by and at the direction of the Department of Treasury and the National Security Council,” the agency’s Bureau of Industry and Security said Monday. Under the regulation, any overseas data center or group of linked facilities must obtain a US license before providing companies headquartered in China, Macau or other restricted countries with more than 12 million aggregate total processing performance of compute, equivalent to roughly 760 Nvidia H100s or 200 Blackwell B300 chips. The restriction covers hardware already installed abroad, giving existing cloud operators 90 days to apply for a license, and asks providers to self-disclose past arrangements where a Chinese firm used a majority of a facility’s compute.
The rule also formally repeals the Biden-era AI Diffusion Rule, eliminating its country tier system, Validated End User authorizations for data centers and export controls on AI model weights. In addition, the regulation tightens foundry due diligence rules by requiring Chinese chip designers to obtain a US license to manufacture 16nm or 14nm and more advanced chips at overseas foundries such as TSMC and Samsung. The remote access restrictions exempt compute used solely for commercial recommender systems and do not apply to data centers located inside the US.




