Regulation•

Washington Shuts the Cloud Route to AI Chips: Data Centers Serving Chinese Firms Now Need a License Above 12 Million TPP

A new Commerce Department rule extends US AI chip export rules to remote cloud access, requiring licenses for data centers that supply large amounts of computing power to China-based customers.

Written By :MINRK
Washington Shuts the Cloud Route to AI Chips: Data Centers Serving Chinese Firms Now Need a License Above 12 Million TPP

Photo by Igor Omilaev on Unsplash

1. A license requirement for renting AI compute to China

The US Commerce Department's Bureau of Industry and Security (BIS) issued an interim final rule on Monday, November 16, 2026, that extends AI chip export rules to remote access. Data centers anywhere that provide more than 12 million aggregate total processing performance (TPP) to customers based in China, Macau or certain other restricted jurisdictions now need a US license.

The measure targets a gap that had existed for years. Chinese companies barred from purchasing the most powerful American AI accelerators could still use them by renting computing capacity from facilities located outside China. The chips never crossed a Chinese border, so the transactions fell outside controls written around physical shipments. The new rule is designed to bring that rented access under the same regime that already governs the hardware itself.

Alongside the remote access provisions, the rule adds five countries to the list of destinations where exports of advanced AI chips require a license: Indonesia, Malaysia, Singapore, Thailand and Brazil. BIS cited an elevated risk that chips sent to those markets could be diverted to restricted users.

2. Why the old controls left a side door open

US export controls on advanced semiconductors were built to answer one central question: where is the hardware going? Licensing decisions focused on the physical destination of a shipment and on the identity of the buyer receiving it. That framework worked reasonably well for preventing a crate of accelerators from landing in a restricted country.

It was far less effective against a business model in which the chips stay put and the customer comes to them over a network connection. A cloud provider in a third country could legally acquire top-tier processors, install them in its own facility and then sell time on those machines to clients in China. Under the previous rules, the controls tracked where the chips were delivered, not who was ultimately using them from a distance.

For an AI developer, the distinction matters little. Training and running large models requires access to computing power, not ownership of the physical servers. Renting capacity in a foreign data center can deliver much of the same practical capability as buying the hardware outright, which is why policymakers came to see remote access as a meaningful weakness in the existing system.

BIS has framed the new rule as an effort to align the treatment of remote access with the restrictions that already apply to physical exports. In effect, the agency is now asking not only where a chip sits but also who is drawing on its processing power.

3. What do the new AI chip export rules require?

The new AI chip export rules require a license for any data center that supplies more than 12 million aggregate TPP to entities based in China, Macau or other covered restricted jurisdictions. Below that threshold, no license is needed for the remote access itself; above it, the provider must seek approval from BIS.

Total processing performance is the metric US regulators use to measure the raw computing capability of advanced chips. By setting the threshold as an aggregate figure, the rule looks at the combined capacity a provider makes available to covered customers rather than at any single processor. A provider serving such clients therefore has to track the total compute it is supplying to them, not just the number of machines involved.

Providers that already serve customers in the affected jurisdictions are not cut off immediately. They have a 90-day window to come into compliance, which in practice means obtaining a license or winding down the relationships that exceed the threshold.

The framework also leans on customer vetting and reporting. Data centers and the suppliers that equip them are expected to apply know-your-customer diligence to establish who their clients are and who controls them, and the rule builds in monthly reporting to BIS. Those obligations are meant to give the agency visibility into who is using American compute abroad, rather than relying solely on where the chips were shipped.

4. The Megaspeed case that forced the issue

The rule is widely seen as a direct response to the controversy surrounding Megaspeed, a Singapore-based cloud provider. US investigators have examined whether the company was involved in diverting or circumventing controls on Nvidia hardware for the benefit of Chinese customers.

The case highlighted the precise weakness the new rule addresses. Advanced chips could be lawfully located in a third country and still serve clients in China through a network connection. Even if every shipment complied with the letter of the export rules, the end users drawing on that compute could be the very parties the controls were meant to exclude.

The Megaspeed affair turned what had been a largely theoretical concern among export control specialists into a concrete political problem. It showed regulators that a sizable volume of controlled hardware could serve restricted users without ever triggering the shipment-based checks that the system relied on.

