JupiteX Get the app
International Relations17 Nov 2026 · about 7 min

US tightens AI chip export controls to deter Chinese companies from accessing compute remotely

The brief

The rule reaches beyond shipping chips to a foreign buyer: it would require licenses for exporting, reexporting, or transferring AI chips and servers to data centers anywhere if they may provide computing to companies in China or other restricted countries. That matters because a company could use U.S. technology abroad without importing the hardware itself. For example, a data center in another country could house advanced chips and sell remote computing capacity to a China-based business. Under the rule, the licensing requirement also reaches indirect access through brokers or other go-betweens. Exporters and data center operators must monitor customers’ ownership and activities to ensure no D:5-group entity runs them. The controls apply to data centers or networks exceeding 12 million total processing performance, roughly the power of 750 Nvidia H100 or H200 chips. Separate new license requirements cover advanced chip transfers to Malaysia, Singapore, Indonesia, Thailand, and Brazil, though destinations at U.S.-based companies are exempt. Existing affected operators have 90 days after the rule takes effect to obtain licenses.

01

What new activity will the US require a license for under this rule?

The rule reaches beyond shipping chips to a foreign buyer: it would require licenses for exporting, reexporting, or transferring AI chips and servers to data centers anywhere if they may provide computing to companies in China or other restricted countries. That matters because a company could use U.S. technology abroad without importing the hardware itself.

For example, a data center in another country could house advanced chips and sell remote computing capacity to a China-based business. Under the rule, the licensing requirement also reaches indirect access through brokers or other go-betweens. Exporters and data center operators must monitor customers’ ownership and activities to ensure no D:5-group entity runs them.

The controls apply to data centers or networks exceeding 12 million total processing performance, roughly the power of 750 Nvidia H100 or H200 chips. Separate new license requirements cover advanced chip transfers to Malaysia, Singapore, Indonesia, Thailand, and Brazil, though destinations at U.S.-based companies are exempt. Existing affected operators have 90 days after the rule takes effect to obtain licenses.

02

How can a company in China use AI chips in an overseas data center without receiving the chips itself?

An overseas data center can hold the chips while a China-based company connects to its computing services remotely. The customer then uses the facility’s processing capacity to run AI work, without importing or physically handling the chips. The rule is designed to address that kind of access, not only hardware deliveries.

For instance, a business operating a data center outside the United States could make its advanced chips available to a Chinese customer over a network. Access could also be arranged through brokers or other go-betweens. The rule covers indirect access as well as direct customer relationships.

The licensing test turns on whether the equipment may provide compute to a company in China or another restricted country. Operators and exporters must monitor customers’ ownership structures and activities so entities from the D:5 group do not run covered services. Facilities above 12 million TPP are covered; the article equates that threshold to about 750 Nvidia H100 or H200 chips.

03

How much computing power triggers the rule, and roughly how many Nvidia H100 or H200 chips is that?

The rule covers a data center or network that can supply over 12 million total processing performance, abbreviated TPP. TPP is an export-control measure the United States uses to assess computing power. The threshold helps define which facilities fall under the new licensing requirements.

The article gives a practical comparison: 12 million TPP roughly equals the maximum computing power of about 750 Nvidia H100 or H200 chips. That means the rule is not limited to a facility holding a single chip or a small number of chips; it focuses on the computing capacity that a data center or network can supply.

The threshold applies to covered data centers and networks, including those that could provide compute to companies in China or other restricted countries. The rule also reaches indirect access through brokers or go-betweens. Operators and exporters must monitor customers’ ownership and activities, while existing operators whose services fall under the rule must obtain licenses within 90 days of its taking effect.

04

How does this rule differ from earlier controls that focused on shipping chips directly to restricted companies?

Existing U.S. controls already require chip companies to get licenses to ship AI chips to companies based in, or owned by a business in, the D:5 group of embargoed countries. Those restrictions focus on the destination or ownership of the company receiving the chips.

The new rule adds a different route: a restricted-country company could access computing power remotely from a data center in another country. It requires licenses for exporting, reexporting, or transferring chips and servers to data centers anywhere if they may provide compute to companies in China or other restricted countries. Brokers and other go-betweens are included too.

The rule also requires exporters and data center operators to watch customers’ ownership structures and activities. It applies to facilities or networks exceeding 12 million TPP, roughly the capacity of about 750 Nvidia H100 or H200 chips. Separately, it adds license requirements for advanced chip transfers to five named countries, with an exception for chips destined for U.S.-based companies.

05

What must existing data center operators do if their services fall under the rule, and by when?

Existing data center operators are not automatically exempt just because their facilities are already running. If the services they provide fall under the new rule, they are required to obtain licenses. The article sets a specific transition period for those operators.

They must get the required licenses within 90 days of the rule coming into effect. The rule applies to data centers or networks that can supply over 12 million total processing performance, and that may provide compute to companies in China or other restricted countries. It also covers indirect access through brokers or other go-betweens.

Operators and exporters are expected to keep an eye on customers’ ownership structures and activities to ensure that entities from the D:5 group do not run covered services. The article does not give a calendar date for the 90-day deadline; it ties the period to when the rule takes effect. The licensing requirement is therefore conditional on a service falling within the rule’s scope.

06

Why does the US consider access to powerful AI computing a national security concern?

The rule frames access to advanced computing as a national security concern because that capacity can be used to train and run advanced Super Intelligence models. It says those models could support military- and intelligence-related activities that may threaten U.S. national security or foreign policy interests.

That concern underpins controls on AI chips, servers, and remote access to data center computing. For example, the rule seeks to prevent companies headquartered in countries of concern from using U.S. advanced SI semiconductors to train advanced SI models. It also reaches access arranged through brokers or other go-betweens.

The rule applies to data centers or networks able to supply over 12 million TPP and may serve companies in China or other restricted countries. It also adds licensing requirements for advanced chip transfers to five other named countries that the Commerce Department considers at higher-than-normal risk of diversion. The article presents these restrictions as a way to limit access to computing power usable for advanced models and related military or intelligence activities.

07

What is AI compute, and why do advanced AI models need so much of it?

AI compute is the processing capacity a computer system uses to handle the calculations involved in artificial intelligence work. The article does not give a technical definition, so this is a standard, general explanation. It describes compute as a service that a data center can provide to a company remotely.

In general, training an advanced model involves processing many examples and adjusting the model based on them; running a trained model also requires computing capacity to produce results. The article refers to both training and running advanced SI models, and its rule aims to limit access to certain high-powered computing resources.

To set a threshold, the U.S. uses total processing performance, or TPP, as an export-control measure of computing power. The rule covers data centers or networks able to supply over 12 million TPP, roughly the maximum capacity of about 750 Nvidia H100 or H200 chips. This makes compute capacity—not just the physical destination of chips—central to the new controls.

This brief was written by AI from the original reporting and checked by other models. Names, figures and quotes come from the source; read it for full context.

Read more in the JupiteX app

Pulse is free. New stories every 4 hours, each one broken into the questions that explain it.

Or read more news on the web