Chinese AI companies are trying to reserve Nvidia-powered computing capacity at data centers outside China before Washington can close the rental route, Reuters reported Thursday, citing three people familiar with the matter. The rule they fear, as described by The Information and relayed by Gizmodo's account of the report and Forbes' summary of the report, would put the licensing burden on cloud operators in seven countries rather than on Chinese renters, and would exempt US-run data centers that block remote access from China. Commerce has published nothing, and every account so far rests on anonymous sources or secondhand summaries.

If the rule is adopted as described, local operators in Singapore, Malaysia, Thailand, Brazil, Indonesia, Norway and South Africa would need a US license before receiving Blackwell-class chips and would have to report who gets them and how they will be used. That makes the story relevant to anyone who runs, finances or rents GPU capacity in those markets, and to anyone who wants to know where the chokepoint in US chip policy would sit if it moves from the chip's buyer to the chip's host.

Reuters Says Chinese Cloud Provider Weighed Locking In Overseas Capacity

One major Chinese cloud provider has discussed internally whether to lock in overseas Nvidia capacity, and some companies have already approached data-center operators about reserving it, Reuters' sources said. Two of the three sources added that smaller AI labs weighing similar moves, including approaches to large rented-compute providers in Southeast Asia.

Reuters did not name the cloud provider or the sources. Nvidia, Megaspeed, the US Commerce Department, Tencent and Alibaba did not immediately respond to its requests for comment. The companies fear that a scandal around Jensen Huang, who resigned as Nvidia chief executive on November 11, could add political momentum to the Remote Access Security Act and prompt tighter limits on remote use of advanced processors, according to the sources.

Is It Legal for Chinese Companies to Rent Nvidia Chips Abroad Today?

Largely yes, under rules as reported. US export controls bar shipments of Nvidia's most advanced AI chips to China, but BIS advisory opinions on cloud computing from 2009, 2011 and 2014 treated providers as not exporting what customers do with their capacity, leaving what lawyers call an apparent gap for remote access to chips. Under that reading, a Chinese lab can lease capacity from an operator in Malaysia or Singapore without the hardware ever crossing a Chinese border.

The practice is well documented. The Financial Times has reported that Alibaba and ByteDance moved training offshore, sending work on their newest language models to facilities in Singapore and Malaysia, according to TechSpot's summary. The Trump administration also told BIS not to enforce the Biden-era AI Diffusion Rule in May 2025.

Washington has narrowed one side of the gap. On May 31, the Bureau of Industry and Security issued guidance that licenses are required for advanced chips going to any entity whose ultimate parent is headquartered in China or Macau, wherever that entity sits. The guidance did not force existing data centers already running such chips to stop, and BIS framed it as a clarification rather than a new rule.

Reported Rule Would Require Licenses for Chip Deliveries to Operators in Seven Countries

The Information reported this week that Commerce is preparing a rule aimed at the rental route. The original article is behind a paywall, so these provisions rest on Gizmodo's account of it and Forbes' summary of the report:

• Licenses: Cloud providers in Singapore, Malaysia, Thailand, Brazil, Indonesia, Norway and South Africa would need a US license before receiving Blackwell-class chips or anything more advanced.
• Reporting: Recipients would have to report who receives the chips and how they will be used.
• A carve-out: US-operated data centers in those countries would need no license if they block remote access for customers in restricted countries such as China. Forbes says they would have to cut off China and Russia and still report their local customers.
• Timing: The rule could be announced in November, according to The Information, while Commerce consults other agencies.

One caveat applies. A separate summary of The Information's article says the draft covers seven additional countries but names only Malaysia, Singapore and Thailand and does not mention a license requirement or a carve-out, so the fuller details rest on the Gizmodo and Forbes accounts.

The design shifts the burden. Under the reported text, the party that needs a license is the local operator, not the Chinese renter, and the party that escapes licensing is the US-run facility next door. Forbes columnist Jon Markman's opinion piece argues for holding Nvidia and the large US clouds building in Southeast Asia, naming Amazon Web Services, Microsoft, Google and Oracle, and calls smaller GPU clouds the riskiest part of the AI trade. Norway's inclusion is notable: it was listed in Tier 1 under the Diffusion Rule, the framework the administration shelved.

Chips Already Installed Could Sit Outside License on New Deliveries

The accounts do not say whether capacity already installed would be covered. If the license attaches to chips an operator receives, as described, hardware racked before the rule takes effect could fall outside it. That is analysis, not reported fact, but it would explain why a lease signed now is worth rushing for: the May 31 guidance likewise left data centers already running such chips alone. A license on deliveries is also a traditional export-control hook, unlike a direct rule on remote access, though the reporting does not say how Commerce would answer the authority doubts described below.

Nvidia Lobbies Against Rule While Tightening Its Own Customer Terms

Nvidia has lobbied against the measure, Gizmodo reports, arguing that blocking remote access would push China to build its own chipmaking capacity and could give it an edge in the AI race.

Nvidia is also tightening its own terms. The Financial Times reported on November 3 that Nvidia told cloud partners to sell at least 51% of services to non-Chinese customers, on new contracts only, citing five people familiar with the matter. Nvidia neither confirmed nor denied the report when AFP asked, and a Singapore-based group told the FT that finding enough non-Chinese customers outside the US is difficult.

