Export controls on chips were always easy to sidestep if you could simply rent the chips instead. Washington now wants to close that gap. The Trump administration is preparing an interim rule that requires licenses to export, reexport or transfer AI chips and servers to any data center in the world that may supply compute to companies in China or other restricted countries.
The targets are firms in the D:5 group of embargoed countries, which includes China, Russia, Iran, North Korea and 18 others. Existing rules already required licenses for shipping chips to companies based in, or owned by businesses in, those countries. This rule extends that logic to access. The text calls it a “logical extension of existing controls.” That’s fair. The old rules governed hardware, and the workaround lived in the cloud.
The thresholds are specific. The rule covers any data center or network that can supply more than 12 million total processing performance (TPP), roughly the peak output of 750 Nvidia H100 or H200 chips. It also reaches indirect access through brokers and go-betweens. Operators and exporters must watch their customers’ ownership structures and activity. Existing operators have 90 days to get licenses. Commercial recommendation systems, like those in e-commerce or social media, are exempt.
Geography matters too. New license requirements apply to advanced chip shipments to Malaysia, Singapore, Indonesia, Thailand and Brazil, which Commerce says carry a higher-than-normal diversion risk. Chips headed to U.S.-based companies in those countries are excluded. If you run GPU capacity in Southeast Asia, this is your problem now.
The timing is not subtle. The rule lands days after Nvidia co-founder Jensen Huang resigned amid a scandal involving an alleged Chinese spy tied to a customer. The government has reportedly wanted to put that customer, Megaspeed, on a special entity list over suspicions it bought thousands of Nvidia chips to resell compute to Chinese businesses. Reuters also reported that Chinese AI firms were racing to lock in overseas capacity before restrictions tightened. So the rush was real, and this is the response.
The rule also rescinds the Biden-era AI Diffusion Rule from January 2025, which set country-level caps on advanced chip exports and limited transfers of certain powerful models. The Trump team said in May it would scrap that framework as a burden on U.S. companies and allies. The replacement is narrower and more targeted: instead of tiering whole countries, it follows who actually gets the compute.
Does it matter? Yes. For Nvidia, AMD and the neoclouds, compliance costs go up, and customer vetting becomes a product requirement. For hyperscalers like AWS, Azure and Google Cloud, which already screen customers, the burden is lighter. The biggest losers are smaller offshore GPU resellers. Enforcement is the open question. Verifying beneficial ownership through layers of brokers is hard, and Chinese labs have shown they will find the next gap. Still, this is the most serious attempt yet to regulate compute as a service, not just a box.



