Comments will be due 60 days after publication, but the rule will be in force during the comment period. In the accompanying press release, BIS stated that the interim final rule (IFR) was written “by and at the direction of the Department of the Treasury and National Security Council,” which is an unusual statement potentially signaling more active involvement by the White House and the US Department of the Treasury in high-profile export control issues.
The rule imposes a new worldwide, catch-all provision requiring a license for the export, re-export, or in-country transfer of any “advanced Super Intelligence (SI) items”[1] when there is “knowledge” that the items are destined for end use in data centers—wherever located—that will provide direct or indirect access to a “significant volume of computing power to or for an entity headquartered in, or whose ultimate parent company is headquartered in” either Macau or a destination specified in Country Group D:5 (including China) by the US Export Administration Regulations (EAR).
SCOPE
The new Remote Access Rule applies to exports, reexports, or in-country transfers to any data centers worldwide—acting individually or “linked” together (broadly defined under the rule)—that will provide such end users with access to more than 12 million “aggregate total processing performance (TPP).”
For context, NVIDIA publishes 1,080 petaFLOPS of dense FP4 compute for a 72-GPU GB300 NVL72 rack, or roughly 4.3 million TPP per rack. On that basis, three GB300 racks (just over 200 GPUs) would exceed the threshold, as would roughly 760 H100 or H200 GPUs.
The threshold measures hardware performance, not GPU-hours, tokens, or spend, and capacity across “linked” data centers is aggregated. The rule is not limited to training: BIS describes the concern as compute for “training or running” advanced models, so inference capacity could also count.
More importantly, it also reaches indirect arrangements, including where a data center’s non-D:5 customer is training models for, or providing models to, a D:5 company. The trigger is where the customer or its ultimate parent is headquartered, not where a contracting entity is incorporated.
‘KNOWLEDGE’-BASED COMPLIANCE AND ENFORCEMENT
Compliance and enforcement under the new end-use restriction will turn on “knowledge,” a term defined under the EAR to include not only actual (i.e., confirmed) knowledge but also “reason to know” and “an awareness of a high probability.” This terminology originates from the Model Penal Code and was adopted by the EAR in 1996 after it was added to the US Foreign Corrupt Practices Act in 1988, and in the latter context has been a driver of significant enforcement and investigative activity since 2004.
Foreign data centers “will be expected to perform robust due diligence on their customers to identify direct or indirect ownership or control by Country Group D:5 or Macau persons.” BIS specifically advises:
To satisfy the requirement that you do not have “knowledge” for purposes of [the new “Prohibition on exports, reexports, or transfers (in-country) for certain advanced computing items for use in data centers for end uses of concern”], you will be expected to obtain and be able to provide sufficient information from the foreign data center regarding that data center’s “know your customer” due diligence checks.
Such diligence checks will include diligence on whether any data center customers are subject to direct or indirect ownership or control by persons located in, headquartered in, or whose ultimate parent is headquartered in, Macau or a destination specified in Country Group D:5.
A foreign data center declining or being unable to provide information sufficient to verify compliance with this end use control shall be considered a red flag that presumptively provides you with “knowledge” unless you are able to resolve that red flag before proceeding with the transaction.
Accordingly, making and defending risk-based due diligence decisions, which are often subjective judgment calls based on imperfect information, will be critically important for trade compliance professionals advising both exporters and data centers. These requirements flow down to subsequent owners of the items that were not parties to the original transaction and renew if there is a change in end use.
INSTALLED CHIPS AND THE 90-DAY TRANSITION
The Remote Access Rule is not limited to future shipments. The act of changing from a nonprohibited to a prohibited end use can now be treated as an in-country transfer that requires authorization. Exporters, reexporters, and transferors must notify recipients of this requirement.
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 those services. Notably, as drafted, the rule requires that a license be obtained, not merely applied for, within that period.
LIMITS ON DE MINIMIS TREATMENT AND LICENSE EXCEPTIONS
BIS clarifies that covered items incorporated into other products are not eligible for de minimis treatment and that no license exceptions are available under the new restrictions. License applications will be reviewed on a case-by-case basis.
