Policy & Regulation

BIS Rule Targets Remote Access to Advanced AI Chips

BIS released an interim final rule on November 16, 2026 requiring licenses for data centers worldwide providing over 12 million aggregate TPP of advanced U.S. AI compute to Chinese and other D:5 entities.

BIS Releases Rule Targeting Remote Access to Advanced AI Chips
BIS Releases Rule Targeting Remote Access to Advanced AI ChipsAI-generated
By Rebecca Stone7 min read

Why it matters

  • BIS released the interim final rule on November 16, 2026, with a licensing threshold above 12 million aggregate TPP.
  • Existing data centers serving covered companies must obtain a license within 90 days of the rule's effective date.
  • The rule adds license requirements for advanced-semiconductor transfers to Indonesia, Malaysia, Thailand, Singapore and Brazil.
  • The rule formally rescinds the never-enforced Biden-era AI Diffusion Rule, announced in May 2025.
  • A commercial recommender system exception excludes generative models that produce novel text, images, audio, video or software.

The U.S. Commerce Department's Bureau of Industry and Security released an interim final rule on November 16, 2026 that requires a license for any data center anywhere in the world providing more than 12 million aggregate total processing performance (TPP) of advanced American AI compute to Chinese companies and other restricted entities. The rule closes the overseas data-center loophole that let firms in Country Group D:5 destinations, including China and Macau, rent advanced computing remotely instead of physically importing restricted chips.

BIS wrote the rule at the direction of the Treasury Department and National Security Council. Its stated purpose is to prevent remote-computing arrangements in third countries from circumventing existing physical-shipment controls on advanced semiconductors. The regulation matters well beyond the chip industry: it converts export controls from a border checkpoint into a regime governing who ultimately consumes computing power, how capacity is aggregated across facilities, and what customers do with models after shipment.

What triggers a license under the new rule?

The licensing threshold of more than 12 million aggregate TPP applies when a person knowingly exports, reexports or transfers covered items to a data center anywhere in the world that provides — or will provide — that level of computing access to companies headquartered in Macau or a Country Group D:5 destination, including China, or whose ultimate parent is headquartered there. Aggregate TPP is defined as the sum of total processing performance across the covered integrated circuits or commodities.

License applications under the data-center end-use control will be reviewed case by case, giving BIS discretion rather than a fixed approval standard.

The threshold can be reached at a single data center or across linked data centers. Linked facilities include digitally or physically connected sites that share resources, balance workloads and replicate data as a unified whole. The definition pulls distributed infrastructure into one capacity calculation, so a company cannot fragment its compute footprint across several nominally separate sites to stay under the limit.

How does the rule reach indirect access?

The control covers more than direct customers in restricted countries. A data-center customer outside Country Group D:5 can bring a transaction within scope if it trains models for, participates in training models for, or supplies models to a D:5 company. This extends the rule's reach to intermediaries that never touch restricted territory themselves but feed restricted firms with the outputs of advanced compute.

Exporters and foreign data centers are expected to conduct know-your-customer checks covering direct and indirect ownership or control by Macau or D:5 persons. The rule specifies a presumption with teeth: a data center's refusal or inability to provide sufficient verification information is a red flag that presumptively supplies the knowledge triggering the prohibition, unless the exporter resolves the concern before proceeding.

Compliance obligations continue after the original shipment. Changing covered equipment from a non-prohibited data-center use to a prohibited one counts as an in-country transfer requiring authorization. For shipments made without authorization because the exporter lacked knowledge of a prohibited end use at the time, the exporter must tell the recipient that any later change of use requires BIS authorization.

Data centers already providing significant compute to covered companies before the rule's publication must obtain a license within 90 days of its effective date to continue operating, or face future export restrictions. The on-display document leaves its Federal Register publication date, effective date and comment deadline as placeholders, so it does not yet establish a calendar date for the requirements to take effect.

What is the commercial recommender exception?

