🔍 Read the full analysis: Does Remote Access To AI Chips Require A US Licence? on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
The U.S. Commerce Department’s Bureau of Industry and Security issued a rule on November 16 requiring licences for certain advanced AI chips used by foreign data centres, including facilities that remotely serve customers in China and other restricted destinations. It is scheduled to take effect November 19, 2026; a 90-day period applies to existing customer relationships, not the rule’s start date.
The U.S. Commerce Department’s Bureau of Industry and Security (BIS) has issued a rule requiring licences for certain advanced chips supplied to data centres that provide computing access to companies in China and other restricted destinations, including when the data centres are located elsewhere. The rule is scheduled to take effect November 19, 2026, and addresses a gap in export controls that previously focused more directly on where chips were shipped than on who used them remotely.
The interim final rule, titled “Addressing Remote Access to Advanced Semiconductors,” adds a data-centre test to the Export Administration Regulations. A licence is required for covered chips when a data centre provides, or will provide, more than 12 million aggregate TPP of computing power to companies headquartered in China, Macau or another country in the U.S. restricted group D:5, or whose ultimate parent is based there. TPP, or total processing performance, is the U.S. measure used for chip computing power.
The threshold applies to data centres anywhere in the world. BIS says capacity is counted across a single facility or linked data centres, including digitally linked infrastructure. The rule also covers some indirect arrangements, such as a customer outside D:5 that trains models for, participates in training for, or provides models to a D:5 company. It applies to chips in export-control categories 3A090 and computers containing them in category 4A090, including when a covered chip is incorporated into a foreign-made server or GPU board.
A separate provision requires a licence for every covered chip export to Indonesia, Malaysia, Singapore, Thailand or Brazil. Shipments to companies headquartered in the United States, or with a U.S. ultimate parent, can qualify for an exception in those five countries if the exporter submits monthly reports. Commercial recommender systems, such as online retail recommendations, are exempt from the data-centre test; chatbots and image-generation services are not included in that stated exemption.
US export rule · explainer
How the new rule closes the cloud route to US AI chips
Buying the chips was already restricted. Renting them abroad was the gap.
The route (schematic)
Before: rent instead of buychips shipped to a data centre in a third country, computing power sold remotely New: a licence gateabove 12,000,000 aggregate TPP for companies headquartered in China, Macau or other D:5 countries Already serving them? 90 daysto obtain a licence, or face future export restrictions
What the rule does, step by step
- BuyCompanies headquartered in China already need a licence to buy advanced US AI chipsexisting rule
- RentRemote access through data centres in third countries was the workaroundthe gap
- GateNew licence test for data centres anywhere above 12,000,000 aggregate TPPsection 744.24
- CheckExporters need the data centre’s know-your-customer checks; no answer is a red flagdue diligence
- FiveEvery advanced-chip export to Indonesia, Malaysia, Singapore, Thailand or Brazil needs a licence, US-headquartered buyers exceptedcountry test
Source: US Commerce Department, Bureau of Industry and Security, interim final rule “Addressing Remote Access to Advanced Semiconductors”, Federal Register document 2026-23529 (public inspection), pp. 4, 10, 24, 30 and 32; BIS press release, Nov 16, 2026. The diagram is a schematic.
Graphic: Powered by Thorsten Meyer AI
Cloud Access Now Falls Under Export Controls
The rule makes the customer’s access to computing, not only the chip’s shipping destination, relevant to export compliance. Cloud providers and data-centre operators may need to identify who ultimately owns or controls customers and assess whether computing capacity is being used on behalf of restricted companies. That can affect sales, service agreements and the information providers must collect before allowing access.
For customers, the measure could limit access to high-performance computing through facilities outside the United States, even where they cannot import the chips directly. The rule does not amount to a blanket prohibition on all cloud computing or all chip exports to the named countries: the provisions are framed as licence requirements, with specified exceptions. Whether licences will be granted, and how quickly, will shape the practical effect.
Products Worth Considering

Sonos Roam (Gen 1) Portable Smart Speaker, White

DJI Mini 4K Drone Combo with 4K Camera, 3-Axis Gimbal, and Two Batteries

LG 27GR83Q-B 27-inch Ultragear QHD IPS Gaming Monitor
As an affiliate, we earn on qualifying purchases.
How the Rule Sets Its Timetable
BIS released the interim final rule on November 16; the Federal Register scheduled publication for November 19, when it is set to become effective. Because it is an interim final rule, the controls apply on that effective date rather than after a prior public-comment period. Comments are due 60 days after publication, according to the source material.
The rule provides a 90-day window for existing relationships: data centres already serving affected customers must obtain a licence within 90 days of the effective date to continue those arrangements. This is not a 90-day delay before the rule begins. Certain shipments already en route that lose their licence-free status may proceed if completed within 30 days of publication. BIS described the rule as written by and at the direction of the Treasury Department and the National Security Council.
““exploring how to facilitate trusted access for Singapore companies””
— Singapore government
Licensing and Enforcement Remain Open
The rule establishes which arrangements require licences, but the source material does not say how BIS will process applications, how long decisions will take or what share of requests may be approved. The details of enforcement, including how regulators will assess linked facilities and indirect model-training relationships in practice, also remain unclear.
The text could be revised after the comment period. It is also not yet clear whether the U.S. will add countries to the separate country-specific requirement. The stated 12 million TPP threshold and the named-country provisions are the rules described in the issued text, not evidence that every remote computing arrangement involving those markets is prohibited.
Comments and Licence Applications Ahead
The rule is scheduled to take effect upon Federal Register publication on November 19, 2026. Affected exporters and data-centre operators will need to determine whether their facilities, customers and linked capacity meet either licence test, while existing customer arrangements face the 90-day licensing window. Exporters are expected to obtain customer due-diligence information from foreign data centres, including details about who owns or controls their customers; inability or unwillingness to provide it is identified as a red flag.
BIS is due to receive public comments 60 days after publication. Those comments and any later agency action may clarify how the thresholds and exceptions will be applied. Until then, providers and customers will have to work from the published rule, with the pace of licensing and enforcement still uncertain.
Key Questions
Does the rule ban all remote access to U.S. AI chips?
No. It creates licence requirements for specified shipments and data-centre arrangements. It does not state that all remote access or every export to the named countries is prohibited.
When does the rule take effect?
The Federal Register lists November 19, 2026 as the publication and effective date. The 90-day period is for certain existing customer relationships to obtain a licence, not a delay in the rule taking effect.
What does the 12 million TPP threshold cover?
It covers aggregate total processing performance available through one data centre or linked data centres to specified companies headquartered in China, Macau or another D:5 country, including companies whose ultimate parent is based there. The test applies to facilities worldwide.
Which countries have a separate chip-export licence requirement?
The rule separately requires licences for covered chip exports to Indonesia, Malaysia, Singapore, Thailand and Brazil. A stated exception applies to qualifying U.S.-headquartered companies or those with a U.S. ultimate parent, if exporters file monthly reports.
Are chatbots covered by the recommender-system exemption?
No. The source describes an exemption for commercial recommender systems, such as online retail recommendations, and says chatbots and image generators are not exempt under that provision.
Source: ThorstenMeyerAI.com