The Remote Access Rule provides a narrow exclusion for computing capacity used exclusively for a “commercial recommender system,” subject to a submission to BIS identifying the covered customer, aggregate TPP provided, and due diligence performed. Generative systems do not qualify for this exclusion, even if they also make recommendations.
Importantly, exporters must separately assess the restrictions under § 744.23(a)(3), which apply independently of the new 12 million TPP threshold. As BIS explains, those restrictions may apply where a covered entity is the direct or indirect purchaser, or the exclusive or primary (more than 50%) remote user, of data center capacity.
NEW ADVANCED COMPUTING RESTRICTIONS ON EXPORTS TO BRAZIL, INDONESIA, MALAYSIA, SINGAPORE, AND THAILAND
In addition to the above end-use restrictions, the Remote Access Rule enacts a separate, country-specific license requirement for advanced computing item exports to Brazil, Indonesia, Malaysia, Singapore, and Thailand that mirrors the same November 2023 license requirement imposed on Country Groups D:1, D:4, or D:5 (including China).
This requirement applies regardless of the 12 million TPP threshold. The new licensing requirement does not apply where the ultimate consignee or end user is headquartered in the United States, or has a US-headquartered ultimate parent, provided the exporter, reexporter, or transferor files monthly reports with BIS under new § 743.10.
LONG-ANTICIPATED FORMAL RECISSION OF THE AI DIFFUSION RULE
The Remote Access Rule formally rescinds the prior Framework for AI Diffusion adopted during the final days of the Biden administration, following the Trump administration’s announcement in May 2025 that it would not enforce the framework.
The rescission eliminates the framework’s associated license exceptions, including License Exception Low Processing Performance and License Exception Artificial Intelligence Authorization, the Data Center Validated End-User authorization, and the controls on closed AI model weights classified under export control classification number (ECCN) 4E091.
The rescission should not be treated as a relaxation of compute controls, since BIS retains the advanced computing hardware controls and License Exception Advanced Compute Manufacturing.
WORLDWIDE ADVANCED COMPUTING FOREIGN DIRECT PRODUCT RULE REMAINS, WITH NEW EMPHASIS
While the AI Diffusion Rule is rescinded, BIS is “retaining the worldwide destination scope of the advanced computing foreign direct product (FDP) rule in § 734.9(h)(2)(i) if there is ‘knowledge’ that a foreign-produced item is destined worldwide or will be incorporated into any ‘part,’ ‘component,’ ‘computer,’ or ‘equipment’ not designated EAR99 to any location worldwide.”
In the Remote Access Rule, BIS underscores the “expansive application” of this particular FDP rule, emphasizing that when an integrated circuit (IC) subject to ECCN 3A090 is incorporated into a foreign-produced component (e.g., a GPU board or server), the IC remains subject to BIS jurisdiction. According to BIS, “[t]hus, even third parties who only receive a 4A090 item [such as a GPU board or server] are on notice that such items contain an ECCN 3A090 IC.”
TIGHTER PRESUMPTION FOR FOUNDRIES AND OSATS
The Remote Access Rule also narrows how front-end fabricators and outsourced semiconductor assembly and test (OSAT) companies can overcome the presumption in Note 1 to ECCN 3A090.a. The transistor-count attestation routes are no longer available where the IC designer is located in, or has an ultimate parent headquartered in, Country Groups D:1, D:4, or D:5 or any of the five newly controlled countries.
New § 744.23(a)(3)(i)(C) separately requires a license where a fabricator or OSAT exports, reexports, or transfers, to or within any destination, an ECCN 3A090.a IC designed by a company that is neither an approved nor an authorized IC designer.
IMPACT ACROSS THE SI (AI) ECOSYSTEM
The rule regulates covered hardware and the persons who export, reexport, or transfer it; it does not license cloud, inference, or agent services as such. The licensing obligation falls on parties that ship, hold, or control the hardware, while the commercial effect reaches every service that depends on that compute. The impact of the AI ecosystem differs:
- Operators of covered hardware outside the United States are the most directly affected: BIS identifies the data center operator as the party that must obtain authorization when the end use changes. Colocation operators that do not own the accelerators are in a less settled position because capacity and customers are measured at the data center level.
