By Hamza Faisal

The United States is preparing to close a major gap in its AI chip export controls: the ability of Chinese companies to rent Nvidia's most advanced processors from data centres outside China. The Commerce Department is finalising rules that would require cloud operators in seven countries to obtain a US licence before receiving Blackwell-class chips, The Information reported this week, with an announcement possible as soon as November.

The plan, led by the department's Bureau of Industry and Security, would shift the chokepoint in American chip policy from the buyer of a processor to the company hosting it. Washington already bars direct shipments of Nvidia's most advanced AI chips to China, but Chinese firms can still train models on the same hardware by leasing computing power abroad, without any chip ever crossing a Chinese border.

Seven countries, one new chokepoint

According to accounts of The Information's reporting by Gizmodo and Forbes, operators in Singapore, Malaysia, Thailand, Indonesia, Brazil, Norway and South Africa would need a US licence before receiving Blackwell-class chips or anything more advanced. Applicants would have to disclose the final recipients and how the chips will be used, and accept subsequent verification.

Data centres run by American companies would get a carve-out: they could skip the licence if they can show they have blocked remote access by customers in China, Russia, Venezuela and other countries under US arms embargoes. Even then, they would still have to report their customers and the scale of computing they use. Commerce has published nothing yet, and the rules are still in interagency consultation.

Chinese firms are racing to lock in capacity first

Reuters reported on Thursday, citing three people familiar with the matter, that Chinese AI companies are already trying to reserve Nvidia-powered computing capacity at data centres outside China before the rule can land. One major Chinese cloud provider has discussed internally whether to lock in overseas capacity.

The rush makes sense if the licence attaches to chips an operator receives: hardware already installed before the rule takes effect could fall outside it. The Bureau of Industry and Security's May guidance similarly left existing data centres alone.

Legal doubts and a bill waiting in the Senate

Whether Commerce can actually regulate remote access to chips is contested. A lawyer at Baker McKenzie told The Information that the department lacks statutory authority under current law to regulate remote computing access. The department's own advisory opinions from 2009, 2011 and 2014 treated cloud providers as not exporting what customers do with their capacity.

Congress has a fix waiting. The Remote Access Security Act would give Commerce explicit authority over cloud-based access to restricted technology. The House passed it by 369 votes to 22 in January, but it is still awaiting action in the Senate. Reuters' sources said the scandal around Jensen Huang, who resigned as Nvidia chief executive on November 11, could add momentum to the bill.

Nvidia has lobbied against the measure, Gizmodo reported, arguing that blocking remote access would push China to build its own chipmaking capacity. The Financial Times reported on November 3 that Nvidia told cloud partners to sell at least 51 percent of services to non-Chinese customers on new contracts.

The dispute reaches into the cabinet: Treasury Secretary Scott Bessent reportedly confronted Commerce Secretary Howard Lutnick over Megaspeed, which Commerce denies. The New York Times reported on November 9 that Commerce's Megaspeed investigation stopped after a Nvidia lawyer contacted the agents, and the matter moved to the Justice Department. Megaspeed denies illegally diverting chips, and no charges have been reported.