# AI Chip Export Controls US rules, administered by the Commerce Department's Bureau of Industry and Security (BIS), that restrict the sale of advanced AI chips, chipmaking tools and related technology to China and other destinations. ## The main instruments - **Performance thresholds.** Since October 2022, chips above set limits on compute and interconnect speed need a licence to ship to China. Nvidia responded with China-specific parts such as the [[Nvidia H800]], which the October 2023 update then also caught. - **Entity List.** A named list of foreign companies that US firms cannot supply without a licence. Adding a Chinese lab or cloud provider cuts it off from US hardware and, through the foreign direct product rule, from much foreign hardware built with US technology. - **Tool and fab controls.** Limits on lithography and other equipment, aimed at slowing China's domestic chip production. - **Hosting and procurement rules.** Newer proposals target where models may be run or bought, for example bans on using certain foreign models in government systems. ## What they did and did not do Controls cap access to the best silicon. They do not cap the use of cloud capacity rented outside China (Malaysia, Thailand, Japan), the flow of model outputs ([[API Distillation]]), or the spread of [[Open Weights]] that already exist. The second-order effect is the article's point: constraint pushed Chinese labs into [[Systems and Compiler Depth]], and that kind of capability compounds. > [!note] Timing matters > Every control applies to what happens next. Weights already released cannot be recalled, so rules land first on the parties still inside the loop, often American companies building on those weights. Related: [[Chokepoints in the AI Stack]], [[China - Risky Bet or Unique Opportunity]], [[The Cross-Pacific AI Chessboard]]