The Commerce Department is closing the cloud loophole in US chip export policy. According to TechCrunch, an interim rule due Monday requires a license before AI chips and servers reach any data center in the world that might supply computing power to companies in China or other restricted countries.

Until now, the controls followed the hardware. Chipmakers needed a license to ship to buyers based in, or owned by a business in, the D

list of embargoed nations. China, Russia, Iran and North Korea sit on it, along with 18 more. Renting capacity in a third country avoided that check. The new rule follows the compute instead of the box it ships in.

The threshold is lower than many readers will assume. The rule covers any data center or network able to supply more than 12 million units of total processing performance (TPP), Washington’s yardstick for computing power. TechCrunch calculates that as roughly the peak capacity of about 750 Nvidia H100 or H200 chips. That is a modest cluster by frontier-lab standards, so the rule reaches well beyond the hyperscalers.

It also reaches the middlemen. Access routed via brokers or other intermediaries counts, and operators and exporters alike must track who owns each customer and what that customer does, so no D

company hides behind an account.

Geography shifts too. Licenses will now be needed for advanced chip exports, reexports or transfers to Malaysia, Singapore, Indonesia, Thailand and Brazil, which Commerce says carry a higher-than-normal risk of diversion. Chips headed to US-based companies in those countries are exempt. Operators already providing services that fall under the rule have 90 days after it takes effect to obtain licenses. Commercial recommendation systems, such as those behind e-commerce or social media, are outside its scope.

The timing is not subtle. TechCrunch says Nvidia co-founder Jensen Huang stepped down only days earlier, after a scandal over an alleged Chinese spy tied to a chip customer. That customer, Megaspeed, is one Washington has reportedly wanted to put on a special entity list, because it believes Megaspeed bought thousands of Nvidia chips to resell compute to Chinese businesses. A Reuters report last week described Chinese AI companies rushing to lock in overseas capacity ahead of tighter limits.

Last week we covered an independent analysis that found Blackwell-chip patterns in six Chinese models. That analysis did not establish any export breach, and this rule does not change that. It does show where policymakers think the leak is: rented compute, not smuggled hardware.

The rule also rescinds the Biden administration’s AI Diffusion Rule from January 2025, which would have set country-level limits on advanced chip exports and restricted transfers of certain powerful models. The Trump administration said in May it would discard that framework as an unnecessary burden on US companies and allies. The replacement trades national caps for scrutiny of who ends up using the capacity. In the rule’s own words, the new control is “a logical extension of existing controls.”

For data center operators in the five newly listed countries, and for any cloud provider with Chinese-owned customers, the 90-day clock turns customer ownership checks from good practice into a licensing condition.

Based on reporting by Ram Iyer for TechCrunch, published 16 November 2026.