Dr Kamran Jalali

BIS's New AI Chip Export Controls Reach the Cloud

BIS AI chip export controls now reach remote cloud computing across national borders.

Introduction

An advanced AI chip no longer has to cross a border for a foreign company to use it. It can sit in a Malaysian data center while a customer in China rents its computing power. That is the gap the US Commerce Department is trying to close.

On November 16, the Bureau of Industry and Security (BIS) released an interim final rule, Addressing Remote Access to Advanced Semiconductors. It is scheduled to take effect upon Federal Register publication on November 19, extending the new AI chip export controls beyond physical shipments to certain remote-computing arrangements.

My view: this is a more direct response to the cloud-computing loophole than another round of chip shipment limits. But it is not a worldwide ban on cloud AI, and it does not mean ordinary people will suddenly lose access to chatbots. The exact trigger is narrower, and more technical, than many headlines suggest.

The New AI Chip Export Controls Target Computing Power

The 12 million TPP threshold is the key number

The rule creates a licensing requirement for certain advanced computing chips and related equipment sent to data centers anywhere in the world. The trigger is what the data center will let a customer do with that hardware.

BIS sets the threshold at more than 12 million aggregate total processing performance, or TPP, for a company headquartered in Macau or a destination in Country Group D:5. TPP is an export-control measure of chip performance. In plain English, it adds up the processing capacity of covered chips. It is not a measure of cloud spending, rented hours, or the number of prompts someone sends to an AI chatbot.

The threshold can be reached at one data center or across linked facilities. The rule includes digitally connected infrastructure, so splitting a large computing operation across sites does not automatically keep it below the limit.

Here is the practical example. A cloud provider installs advanced chips in a data center in a country that is not restricted. A company headquartered in China rents a large block of that computing power. If the covered threshold and other conditions are met, the exporter or transferor may need a license even though the chips never entered China.

This is not a rule against every overseas data center. It applies to specified advanced items and end uses when the relevant party knows the capacity will serve a covered company.

The rule follows indirect customers too

The phrase "to or for" a restricted company is crucial. The rule is not limited to direct cloud contracts.

A third-country customer could rent capacity to train a model for a restricted company. The rule covers certain indirect arrangements where a non-restricted customer trains, helps train, or provides a model to a restricted company.

That makes customer checks central. BIS expects data centers and other companies in the supply chain to examine customer ownership and control, including indirect ownership and parent-company links. The name on an invoice may not tell the full story.

The rule can also matter after the chips arrive. If a data center later begins serving a restricted company, its operator or another party making the transfer may need authorization. Compliance can follow the hardware through later changes in use.

Five Countries Face New Chip Licensing Requirements

Southeast Asia is now part of the control map

The rule adds Indonesia, Malaysia, Singapore, Thailand, and Brazil to a separate set of advanced-semiconductor licensing requirements. BIS says these destinations present a higher-than-normal risk that controlled chips could be diverted to unauthorized users.

This does not mean every shipment of every processor to these countries is automatically blocked. The requirement applies to specified advanced computing items covered by US export rules. There is an exclusion for shipments to companies headquartered in the United States, or whose ultimate parent is headquartered there, when reporting conditions are met.

The choice of countries is revealing. Malaysia and Singapore are important data-center and semiconductor hubs. The Associated Press reported that US and regional authorities have been examining suspected diversion involving Singapore-based Megaspeed International, including servers routed through Singapore and Malaysia. AP also reported that Nvidia had put Megaspeed on its no-sell list after earlier inquiries.

Those reports do not prove every Southeast Asian data center is a diversion route. They show why regulators look beyond the shipping address: a chip can be installed in one country and used by a customer in another.

The new rule gives BIS another way to examine that chain. It can require a license based on the destination of the hardware, the identity of the end user, and the way the computing power will be used.

Existing data centers have a 90-day clock

The rule also reaches existing services. Data centers that were already providing significant computing power to covered companies before publication must obtain a license within 90 days of the effective date if they want to continue.

Operators with long-term contracts may still need authorization to keep serving a restricted company.

BIS will review license applications case by case, and approval is not guaranteed. A contract signed before the rule will not automatically protect an operator.

The rule's use of "knowledge" adds another layer. Companies may need to act on information that gives them reason to know who controls a customer or where capacity is going. Weak screening could become a legal risk.

The Strange Part: BIS Is Also Repealing an AI Rule

The old AI Diffusion Rule is being formally rescinded

There is a twist. In the same package, BIS formally rescinds the Biden-era AI Diffusion Rule, a framework announced in January 2025 that the Trump administration had already said it would not enforce.

That earlier framework used country tiers and other controls to manage the spread of advanced AI chips and certain model weights. Model weights are the learned numerical settings inside an AI model that help determine its outputs.

The new approach removes those provisions and replaces them with more focused destination and end-use controls, including a specific rule for remote access and new licensing requirements for five countries.

So "the US repeals AI chip controls" would be a misleading headline. BIS is changing the design of the controls, not abandoning export oversight. The focus is shifting toward where covered chips go, who ultimately uses them, and whether computing power is supplied through a foreign data center.

The recommender-system exception is narrow

The rule includes an exception for computing power used exclusively for commercial recommender systems. These tools rank, filter, or suggest products, services, or content based on user behavior, transaction history, or item details. An online store suggesting shoes based on a shopper's browsing is a simple example.