Bessent Reportedly Confronted Lutnick Over Megaspeed; Commerce Denies It

Singapore-based Megaspeed sits at the center of the political dispute. Reuters reported that the company denies illegally diverting chips to China. According to Gizmodo's account of Axios, Treasury Secretary Scott Bessent confronted Commerce Secretary Howard Lutnick over Megaspeed, and Lutnick reportedly warned that tighter restrictions would "start a trade war with China." Gizmodo says Axios and the New York Times both reported tensions between the two on November 11. The Commerce Department denies the confrontation took place.

Both officials have technology-policy records that bear on the dispute. Bessent has said that by 2028 the United States will hold about 80% of global computing power. Twenty-five congressional Democrats asked Commerce's inspector general to examine whether Lutnick's family ties to AI data centers conflict with his official actions in December 2025. Commerce officials said Lutnick had no direct role in the projects and met all ethics obligations.

Commerce Paused Megaspeed Probe After Nvidia Lawyer Contacted Agents, Times Reports

The New York Times reported on November 9, as summarized by The Next Web's account, that Commerce's Megaspeed investigation stopped after an Nvidia lawyer contacted the agents on the case, and that the matter moved to the Justice Department, which is running a criminal inquiry, according to people familiar with the matter. The Next Web said it had not independently verified the inquiry. The Justice Department asked Commerce to pause its work, and Commerce complied. A Commerce spokesman told the Times the department handled the case strictly by the book, and Nvidia spokesman John Rizzo described the company's contacts with officials as a way to answer questions and request guidance. Megaspeed denies illegally diverting chips, and no charges have been reported.

Other agencies have proposed adding Megaspeed to the Entity List, Commerce's trade blacklist, but Commerce has not acted and has added no company to the list in more than a year, the Times reported.

Reuters Reports Alibaba Subsidiary Financed Megaspeed-Linked Firm

Singapore business filings reviewed by Reuters show that Alibaba subsidiary Apex Enterprise Solutions provided financing to Speedmatrix, a Malaysian company linked to Megaspeed. Asia Times reported in June that Megaspeed bought nearly $2 billion of chips from Nvidia through a Malaysian subsidiary.

Two sources also told Reuters that Alice Huang, a former Megaspeed executive whom the Times has linked to Jensen Huang, had business and investment ties with several leading Chinese technology companies, including Tencent and Alibaba. Reuters could not verify the allegation that she is a Chinese intelligence asset, and China denies it. Alibaba and Tencent did not immediately comment.

Reuters Reports Gray-Market Stockpiling and PLA-Linked H200 Purchase

The expected crackdown is also lifting demand for restricted chips on China's gray market, Reuters reported, citing two sources. Traders and intermediaries have begun stockpiling processors, fearing a broader US move against smuggling of chips and AI servers. Reuters also reported that a Chinese university with deep ties to the People's Liberation Army bought Nvidia H200 chips last month despite being on a US export blacklist, according to procurement records.

Senate Bill and Authority Doubts Shape Whether Any Rule Holds Up

A second track runs through Congress. The Remote Access Security Act, H.R. 2683, would give Commerce explicit authority over cloud-based access to restricted technology. The House passed it 369-22 on January 12, and it awaits Senate action. Reuters' sources said the companies fear the Huang controversy could add momentum to it.

The authority question matters for any Commerce rule. A Baker McKenzie attorney told The Information that Commerce lacks statutory authority under current law to regulate remote computing access, according to a summary of that article, which suggests a rule could face legal challenge. A separate legislative vehicle is slower: the House passed its FY2027 defense authorization bill in July, the Senate has not, and a compromise text is expected during the lame-duck session alongside a December 11 government funding deadline, though Reed Smith's tracker of trade provisions does not mention AI chip remote-access measures.

Unnamed Sources and Unpublished Text Leave Four Questions Open

• Reuters sourcing: All of it is anonymous, and the Chinese cloud provider is unnamed.
• Link between the two reports: Reuters ties the Chinese rush to the Huang scandal and the Senate bill. The seven-country licensing rule comes from The Information. Whether it prompted the reservations is inference.
• The rule itself: It has not been published, and its scope, country list and timing could change.
• Any Commerce action on Megaspeed: The Times reported that other agencies have proposed adding it to the Entity List. Commerce has not acted.

Nvidia is scheduled to report quarterly earnings at about 4:20 p.m. ET on Tuesday, November 17, when investors may get more detail on how much China-linked demand is in its numbers.

Frequently Asked Questions

Can Chinese companies legally rent Nvidia chips in another country?

Under rules as reported, yes. Export controls bar shipping the most advanced chips to China, but BIS has said it does not treat cloud computing services as exports, and the reported rule has not been published.

Which countries would the reported Commerce rule cover?

Singapore, Malaysia, Thailand, Brazil, Indonesia, Norway and South Africa, according to Gizmodo and Forbes accounts of The Information's report. A separate summary names only three of them.

Would capacity Chinese firms reserve now still work after a rule?

The reporting does not say. If a license applies to chips an operator receives, hardware installed earlier might fall outside it, but that is analysis, and the rule text is unpublished.

What does the Remote Access Security Act do, and where does it stand?

It would give Commerce explicit authority over cloud-based access to restricted technology. The House passed it 369-22 in January, and it awaits Senate action.