On 7 October 2022 the US Department of Commerce’s Bureau of Industry and Security (BIS) announced a new set of export controls aimed at China’s access to advanced computing chips and the equipment used to make them. The interim final rule was published in the Federal Register on 13 October 2022 (87 FR 62186), under the title “Implementation of Additional Export Controls: Certain Advanced Computing and Semiconductor Manufacturing Items; Supercomputer and Semiconductor End Use; Entity List Modification.” It is the policy most often described as the opening move of the modern chip war.
The rule did several things at once. It restricted the export to China of the most capable AI accelerators (defined by performance thresholds covering compute power and chip-to-chip interconnect bandwidth), broadened controls on advanced semiconductor manufacturing equipment, added new restrictions tied to supercomputer and semiconductor end uses, placed twenty-eight Chinese entities on the Entity List, and required US persons to obtain licenses before supporting the development or production of certain integrated circuits in China. A Temporary General License was included to limit near-term supply-chain disruption for goods ultimately destined elsewhere.
The performance thresholds mattered enormously for industry. Nvidia’s top data-center GPUs of the period exceeded the limits, so the company designed cut-down export variants (the A800 and later H800) to stay under the bandwidth ceiling and keep selling into China. Subsequent BIS rules in October 2023 and December 2024 tightened the thresholds further, progressively closing those gaps. The controls are the reason “how many chips, and which ones, can a Chinese lab legally buy” became a central question in AI strategy.
The controls form the backdrop to several later library entries. When DeepSeek released competitive models reportedly trained at low cost, part of the story was that it had done so under hardware constraints these rules created. The episode is a clear case of compute itself, rather than algorithms or data, becoming the lever of state power over AI. The rule does not ban AI research or restrict software and model weights; it is an export-control measure targeting physical hardware and the people and tools that produce it.