The exception does not cover systems designed to generate new text, images, audio, video, software, or other content in response to user input. A generative AI model does not become exempt just because it also recommends products or content.

That distinction matters. A data center cannot simply label a general-purpose AI service as a recommendation engine and assume the exception applies. The use must fit BIS's definition and meet its reporting requirements.

What This Means for AI Companies, Cloud Providers, and Users

The first pressure point is compliance, not an immediate shortage of AI tools. Data centers, chip exporters, server makers, and cloud providers will need better information about customers and indirect end users. That can slow deals and raise legal costs.

The second pressure point is capacity planning. A large AI customer may not be able to shift workloads between connected data centers as freely if the move changes its licensing position.

There may also be effects on where new data centers are built. If a country becomes harder to use as a route for supplying restricted customers, cloud companies may reconsider future hardware locations. That is a possible business response, not a confirmed outcome.

For everyday users, the direct effect is much less clear. The rule does not say that a person in Pakistan, Europe, or the United States cannot use a public AI chatbot because its servers contain American chips. It regulates certain exports, transfers, and data-center end uses involving specified advanced computing items and restricted corporate customers.

Providers could change prices, capacity, or enterprise customers. But the rule does not automatically raise subscription fees or shut down popular AI services.

The Strongest Counterpoint: Rules Do Not Guarantee Enforcement

The strongest counterargument is simple: the rule cannot monitor every cloud contract. It depends on companies identifying customers accurately, despite complex ownership structures, brokers, and cross-border service agreements.

The Center for Strategic and International Studies warns that unclear due-diligence expectations could lead to over-compliance. Providers may reject legitimate customers rather than risk serving a prohibited one.

The rule can make remote access harder, but it cannot erase global demand for advanced computing. Companies may seek different hardware, suppliers, or technical methods.

Still, the rule closes a gap. Renting computing time abroad can deliver much of the same capability as owning the chips. BIS is trying to regulate the service the hardware provides, not only the box that carries it.

The test is whether BIS can enforce the rule without making legitimate cloud services unnecessarily difficult.

FAQ

What are the new US AI chip export controls?

BIS has introduced licensing requirements for certain advanced chips and related equipment used by data centers that provide large amounts of computing power to companies headquartered in China and other restricted destinations. It also adds licensing requirements for covered advanced chips sent to five additional countries.

What does 12 million TPP mean?

TPP means total processing performance, a technical measure used to compare the computing performance of covered chips. The rule adds the TPP of relevant chips together and uses more than 12 million as the threshold for certain remote data-center access arrangements.

Does the rule ban China from using all foreign cloud services?

No. It targets specified advanced computing items and data-center arrangements that cross the rule's threshold and meet its end-user conditions. It is not a blanket ban on every cloud service or AI application.

Which countries face the new chip licensing requirements?

The five additional destinations are Indonesia, Malaysia, Singapore, Thailand, and Brazil. The rule includes a reporting-based exclusion for certain shipments to US-headquartered companies.

When do existing data centers need a license?

Data centers already providing covered computing power to restricted companies must obtain a license within 90 days of the rule's effective date if they want to continue those services. The rule is scheduled to take effect upon Federal Register publication on November 19, 2026.

Does the rule affect ChatGPT or other AI chatbots?

It does not directly prohibit ordinary people from using AI chatbots. It regulates certain exports and transfers of advanced computing items and their use in data centers serving restricted corporate customers.

Conclusion

The most important change in BIS's new AI chip export controls is not another border check for a box of GPUs. It is the attempt to control who can rent the computing power after those GPUs have been installed thousands of miles away.

The 12 million TPP threshold, linked-data-center rules, and 90-day deadline for existing services turn remote compute into an export-control issue. That could make it harder for restricted companies to reach advanced US technology through overseas cloud providers. It could also create extra friction for legitimate data-center businesses.

BIS has drawn a new line around access to AI computing power. The question now is whether companies can enforce that line without blocking too much legitimate cloud activity.

If you operate or rent cloud computing capacity, would you accept stricter customer checks to close this loophole, even if they slow down legitimate services?

Disclaimer: This article is for information and education only. It is not legal, financial, or investment advice. Affected businesses should consult qualified export-control counsel.

Did you enjoy this article?


Crypto Strategist
Crypto Strategist

I am Dr. Kamran Jalali, Crypto researcher & educator. Deep analysis on crypto trends, AI tokens, RWA, and smart money, in plain language. No hype. Just honest research to help you make smarter decisions.


Dr Kamran Jalali
Dr Kamran Jalali

Most people lose money in crypto not because the market is against them — but because nobody ever taught them the rules of the game. I am Dr. Kamran Jalali. I write about crypto in plain, simple language that anyone can understand — no confusing jargon, no hype, no false promises. Here you will find honest breakdowns of how crypto really works, why traders fail, how to protect your money, and how to make smarter decisions in the digital asset world. Whether you are completely new to crypto or have been in

Publish0x Publish0x

Reward the author with $0.01 in crypto, and earn yourself as you read!

20% to author / 80% to me.
Rewards are FREE. Publish0x pays them, not you.

Page not displaying correctly?