Publication

BIS issues rule restricting remote access to advanced AI chips

AI-digital.jpg AI-digital.jpg

On November 16, 2026, the Bureau of Industry and Security (BIS) issued an Interim Final Rule (IFR) titled “Addressing Remote Access to Advanced Semiconductors.” The rule amends the Export Administration Regulations (“EAR”, 15 C.F.R. Parts 730–774) to expand restrictions on Chinese and other companies’ access to advanced computing power through third-country data centers, while emphasizing that certain remote-access arrangements were already subject to licensing.

The IFR also creates new destination-based export license requirements for advanced compute integrated circuits (ICs) destined to countries that BIS views as diversion hubs and removes methods for certain foreign IC designers to overcome the presumption that their chips are classified under export control classification number (ECCN) 3A090.a. Notably, it formally rescinds the Biden-era “AI Diffusion Rule,” which BIS announced would cease to be enforced in May 2025.

The rule takes effect on publication in the Federal Register, expected to be November 19, 2026. Public comments will be accepted for 60 days after publication. Because the rule takes effect on publication, companies should assess their exposure before the comment period closes.

The IFR has four main components:

  • Data center controls – A new license requirement for advanced chips going to data centers anywhere in the world that supply large amounts of computing power to Chinese (or other Country Group D:51 or Macau) headquartered companies

  • Five new diversion-risk countries – A new license requirement for advanced chips going to Indonesia, Malaysia, Singapore, Thailand and Brazil

  • Foundry and chip packaging changes – Reducing the methods available for a front-end fabricator (fab) or Outsourced Semiconductor Assembly and Test (OSAT) providers to overcome the presumption that an “applicable advanced logic IC”2 is classified under ECCN 3A090.a

  • Rescission of the AI Diffusion Rule – Removal of the earlier framework’s broad worldwide licensing structure, country tiers, model weight controls and data center validated end-user (VEU) arrangements in favor of more concentrated restrictions on particular destinations, beneficiaries and uses.

New controls on data centers serving China-linked customers

  • What is restricted – When the rule takes effect, a license will be required to export, reexport or transfer specified advanced computing chips, servers and related items subject to the EAR where the transaction party has “knowledge,” as defined in the EAR, that the items are destined to a data center that is providing or will provide companies headquartered in China, Macau or another Country Group D:5 destination – or whose ultimate parent is headquartered there – access to more than 12,000,000 aggregate total processing performance (TPP).3 The EAR’s knowledge standard extends beyond actual knowledge to include awareness of a high probability, conscious disregard and willful avoidance.

  • Linked facilities count – The threshold applies across “linked data centers,” including facilities connected physically or digitally to share resources, balance workloads and replicate data as a unified whole. Spreading capacity across sites that satisfy this definition does not avoid the control.

  • Intermediaries are covered – The rule also reaches access provided indirectly, for example to a non-Chinese customer that is training or supplying AI models for a Chinese company.

  • The obligation travels with the chips – Later reexports and in-country transfers remain subject to review even where the current holder did not participate in the original export. Incorporation into a foreign-made server does not necessarily remove the chips from EAR jurisdiction. A later change to a covered end-use can require authorization. For covered transactions made on or after effectiveness without authorization because the transaction party lacks knowledge of the covered end-use and no other license requirement applies, the supplier must notify recipients of the authorization and customer-diligence requirements applicable to subsequent changes in use.

  • Existing relationships – The IFR directs data centers already providing significant computing power to covered customers before publication to obtain a license within 90 days of effectiveness to continue those services. This provision is not a general grace period. Importantly, BIS says existing controls already covered certain purchasing and exclusive or majority remote-use arrangements, and encourages voluntary self-disclosures where violations occurred. Companies should, therefore, review relevant past transactions as well as future activity.

  • License review – BIS will review applications on a case-bycase basis.

  • Commercial recommenders – A conditional exclusion covers computing access used exclusively for narrowly defined commercial recommender systems, such as tools recommending existing products or content. Generative AI systems do not qualify. Reliance requires a submission to BIS, and the exclusion does not override separate requirements under § 744.23.

Heightened due diligence expectations

The IFR shifts significant compliance responsibility onto chip suppliers and data center operators. Suppliers are expected to obtain, and be able to produce, information about each foreign data center’s know-your-customer (KYC) due diligence processes, including whether a data center’s customers are directly or indirectly owned or controlled by persons located in, headquartered in, or whose ultimate parent is headquartered in China (including Macau) or other D:5 destination. If a data center refuses or is unable to provide that information, this presents a red flag that the exporter must resolve prior to shipping. BIS states that this diligence is meant to prevent companies from evading the rule through corporate restructurings or spinoffs.

These requirements operate independently of the separate Affiliates Rule, whose implementation is suspended until January 10, 2027. The new control can apply to data center customers that are not listed parties, based on their headquarters or their ultimate parent’s headquarters.

