U.S. expands AI chip curbs to block remote access

The planned rule would require licenses for some overseas data centers and chip transfers if compute could reach companies in China, Russia, Iran or North Korea.
The Trump administration is preparing a new round of export controls aimed at AI chips and overseas data centers, trying to close what it sees as a loophole in existing U.S. restrictions. The main focus is remote access: Washington wants to make it harder for companies — especially in China — to use computing power located outside the United States. That matters because the pressure would no longer stop at physical shipments and would extend to cloud and data center services.
The move comes as the U.S. tries to prevent controlled AI hardware from being used indirectly by companies in countries it already treats as restricted. That includes China, as well as Russia, Iran, North Korea and other countries in the D:5 group. The Commerce Department’s interim rule would require licenses for exports, reexports or transfers of AI chips and servers to data centers anywhere in the world if they may provide compute to companies in those countries.
Remote access is now part of the export-control push
The new rule is meant to address a gap that existing controls do not fully cover. The U.S. already requires licenses to ship AI chips to companies based in, or owned by businesses in, the D:5 group. Under the new approach, though, the government is also looking at whether a data center abroad could still provide computing power to those same companies through remote access.
According to the rule, the restrictions are a “logical extension” of current controls. The administration says the aim is to stop companies headquartered in countries of concern from using U.S. advanced Super Intelligence semiconductors to train advanced AI models.
In practice, that means the government is no longer focused only on where chips are physically sent. It is also looking at who can use the compute after the hardware is installed, even when access happens through networks rather than direct ownership of the equipment. As IT-PUB News reports, the proposed controls would reach beyond shipments and into the way overseas computing infrastructure is used.
Data centers above the TPP threshold would need scrutiny
The controls are tied to a specific computing-power threshold. They apply to data centers or any network that can supply more than 12 million total processing performance, or TPP, which is the export-control metric the U.S. uses to measure computing power. The source notes that 12 million TPP is roughly equivalent to the maximum computing power of about 750 Nvidia H100 or H200 chips.
The rule also covers indirect access through middlemen or brokers. Exporters and data center operators would have to examine customer ownership structures and activities to make sure the service is not being run by someone from the D:5 group.
That broadens the compliance burden well beyond a simple shipping rule. Companies would need to look not just at the direct buyer, but also at who ultimately controls the customer and how the compute might be used.
Malaysia, Singapore and Brazil face new licensing checks
The Commerce Department is also adding new licensing requirements for chip exports, reexports or transfers of advanced semiconductors to Malaysia, Singapore, Indonesia, Thailand and Brazil. The department said those countries were included because they “present a higher-than-normal risk of diversion.”
There is an important carve-out. For those countries, the licensing requirement will not apply to chips destined for companies based in the U.S. So this is not a blanket ban, but it does add another layer of review for destinations Washington sees as higher risk.
Existing data center operators that fall under the new rule will have 90 days after it takes effect to obtain the required licenses. That gives companies a short window to adjust contracts, customer screening and service arrangements.
The rule targets overseas demand for AI compute
The timing matters. Reuters reported last week that Chinese AI companies are rushing to secure computing capacity overseas amid fears of tighter U.S. restrictions. The new rule appears designed to make that route harder.
The administration’s concern is not only commercial. The rule says access to this level of computing power can be used to train and run advanced AI models capable of supporting military and intelligence-related activities that could threaten U.S. national security or foreign policy interests.
The source also ties the move to recent controversy around Nvidia. It says the administration is preparing the rule days after Nvidia co-founder Jensen Huang resigned amid a scandal involving an alleged Chinese spy linked to one of the chipmaker’s customers. The U.S. government has reportedly wanted to place that customer, Megaspeed, on a special entity list, believing it had been buying thousands of Nvidia chips to provide compute to Chinese businesses in a way that bypasses existing restrictions.
The administration drops the AI Diffusion Rule
The new rule also rescinds the Biden administration’s AI Diffusion Rule from January 2025. That earlier framework sought to impose country-level limits on exports of advanced AI chips and to restrict transfers of certain powerful AI models.
The Trump administration had already said in May that it would discard that framework, arguing that it placed unnecessary restrictions on U.S. companies and allies. The new interim rule replaces it with a different model centered more on licensing, data center access and diversion risk.
Notably, the rule does not apply to commercial recommendation systems, such as those used in e-commerce or social media. That narrows its reach and suggests the government is concentrating on high-end AI training and compute infrastructure rather than consumer-facing ranking and suggestion tools.
The intervention is narrower than the Biden-era framework, but still significant for the global AI supply chain. For chipmakers, cloud providers and data center operators, the practical challenge will be proving that advanced compute is not being routed to restricted buyers. For companies outside the U.S., access to American AI hardware and infrastructure could become more complicated — especially when ownership structures or end users are difficult to verify.