- A provider operating its own hardware abroad stands in the same position as a cloud provider. One that rents capacity is a potential intermediary, and the rule does not say whether ordinary application programming interface (API) access for covered customers counts, or how aggregate TPP is attributed to users of a shared, token-metered service.
- A provider that only consumes inference through an API ordinarily has no licensing obligation of its own but should expect suppliers to flow down end-customer disclosure and usage restrictions; it may be an intermediary if it provisions GPU capacity or runs training or fine-tuning jobs for covered customers.
- Investors and lenders are not directly regulated but bear the resulting offtake, collateral, and covenant risk.
BEWARE LEGACY ASSUMPTIONS
It is critically important to read this new Remote Access Rule in the context of geopolitical developments, national developments, and enforcement trends. Companies should expect BIS, for example, to look to enforce the due diligence requirements for their own sake, regardless of whether there was a provable violation of other provisions of the Remote Access Rule or other US export controls.
Companies should also expect BIS to continue to leverage the “acting with knowledge” prohibition (15 CFR 764.2(e)) and General Prohibition 10 (taking any action, essentially, with “knowledge” of a past, present, or future violation) in enforcement actions.
CONGRESSIONAL INVESTIGATIONS AND PARALLEL OVERSIGHT
The Remote Access Rule also arrives amid heightened congressional scrutiny of advanced semiconductor exports, third-country diversion risks, and remote access to US-controlled computing capacity. Congressional committees have recently sought information, documents, and briefings concerning advanced semiconductor transactions, customer due diligence, and the adequacy of US export control enforcement.
For example, congressional scrutiny has extended directly to companies, including a June 2026 inquiry by Senator Elizabeth Warren into NVIDIA’s export control compliance practices and public representations concerning potential semiconductor diversion; most recently, Senators Elizabeth Warren and Christopher Coons requested preservation of executive branch records relating to advanced semiconductor exports and potential diversion.
Companies should anticipate that congressional inquiries may extend to what companies knew, when they knew it, how they identified and addressed potential red flags, whether commercial considerations influenced compliance decisions, how those decisions were escalated and documented, and what was communicated to government officials.
Such scrutiny may implicate semiconductor manufacturers, exporters, cloud service providers, data center operators, and other supply chain participants, and may proceed independently of BIS or US Department of Justice investigations.
Accordingly, companies should consider congressional investigative risk alongside traditional export control exposure, including the importance of contemporaneous documentation, appropriate record preservation, protection of privileged communications, and coordination of responses to potentially overlapping government inquiries.
NEXT STEPS AND HOW WE CAN HELP
Companies should assess their exposure to the Remote Access Rule by mapping the location of covered hardware, contractual counterparties, ultimate beneficiaries of computing capacity, aggregate capacity across linked facilities, and existing export authorizations before entering into new commitments or modifying existing arrangements.
Wholly US-hosted computing warrants a separate jurisdictional analysis because an in-country transfer is a change in end use or end user within a foreign country; however, a US provider’s ownership does not exempt its overseas facilities.
Open questions for comment or BIS guidance include whether the 12 million TPP threshold is measured per customer or across all covered customers, how shared or token-metered computing capacity is attributed, and how the 90-day transition period will operate in practice.
Morgan Lewis is at the forefront of helping clients design and implement risk-based due diligence and other controls in the context of risk-based export controls and other regulatory regimes. Our lawyers are advising on the Remote Access Rule’s scope, customer diligence, contract terms, license applications, and comments to BIS, and can help clients turn the rule’s broad “knowledge” standard into a documented, defensible compliance process.
While a broad definition of “knowledge” makes US government enforcement easier by eliminating the need to prove “actual knowledge” of a violation, it can also be a principled basis on which companies subject to US export controls can make and defend risk-based compliance decisions.
For questions about this development, please contact the authors of this LawFlash.