The US government is closing a gap that has been debated for years: under a new AI chip remote access rule, renting advanced US computing power through data centres in third countries now requires a licence in many cases. On 16 November 2026, the Bureau of Industry and Security (BIS) at the US Department of Commerce released an interim final rule. It is meant to stop companies from China and other restricted countries from simply renting powerful US chips instead of buying them. The rule also adds new licence requirements for five countries, including Singapore and Malaysia.
What BIS announced on 16 November
According to the BIS press release, the rule was “written by and at the direction of” the Department of the Treasury and the National Security Council. Its purpose is to ensure that existing restrictions on sensitive US technology cannot be circumvented through remote computing arrangements involving data centres in third countries. BIS points to progress in cloud computing and AI infrastructure, which has made it increasingly possible to use advanced US computing capacity remotely, including through data centres outside the countries that face export restrictions.
The rule text (RIN 0694-AK50) is titled “Addressing Remote Access to Advanced Semiconductors”. One detail stands out: based on Executive Order 14434, the rule consistently uses “Super Intelligence” (SI) instead of “Artificial Intelligence”. It refers to the same high-performance chips and servers that were previously controlled as AI hardware.
AI chip remote access rule: the new licence requirement in detail
The core of the rule is a new section, § 744.24, in the US Export Administration Regulations (EAR). A licence will be required to export, reexport or transfer advanced chips or servers when there is knowledge that they will be used in a data centre that supplies significant computing power to a specific group of customers. The key points:
- Threshold: the rule covers data centres that provide more than 12 million “aggregate TPP” (total processing performance), either at one site or across digitally linked sites that operate as a unified whole.
- Customers: it targets companies headquartered in Macau or a Country Group D:5 destination, as well as firms whose ultimate parent company is headquartered there.
- Worldwide scope: the requirement applies wherever the data centre is located.
- Indirect routes: a customer outside that group also counts if it trains models for a covered company, participates in such training or provides models to it.
- Review: licence applications are reviewed case by case.
To put the threshold in context, TechCrunch estimates that 12 million TPP roughly equals the maximum computing power of about 750 Nvidia H100 or H200 chips. According to the report, Group D:5 includes China, Russia, Iran and North Korea, plus 18 other countries.
The rule also applies after delivery. If an overseas data centre starts supplying significant computing power to such companies only after the hardware was shipped, the owner, the operator or whoever transfers the hardware also needs a licence. Data centres that already provided such services before publication and want to continue must obtain a licence within 90 days of the rule taking effect.
Due diligence for operators and exporters
The rule shifts part of the enforcement burden to data centres, foreign foundries and chip exporters. They must vet their customers, including any direct or indirect ownership or control by persons in D:5 countries or Macau. Through an “is informed” process, BIS can also notify companies individually or generally that an unacceptable risk exists. According to the rule text, licence exceptions cannot overcome the new prohibition unless BIS states otherwise.
There is one explicit carve-out: computing power used exclusively for commercial recommender systems is not covered by the prohibition. This means systems such as those used by online retailers and social networks to rank, filter or suggest products, services or content. The exception comes with reporting requirements, and it does not cover systems designed or used to generate new text, images, video or software.
New licence requirements for five countries
The second part of the rule concerns shipments to specific countries. Exports, reexports and transfers of advanced semiconductors to Indonesia, Malaysia, Singapore, Thailand and Brazil will require a licence. The Commerce Department says these countries present a “higher-than-normal risk of diversion”. The Secretary of Commerce can add further countries in consultation with the Secretary of State or the National Security Advisor.
Chips destined for companies headquartered in the United States, or with a US ultimate parent, are exempt from this five-country requirement, provided certain reporting requirements are met. For countries that tolerate evasion, the rule lists possible consequences: import restrictions, additional licence requirements, a downgrade in the export control framework, and Entity List designations or temporary export bans for the people involved.
What else changes
Two further changes are less visible but matter for the semiconductor industry. First, the rule tightens the presumptions that contract chipmakers and packaging providers face when exporting advanced-node logic chips. Chip designers in countries subject to licence requirements will have fewer ways around that hurdle.
Second, the rule formally rescinds the regulations of the “AI Diffusion Rule” published by the Biden administration on 15 January 2025. That framework set country-tier caps on chip exports. According to BIS, it was never enforced, and the agency had announced its rescission in May 2025. So the new rule does not bring back the old tier model; it relies on targeted controls by customer, end use and destination instead.
Timeline: what is still open
The rule takes effect when it is published in the Federal Register, which also starts a 60-day comment period. The released rule text still contained placeholders instead of concrete dates. The 90-day window for existing data-centre services also starts only once the rule takes effect. Anyone who may be affected should wait for the final Federal Register version and plan deadlines from there.
Politically, the rule lands at a tense moment. TechCrunch links it to the resignation of Nvidia CEO Jensen Huang, which we covered in our article on the Nvidia CEO transition. The report also notes that Reuters had previously said Chinese AI companies were rushing to secure computing capacity abroad. Officially, however, BIS justifies the rule with national security in general terms, not with any single case.
What it means for cloud providers and developers
For most developers in Europe, little changes right away. The rule targets large-scale computing capacity and customers headquartered in specific countries. It is a different story for operators of GPU clouds and data centres that run US hardware and serve international customers. They need to know who their customers are, who owns them and what the computing power is used for. That also applies to brokers and intermediaries who resell capacity.
Companies that use or build data centres in Southeast Asia or Brazil also face the new licence requirement for chip shipments. That can delay projects even when the end customers have no link to D:5 countries. How much this affects GPU prices and availability remains to be seen. Financing of the AI buildout is already under scrutiny, as the latest Fed Financial Stability Report shows, and large providers keep adjusting their plans, as the case of Microsoft and Oracle on AI cloud capacity recently showed.
In practice, anyone who offers or brokers GPU capacity should review customer checks and contract clauses, document ownership structures and watch for the Federal Register publication. Anyone who only buys capacity should expect providers to ask for more information about their company and intended use.



