BIS Advanced-Computing + Semiconductor Export Controls (Oct 7 2022)
regulationAI & Compute · Money & Finance · Nations & Geopolitics
A single U.S. rule quietly put the entire global computer-chip industry under Washington's national-security thumb.
Who they are
The U[1].S. advanced-computing and semiconductor export controls issued October 7, 2022 by the Commerce Department's Bureau of Industry and Security.
What they do
The engine treats it as the regulatory leash binding the world's chip makers to U.S. security goals.
How it works
The rule restricts exports of advanced computing chips and chip-making tools, mainly aimed at China and Russia; it was expanded in January 2023 to cover Macau and updated again in 2023 and 2024. Its original trigger measured chip interconnect speed (600 GB/s), but when Chinese-market chips like the A800/H800 were tweaked to slip under that line, Commerce dropped the interconnect metric in October 2023 and switched to performance density. Chip giants like Nvidia, ASML, TSMC, Micron, SK Hynix, and Samsung all operate under this oversight.
Why it matters
The engine flags it as a codified carve-out lever — an announced rule that governs the whole compute industry — and shows the state watching chip capability so closely it kept rewriting the rule as companies tried to dodge it.
What's new (Aug–Oct 2026)
Aug 19, 2026 — Small batches of Nvidia H200 chips were allowed into mainland China, Business Standard reported; Tom's Hardware reported on Aug 21 that ByteDance and Tencent each received about 10,000 H200s, with China's planning agency (NDRC) approving each purchase. (business-standard.com ; tomshardware.com)
Aug 24, 2026 — Taiwanese prosecutors charged nine people, including one from Nvidia and two from Super Micro; 74 servers reached China. (theguardian.com)
Sept 9, 2026 — C4ADS posted its "covert compute" report, dated September 9, 2026. (c4ads.org)
Oct 2, 2026 — The South China Morning Post reported on a report co-written by Nicholas Brown and Saif Khan saying older DUV chip-printing machines "can be adapted to manufacture 7-nm logic chips and advanced memory for AI processors". (scmp.com)
Aug 28, 2026 — Reports said the Trump administration is preparing a rule aimed at Chinese access to remote AI servers, such as renting Nvidia chip time through Thailand and Singapore. The rule is a draft and the Commerce Department has not confirmed it. (tomshardware.com ; techtimes.com)
The engine's record — word for word
Issued October 7 2022 by the Bureau of Industry and Security (Commerce Department) as amendments to Export Administration Regulations (EAR). Federal Register 2022-21658. Restricts exports of advanced computing integrated circuits, computer commodities containing such ICs, and certain semiconductor manufacturing items — primarily targeting China + Russia. Expanded January 18 2023 (Federal Register 2023-00888) to include Macau (closing the autonomous-region loophole). Updated October 17 2023 + December 2024 with additional restrictions. Engine relevance: the regulatory architecture binding the global compute oligopoly to US national-security objectives at the export-licensing tier. Nvidia, ASML, TSMC, Micron, SK Hynix, Samsung all operate under direct Commerce/IC oversight per this statute. Per
Report #93 H4 findings #021 + #022 — primary-source verified via Federal Register. [Seam: Aligned-To-Whom? codified-exemption instance — an announced rule decoupled from operating reality by a written carve-out / waiver / immunity / 13G-passivity / tax-exemption (the master-key lever). (b)+(c); intentional-single-operator gated.] [Report #154] (Report #154) The Oct 7 2022 rule's original ECCN 3A090 trigger was INTERCONNECT bandwidth >=600 GB/s — the state identifying data-transit, not raw compute, as the capability vector. When the A800/H800 severed interconnects to slide under it, BIS REMOVED interconnect as a metric (Oct 17 2023) and pivoted to Performance Density (TPP/die-area). The interconnect throttle is the compute-layer lever of two_tier_ai / AI Capability Bifurcation (interconnect_binding_constraint).
[Live pass Aug 11 2026] Third loophole-to-rule-pivot iteration: BIS is reviewing Chinese firms' legal OFFSHORE rental of Nvidia compute (IAPS estimates >=60% boost to China's effective access in 2026), building two country lists — the interconnect gate bypassed by remote access, announcement layer controlling while the substrate flows.
[Live pass Oct 6 2026] ~2026-08-19 — Small batches of Nvidia H200 chips allowed into mainland China (Business Standard 19 Aug); ByteDance and Tencent each took delivery of roughly 10,000 H200s under NDRC per-purchase approval (Tom's Hardware 21 Aug). (
business-standard.com ;
tomshardware.com) | ~2026-08-24 — Taiwanese prosecutors charged nine people (Monday) including one from Nvidia and two from Super Micro; 74 servers reached China. (1 source) (
theguardian.com) | 2026-09-09 — The C4ADS report page (c4ads.org/reports/covert-compute) carries the date "September 9, 2026". (1 source) (
c4ads.org) | 2026-10-02 — SCMP: report co-authored by Nicholas Brown and Saif Khan says DUV tools "can be adapted to manufacture 7-nm logic chips and advanced memory for AI processors". (1 source) (
scmp.com)
[Live pass Oct 6 2026] News bearing on what this record said to watch for "BIS is reviewing Chinese firms' legal OFFSHORE rental of Nvidia compute": 2026-08-28 — Reports: Trump administration preparing a rule targeting Chinese access to remote AI servers (renting Nvidia GPU time via Thailand and Singapore). Draft, not confirmed by Commerce. (
tomshardware.com ;
techtimes.com)
Follow the trail
Walk this on the live map →