New US AI Chip Export Controls Target Remote Access via Foreign Data Centers
Data centers anywhere will need a US license to rent big AI clusters to firms based in China and other embargoed countries, and five more nations face chip license rules.
Published 4 min read

The US Commerce Department has issued a rule that treats renting AI computing power much like shipping the chips themselves. Under the interim final rule released by the Bureau of Industry and Security (BIS) on 2026-11-15, data centers anywhere in the world will need a US license to supply large amounts of advanced chip compute to companies headquartered in China, Macau or other US arms-embargoed countries.
BIS says the goal is to stop existing chip restrictions from being bypassed through cloud and remote computing deals in third countries. The agency also says the rule was written by and at the direction of the Treasury Department and the National Security Council. It takes effect on the day it appears in the Federal Register, with a 60-day comment period after that.
What the new AI chip export controls change
The rule text (RIN 0694-AK50) has four main parts. Following Executive Order 14434, it uses the term "Super Intelligence" (SI) where earlier rules said "artificial intelligence", but it covers the same class of advanced AI chips and servers.
| Part | What changes | Who is affected |
|---|---|---|
| Data center control | License needed when advanced chips go to a data center that gives more than 12,000,000 aggregate TPP of compute to covered companies | Cloud and data center operators, chip exporters, foreign foundries |
| New country licenses | License needed to ship advanced AI chips to Indonesia, Malaysia, Singapore, Thailand and Brazil | Exporters to those five countries; US-headquartered buyers exempt if they file reports |
| Chipmaking | Fewer ways for chip designers in license-required countries to get advanced-node logic chips made | Front-end fabs and chip packagers |
| AI Diffusion Rule | Biden-era rule from January 2025 formally rescinded | All exporters; the rule was never enforced |
How the data center license works
The license requirement applies when there is "knowledge" that advanced chips will be used in a data center, or a group of linked data centers, that provides more than 12,000,000 aggregate TPP to companies headquartered in Macau or in Country Group D:5. That group covers the countries under US arms embargoes, including China and Russia. The test looks at the ultimate parent company, so a subsidiary registered elsewhere does not escape it.
TPP, or total processing performance, is the export control measure of a chip's raw arithmetic. By our calculation, using the commonly cited figure of about 15,832 TPP for one Nvidia H100, the threshold equals roughly 760 H100s, or about 330 Blackwell B200 chips. That is our estimate, not an official BIS conversion, but it shows the rule targets serious training clusters rather than a single server.
The rule also covers indirect routes. A data center that knowingly serves a customer outside D:5 which trains or supplies AI models for a D:5 company falls under the same license requirement. Data centers, foundries and chip exporters must check their customers, including any direct or indirect ownership or control.
Operators already selling that level of compute to covered companies before the rule is published must apply for a license within 90 days of the effective date to keep doing so, or "face future export restrictions", BIS says. Applications will be reviewed case by case. Commercial recommender systems, such as those used by online retailers and social networks, are exempt.
Why Malaysia, Singapore and three others were named
BIS describes Indonesia, Malaysia, Singapore, Thailand and Brazil as countries with a higher-than-normal risk of chips being diverted. Exports to them will now need a license unless the buyer is a company headquartered in the US, or owned by a US parent, that meets new reporting requirements. The Commerce Secretary can add more countries later.
The rule also warns that countries tolerating evasion could face import restrictions, new chip license requirements, a lower status in the US export control system, and Entity List designations for people involved. Bloomberg, which reported on the rule on Monday, framed it as a limit on remote access from countries such as Singapore and Malaysia.
The move follows months of pressure over chips reaching Chinese projects through third countries. We reported earlier how restricted Nvidia chips still anchor $3.2 billion in China AI projects, and how Commerce Secretary Howard Lutnick had pushed for softer chip rules.
What happens next
Shipments already on their way when the rule is published can continue under the old terms if they are completed within 30 days. Otherwise the rule applies from publication, with the 90-day window for data centers that already serve covered companies. Companies can send comments through regulations.gov during the 60-day window.
The bigger change is in what Washington now controls. Earlier US AI chip export controls focused on where chips were shipped. This rule adds a second question: who is using the computing power after the chips are installed. For cloud providers in Southeast Asia, that means checking the ownership of every large AI customer, not just the address on the shipping label. The rule also sits alongside the US-led Pax Silica effort to build a shared AI protection list with allies.