The Trump administration is preparing new export controls aimed at a way companies could get around restrictions on AI chips: accessing computing infrastructure remotely from another country. The proposed interim rule would require licenses for certain chips and servers sent to data centers around the world when they could provide compute to companies in China or other restricted countries.
Licenses would follow computing capacity overseas
Existing U.S. rules require chip companies to obtain a license before shipping AI chips to companies based in, or owned by businesses in, the D:5 group of embargoed countries. That group includes China, Russia, Iran, North Korea, and 18 others.
The new rule extends licensing to exports, reexports, or transfers of AI chips and servers to data centers anywhere in the world if those facilities may supply compute to companies in China or other restricted countries. In the rule’s words, it is “a logical extension of existing controls.”
The controls apply to data centers or networks capable of supplying more than 12 million total processing performance, or TPP, an export-control measure of computing power. The source estimates that level is roughly equivalent to the maximum computing power of about 750 Nvidia H100 or H200 chips.
Remote access is central to the policy. The rule also addresses access through intermediaries or brokers, and requires exporters and data center operators to monitor customer ownership and activities so they can determine whether a facility is being run by someone from the D:5 group.
Five more destinations face chip licensing requirements
The rule adds licensing requirements for exports, reexports, or transfers of advanced semiconductors to Malaysia, Singapore, Indonesia, Thailand, and Brazil. The Commerce Department said the countries “present a higher-than-normal risk of diversion.”
There is a stated exception for chips destined for companies based in the U.S.: the new license requirements for these five countries will not apply to those shipments. The rule therefore combines a broader check on where advanced chips go with a specific carve-out for U.S.-based companies.
Data center operators already providing services covered by the rule have 90 days from the rule taking effect to obtain licenses. That gives existing providers a defined period to address the new requirements, while the rule’s scope reaches facilities that can supply the specified level of computing power.
Policy change follows concerns about diversion
The announcement comes amid concern that overseas computing capacity could help Chinese AI companies access chips despite existing restrictions. Reuters reported last week that Chinese AI companies were rushing to secure computing capacity overseas amid fears of tighter U.S. controls.
The article also describes U.S. government concerns about Megaspeed, a customer the government has reportedly wanted to place on a special entity list. The government believes Megaspeed has been buying thousands of Nvidia chips to provide compute to Chinese businesses and get around existing export controls.
The rule is presented as a national security measure. It says access to computing power at the covered scale could support training and running advanced SI models, including models capable of supporting military- and intelligence-related activities that could threaten U.S. national security or foreign policy interests.
The rule replaces an earlier AI framework
The interim rule rescinds the Biden administration’s AI Diffusion Rule from January 2025. That framework sought to set country-level limits on exports of advanced AI chips and restrict transfers of certain powerful AI models.
In May, the Trump administration said it would discard that framework, arguing it imposed unnecessary restrictions on U.S. companies and allies. The new rule instead focuses on the use of certain SI items in data centers and at computing power levels that can be used to train or run advanced SI models.
One category is expressly excluded: the rule does not apply to commercial recommendation systems, including systems used in e-commerce or social media. That distinction limits the stated scope while the controls target access to large-scale computing infrastructure that could support advanced model development or operation.