US Closes AI-Chip Export Loophole for China‑Headquartered Firms

The U.S. Commerce Department’s Bureau of Industry and Security issued guidance on May 31, 2026 clarifying that exports of advanced AI‑computing items to entities whose ultimate parent is headquartered in China or Macau now require BIS licenses, even when the receiving subsidiary is located outside those jurisdictions.

The step restates a licensing requirement first introduced on Nov. 17, 2023 and folded into the AI Diffusion Rule in January 2025, and it closes a practical enforcement gap that arose after BIS announced a non‑enforcement posture for parts of the Diffusion Rule in May 2025.

Reporting by Reuters and other outlets named U.S.‑made models in the scope of the guidance, citing industry and government sources — including families of NVIDIA Rubin and Blackwell processors and AMD’s MI350x class of accelerators as covered advanced items that will generally require licenses when ultimately destined to China‑headquartered firms.

BIS said the clarification was prompted by questions about whether the preexisting, parent‑headquarters license rule continued to apply to affiliates located in third countries; the agency answered that it does, effectively closing a route companies had used to re‑export advanced processors through foreign subsidiaries.

The move has immediate consequences for cloud providers, hyperscalers, and original equipment manufacturers that ship GPUs and servers across borders and to corporate customers in multiple jurisdictions. Companies now must screen buyers by ultimate parent location, not only by shipping address, before exporting ECCN‑listed hardware.

BIS also added an operational caveat: bona fide data‑center operators that are otherwise compliant don’t need to stop using, servicing, or disposing of controlled advanced computing items because of the guidance until BIS provides further notice. That carve‑out is meant to limit disruption to running cloud services.

Industry sources quoted by Reuters and others estimated the gap may have permitted substantial volumes of advanced chips to reach China‑linked projects through overseas subsidiaries — one supply‑chain source said the totals could be in the hundreds of thousands, though such figures are not publicly verifiable.

For exporters the practical rules are unchanged in form but stricter in effect: advanced computing items identified under EAR ECCNs such as 3A090 and related 4A090 categories remain controlled, and license applications should be submitted unless a narrow license exception applies. The BIS guidance points exporters to EAR parts that govern license filings and voluntary disclosures.

Legal and compliance teams will face a sharper burden because screening must account for corporate ownership and the headquarters of ultimate parents, not just the immediate purchaser’s registration country. License review timelines and the availability of exceptions will shape whether firms can complete planned sales or must delay shipments for BIS review.

The industry reaction was cautious. Public statements from major chipmakers and cloud providers were limited at publication time, and Reuters noted earlier reporting that some approvals to sell less capable models to Chinese firms had not led to completed deliveries — a reminder that export approvals, on paper, do not always translate quickly into hardware deployments.

This guidance fits into a broader U.S. strategy since 2022 to restrict China’s access to the most advanced AI compute and to close workarounds such as offshore routing and remote access to cloud‑hosted chips. Policymakers and market participants will now watch how BIS implements license reviews and whether further agency action or congressional measures follow.