AI chip remote access rule requires licenses for China-linked compute
A new AI chip remote access rule from the US Commerce Department requires licenses when overseas data centers would sell large amounts of compute to Chinese firms.

Washington has moved to close the cloud loophole in its chip curbs. The AI chip remote access rule, an interim final rule announced by the Commerce Department's Bureau of Industry and Security on Nov. 15, requires licenses to ship or transfer advanced AI chips and servers to any data center worldwide that may sell computing power to companies in China or other restricted countries.
What happened
The control covers data centers or linked networks able to supply more than 12 million total processing performance, or TPP, the export metric for computing power. TechCrunch said that is about the peak output of 750 Nvidia H100 or H200 chips. It targets firms headquartered in Country Group D:5, which includes China, Russia, Iran and North Korea, along with Macau, and also reaches indirect access through brokers. Operators must check customers' direct and indirect ownership.
A separate requirement adds licenses for exports, re-exports and transfers of covered chips within Malaysia, Singapore, Indonesia, Thailand and Brazil, which Commerce said present a higher-than-normal risk of diversion. Recipients headquartered in the U.S. are excluded if exporters file monthly reports, The Defiant said. There is also an exception for compute used only for commercial recommendation systems, but not for systems that generate text, images, video or software.
Why it matters
Why does the AI chip remote access rule matter now? It lands days after Jensen Huang quit as Nvidia's chief executive amid a scandal tied to Megaspeed, a customer the U.S. reportedly wants on an entity list because it is believed to have bought thousands of Nvidia chips to serve Chinese businesses. Reuters reported last week that Chinese AI companies were racing to lock in overseas capacity before tighter limits arrived, according to TechCrunch. According to the rule text, the aim is to stop firms based in countries of concern from tapping U.S. chips to train advanced models.
Remote access was not fully outside the rules before. BIS guidance from May 2025 warned that supplying overseas cloud providers could need authorization if model training for Chinese parties served military-intelligence or weapons ends. The new test is based on computing capacity and the customer, without that end-use link. For background, see our coverage of the Bessent Nvidia deadline and the Megaspeed probe.
What's next
The rule takes effect on publication in the Federal Register, and as of Monday morning there was no matching entry, The Defiant said. Shipments already en route get 30 days under previous licensing terms, and data centers already serving covered customers get a longer transition. License applications will be reviewed case by case. Cloud operators in Southeast Asia now face the AI chip remote access rule as a compliance cost, and Nvidia's guidance on Tuesday may show how much it bites. TechCrunch's report is here.
This article is for information only and is not investment advice.