New license requirement for five countries

  • When the rule takes effect, a license will be required to ship advanced AI chips and related items to or within Indonesia, Malaysia, Singapore, Thailand and Brazil, which BIS identifies as presenting a higher-than-normal risk of diversion. Applications will be reviewed on a case-by-case basis, considering the strength of each country’s export controls and its cooperation with BIS.

  • Limited exclusion for US-headquartered recipients – The five-country destination requirement does not apply where the ultimate consignee or end-user is headquartered in the US, or its ultimate parent is, provided the exporter, reexporter or transferor submits monthly reports to BIS identifying the transaction parties and relevant parent companies, destinations, product details, quantities and transaction dates. Separate end-use and other licensing requirements still apply.

Changes for foundries, chip packagers and fabless IC designers

These changes apply to Note 1 to ECCN 3A090.a, and are independent of the new remote-access control restrictions described above. Since 2025, fabs and OSAT providers seeking to export, reexport or transfer (in-country) applicable advanced logic ICs have faced a presumption that the chips fell within ECCN 3A090.a. The note provided three methods to overcome the presumption: (i) attestation from an approved or authorized IC designer (as defined in the EAR), (ii) in cases where the fab is also the OSAT, attestation from the fab that the chips did not meet certain technical specifications and (iii) attestation from an approved OSAT (as listed in the EAR) that the chips did not meet certain technical specifications. The IFR, however, removes pathways (ii) and (iii) when the IC designer is located in, or its ultimate parent is headquartered in, D:1, D:4, D:5 (including China) or one of the five newly covered countries. The separate pathway for approved or authorized designers remains.

Rescission of the AI Diffusion Rule

The IFR formally removes the AI Diffusion Rule’s three-tier country framework, its controls on AI model weights and related license exceptions and data center authorizations. In its place, BIS has adopted what it calls a “more focused and effective” approach based on targeted destination and end-use controls. Under the specified regional-stability review policies in § 742.6, covered transactions involving D:5 destinations and Macau generally face a presumption of denial, while other covered D:1/D:4 destinations and the United Arab Emirates generally receive a presumption of approval, subject to specified exceptions and other applicable controls. These policies should be distinguished from the case-by-case review specified for the new remote-access control.

Who is affected

  • Cloud service providers and data center operators outside the US, especially in Asia, the Middle East and Latin America, and those that sell capacity through resellers

  • US hyperscalers and AI infrastructure companies with overseas facilities, particularly in the five newly listed countries

  • Chip designers, server manufacturers and distributors involved in covered transactions, which face customerdiligence and recipient-notice requirements

  • Fabs and OSATs that serve designers based in affected countries

  • AI developers and corporate users that buy largescale computing through third-party data centers, and companies with ownership ties to China

Practical takeaways

  • Assess data center customers – Data center and cloud operators outside the US should identify China-linked customers, including customers reached through resellers. Operators should also check whether the capacity those customers use exceeds the rule’s threshold.

  • Address existing “commercial recommender” services promptly – Affected operators of data centers that provide compute capacity for commercial recommender systems should promptly assess the 90-day existingservices provision and any licensing obligation. The IFR obligates data center operators to obtain a license within that period; submitting an application alone is insufficient.

  • Strengthen customer diligence – Data center operators should capture where their customers and those customers’ ultimate parents are headquartered. Chip and server suppliers should obtain this information before shipping.

  • Confirm classifications – Chip designers, server manufacturers and distributors should confirm their products’ export classifications. Where a classification is uncertain, they should consider requesting a formal BIS classification (CCATS).

  • Prepare for reporting and customer reviews – Exporters relying on the exclusion for US-headquartered customers in the five new countries should be ready to file monthly reports. Foundries and chip packagers should review their designer customers based in affected countries.

  • Update contracts and check shipments in transit – Companies across the supply chain should build the required notices and end-use terms into their agreements. They should also confirm that shipments relying on the rule’s savings clause are completed by December 19, 2026 (30 days after expected publication in the Federal Register on November 19, 2026). The savings clause applies only to qualifying shipments losing license-exception or No License Required (NLR) eligibility that were already enroute aboard a carrier to a port under actual orders on publication.

  • Consider commenting – Companies with significant data center or foundry exposure should consider submitting comments within the 60-day window, including on aspects of the rule that may need clarification.

If you have any questions about the IFR or its impact, please contact a member of the International Trade & Foreign Investment team.


  1. See Supplement No. 1 to Part 740, Title 15.

  2. “Applicable advanced logic integrated circuits” are logic integrated circuits produced using the “16/14 nanometer node” or below, or using a non-planar transistor architecture. 15 C.F.R. Part 772.

  3. The IFR adds a new definition for “aggregate TPP” to mean the sum of the TPP (2 × ‘MacTOPS’ × ‘bit length of the operation’, aggregated over all processing units on the IC) of all of the ICs or commodities specified in ECCNs 3A001.z; 3A090 (except for 3A090.c); 4A003.z; 4A004.z; 4A005.z; 4A090; 5A002.z; 5A004.z and 5A992.z in the group.