5. Five countries move onto the license list

The decision to require licenses for advanced AI chip exports to Indonesia, Malaysia, Singapore, Thailand and Brazil reflects how the data center map has shifted. Singapore, Malaysia, Thailand and Indonesia have become popular hubs for new computing capacity, attracting investment thanks to their location, connectivity and access to power and land.

For years, that growth was treated largely as a commercial success story for the region and for the American chipmakers supplying it. The rule signals a different view in Washington: rapid data center expansion close to China is now also seen as a potential channel through which controlled technology could leak to restricted users.

Brazil's inclusion stands out. It is not one of China's neighbors, and adding it to the list suggests BIS is looking beyond Southeast Asia and taking a wider view of where diversion risk may lie. Any country with a fast-growing data center sector and the potential to serve offshore clients could, in principle, attract similar scrutiny.

6. Who feels the impact first

The most immediate burden falls on data center operators in the five named countries. They now face license applications, more rigorous customer screening and recurring reports to a US agency. Each of those steps adds administrative cost, and the extra checks may slow down sales to some clients while compliance teams work through the new requirements.

Operators that have built their business around demand from Chinese customers face the hardest adjustment. For them, the economics could change quickly. A 90-day window is a short period in which to replace a major customer base, and providers that cannot secure a license may have to give up revenue they had counted on.

Chipmakers, Nvidia among them, could also see market volatility as the controls take hold. Cloud providers in Southeast Asia have been significant buyers of advanced accelerators, and any slowdown in their purchases, or any shift in how they allocate capacity, could affect demand for the most powerful processors.

7. Can the 12 million TPP threshold be gamed?

Any fixed threshold invites attempts to work around it, and the 12 million TPP line is no exception. Customers seeking to avoid a license requirement could try to split their demand into smaller rentals spread across several providers, keeping each relationship below the limit.

Another possibility is routing access through intermediaries located in countries that are not on the new license list. A company in an unlisted jurisdiction could, in theory, rent capacity and pass it on to a restricted end user, adding a layer of separation between the data center and the party actually using the compute.

These risks explain why the rule pairs the threshold with customer vetting and reporting. A cap on its own can be engineered around; a cap backed by diligence on who customers really are, and by regular reports that regulators can cross-check, is harder to evade. How effective that combination proves will depend on how rigorously it is applied.

8. Enforcement will decide whether the rule works

Know-your-customer requirements and monthly reports are only as strong as the review behind them. Paperwork that is filed but never examined does little to stop determined buyers. For the new regime to change behavior, someone at BIS has to read the reports, spot anomalies and act on them.

The Megaspeed episode is a reminder of how much can pass unnoticed when oversight lags behind the market. Controlled hardware was able to serve restricted customers precisely because the system was not looking at who was using it. The new rule closes that blind spot on paper, but its real test will be whether the agency has the capacity and the will to follow through.

9. What happens next

Three developments will shape how the rule plays out. The first is formal publication in the Federal Register, which will set out the final text of the interim rule and give companies a definitive version to work from.

The second is how data center operators respond during the 90-day compliance window. Some may apply for licenses, others may restructure or drop customer relationships, and their choices will show how much business was flowing to China through rented compute.

The third is enforcement. Whether BIS uses the new reporting and vetting tools to pursue violations will determine whether the remote access route is genuinely closed or merely narrowed. For chipmakers, cloud providers and the investors who follow them, the coming months will reveal how far Washington is prepared to go in policing not just where AI chips are sold, but who gets to use them.

Related Articles

NEWSLETTERS

Don't miss another story.

Subscribe to the MINRK Newsletter today.

Tick the newsletters you want. No selection? You'll get MINRK Headlines.

By signing up, you will receive emails about MINRK products and you agree to our terms of use and privacy policy.

MINRK Daybook

Market analysis for crypto traders and investors.

EVERY WEEKDAY

MINRK for Advisors

Defining crypto, digital assets and the future of finance for financial advisors.

EVERY THURSDAY

MINRK Protocol

Exploring the tech behind crypto one block at a time.

WEEKLY

MINRK Long & Short

A must read for institutions. Insights, news and analysis delivered weekly.

EVERY WEDNESDAY

MINRK Headlines

The biggest crypto news and ideas of the day.

EVERY WEEKDAY

MINRK Policy

Examining the intersection of cryptocurrency and government.

WEEKLY

MINRK Research

Join thousands of readers who rely on MINRK for data-driven insights on the latest digital asset trends.

MONTHLY