The prohibition does not apply when computing access is exclusively for a commercial recommender system and the required information is submitted to BIS for sharing with the End User Review Committee. The exception covers systems exclusively designed and used to rank, filter or suggest commercial products, services or content.

The boundary is narrow. The exception excludes systems that generate novel text, images, audio, video, software or other content, including recommendation products that incorporate such models. A shopping site's ranking engine qualifies; a recommender that embeds a generative model does not.

The submission to BIS must identify the covered company, report the sum of TPP provided, and describe the due diligence establishing exclusive recommender use.

Which new destinations face license requirements?

The rule separately adds license requirements for covered advanced-semiconductor exports, reexports and transfers to or within Indonesia, Malaysia, Thailand, Singapore and Brazil. Commerce identifies these five destinations as presenting a higher-than-normal diversion risk — a signal that U.S. agencies suspect restricted chips are moving through Southeast Asian and Latin American hubs on their way to China.

Applications will be reviewed case by case, with consideration of each destination's export-control regime and its enforcement coordination with BIS.

One carve-out applies. Transactions for ultimate consignees or end users headquartered in the United States, or whose ultimate parent is U.S.-headquartered, are excluded from the five-destination requirements — but only if the exporter meets new reporting duties. Monthly reports must identify transaction parties and their parents, destinations and end users, product classifications, names and stock-keeping units, TPP, quantities and transaction dates. Reports are due by the first business day of the month following the transactions they cover.

What changes for foundries and the AI Diffusion framework?

The rule narrows the routes by which front-end fabricators and outsourced semiconductor assembly and test companies can overcome the presumption that applicable advanced logic chips from certain non-approved or non-authorized designers fall under the advanced-chip classification. Designers located in, or with ultimate parents headquartered in, specified D:1, D:4 or D:5 destinations — or in the five new high-risk destinations — lose that eligibility. Foundries and packaging firms serving those designers now face a higher classification bar.

At the same time, the rule removes the former worldwide license requirement imposed on the covered items for regional-stability reasons and recasts coverage through destination and end-use controls. It formally rescinds regulations implementing the Biden-era AI Diffusion Rule, which was never enforced and whose rescission BIS announced in May 2025. The rescission eliminates:

  • The model-weights control and its foreign-direct-product rule
  • The AI Diffusion country tiers
  • The Artificial Intelligence Authorization license exception
  • The Low Processing Performance license exception
  • The Data Center Validated End User Authorization

The AI Diffusion Rule had sought to cap how much advanced AI compute each country could receive under a tiered global structure. Its formal removal, combined with the new end-use controls, marks a shift in U.S. strategy: from limiting compute by geography to limiting it by who ultimately holds the keys.

Commerce's stated rationale is that large-scale advanced computing can support military and intelligence activities, including weapons development, cyber-espionage, targeted cyber disruption, algorithmic theft and weaponized disinformation.

Why the rule matters

The practical consequence is a compliance regime focused on three things the previous framework did not capture: the ultimate users of computing power, aggregated capacity across linked facilities, and later customer arrangements that can change the regulated end use after shipment. Cloud contracts, model-training partnerships and resale arrangements now sit inside the export-control perimeter.

For data center operators outside the United States, the 90-day licensing clock for existing service to covered companies is the immediate operational issue. For chip exporters, the five new destination controls and monthly U.S.-parent reporting duties add administrative weight to every shipment through Southeast Asia and Brazil. And for AI labs, the rescission of the AI Diffusion Rule removes one layer of restriction while the new end-use rule tightens another — meaning the compliance question is no longer only where chips go, but who ultimately computes on them.

Source: unite.ai

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

728 articles

Related articles

  1. New U.S. Rules Close Cloud Loophole in China AI Chip Export Controls
  2. US Moves to Close Cloud Loophole as China Builds Distillation Pipeline
  3. White House tightens AI chip exports to China via overseas data centers
  4. Chinese AI Firms Rush to Lock Up Overseas Nvidia Capacity
  5. Moonshot AI Sought Nvidia Blackwell Capacity in Europe, Sources Say

« Previous articleNext article »