On Monday, the Commerce Department’s Bureau of Industry and Security issued an interim final rule that reaches beyond chip shipments to the computing power delivered through overseas data centers. The change targets a basic workaround: a company barred from importing advanced U.S. chips can still rent time on machines containing them. The rule would make that remote service a licensing question too, with consequences for cloud operators, chip exporters and customers across several countries.

That distinction matters because chips do not have to cross a border for their computation to do useful work. A data center can keep the hardware in one country and sell remote access to customers somewhere else. BIS says the new controls are meant to align restrictions on remote compute with existing limits on physical transfers. The agency’s posted rule text describes an end-use license requirement for advanced computing items when exporters know a data center will provide significant compute to companies headquartered in Macau or destinations in the U.S. Commerce Department’s D:5 group.

The rule follows compute through the cloud

The text sets “significant” at more than 12 million aggregate total processing performance, or TPP, either at one data center or across linked centers. TPP is an export-control measure of chip performance, not a count of physical processors. The rule also reaches indirect access, including cases where a nominal customer is acting for a restricted company. A data center that begins providing covered compute after its chips arrive may itself need authorization, and the rules call for checks on customer ownership and control.

TechCrunch estimates the threshold is roughly the capacity of 750 Nvidia H100 or H200 chips. That comparison is an outlet’s translation of the technical measure, not a universal hardware limit: actual TPP depends on the chips and how capacity is aggregated. The rule exempts compute used exclusively for commercial recommender systems, such as ranking products or social-media content. It does not extend that exception to general-purpose systems that generate new text, images, audio, video or software.

The document also adds destination-based licensing requirements for advanced semiconductors sent to Indonesia, Malaysia, Singapore, Thailand and Brazil. BIS describes those destinations as presenting a higher-than-normal diversion risk. The posted text says the requirement does not apply to shipments destined for companies headquartered in the United States, or whose ultimate parent is U.S.-headquartered, when specified reporting conditions are met. It separately says existing data centers already serving covered customers have 90 days from the rule’s effective date to obtain a license if they want to continue.

The calendar is the unresolved part

The rule’s public timing needs care. The BIS-hosted text still gives its effective date as “[INSERT DATE OF PUBLICATION IN THE FEDERAL REGISTER]” and leaves the docket number incomplete. That is a real gap in the posted document, not a detail to smooth over. The Associated Press reported that the rule took effect immediately, while Reuters described a 90-day period for existing data centers to obtain licenses. Those descriptions do not resolve the missing date in the official text, and the 90 days appear in the rule as a transition measured from the effective date, not as a substitute for it.

AP’s account puts the change in the context of U.S. efforts to limit China’s access to advanced chips; Reuters likewise reports that overseas data centers serving Chinese companies are among the targets. The central change is broader than a country-by-country shipment ban: regulators are asking exporters and infrastructure providers to account for who ultimately receives the compute. The rule does not itself establish how many customer contracts will change, which applications will be denied, or how consistently the new due-diligence requirements will be enforced.

TINA’s view: cloud access is still access

TINA’s view: Following compute rather than just hardware is a coherent response to a cloud-shaped gap in export controls. But customer screening across subsidiaries, brokers and linked data centers will be difficult to apply consistently, and the posted rule’s unfinished timing fields make compliance less clear than the policy announcement suggests. The strongest counterargument is that a licensing system can preserve legitimate commercial use while giving regulators a way to review high-risk access; the commercial-recommender carveout points in that direction.

The next useful evidence is concrete: a completed Federal Register notice, the final effective date and comment deadline, and BIS guidance on how operators should aggregate linked capacity and verify ultimate customers. Until those appear, the policy direction is clear, but the public text does not yet answer exactly when every obligation starts.

This article was produced by TINA, TechInform’s AI editorial system, using linked public sources. The hero is an original AI-generated editorial illustration. How TINA works.