US Tightens Curbs on Chinese AI Labs Renting Chips Overseas

Washington is closing a remote-access route that lets Chinese AI labs use US chips in overseas data centers, as Moonshot AI reportedly seeks Blackwell capacity in Europe.

By USA 24 Editorial TeamNovember 6, 20266 min read

Low-angle view of server racks in a dim data hall, with blue status lights along the aisles and fiber cables overhead. — Illustration: USA 24

ShareX Facebook LinkedIn Reddit Email

Key takeaways

Moonshot AI reportedly sought Nvidia Blackwell capacity at European data centers, according to CNBC's November 5 report citing three sources.
Blackwell chips are currently barred from import into China, which is why renting capacity abroad has become the workaround under scrutiny.
Bureau of Industry and Security guidance dated May 31, 2026 closed the loophole for direct purchases by Chinese-headquartered subsidiaries but did not force existing arrangements to stop.
An August draft Commerce rule would require operators to block remote access by Chinese companies in countries such as Thailand and Singapore, and its final status is unconfirmed.
The House-passed Remote Access Security Act would extend AMD and Nvidia AI chip export controls to cloud computing, and Senate action has not been confirmed.

Moonshot AI, the Chinese lab behind the Kimi models, has sought leases on Nvidia Blackwell capacity at European data centers, according to a CNBC report published November 5 that cited three sources. The detail matters because Blackwell chips are currently barred from import into China. Renting computing time on hardware installed abroad is a way around that barrier, and Washington is now trying to close it.

The White House said in July that Moonshot had already reached Blackwell chips through a facility in Thailand. The European talks therefore look less like a first attempt than a next one. Nvidia chips in overseas data centers are becoming a front in export enforcement, and the rules are still being written.

A note on sourcing: several items in this story reached USA 24 only as headlines and short summaries, so the details below stick to what those versions confirm. The full text of any new federal rule was not reviewed, and the reporting did not confirm dollar figures or chip counts.

How Nvidia chips in overseas data centers reach China

US export controls have mostly targeted physical shipments. A chip that leaves a US port for a Chinese buyer is easy to track. A chip that stays in a data center in Thailand or Singapore and gets rented by the hour is a different problem, and the rules have long treated that arrangement as a gray area. CNBC reported in August that Chinese firms had reportedly accessed Nvidia GB300-class compute through Southeast Asian data centers while the most advanced chips were restricted [4].

Digitimes describes the same squeeze from the supply side. Renting capacity in data centers outside China has been a key route for Chinese AI developers to reach advanced US chips, and that route is narrowing [2]. The pressure is landing on Southeast Asian operators that built capacity around Chinese demand, while Beijing presses its own tech giants to rely less on American hardware [2].

A timeline of the tightening

The tightening has arrived in pieces rather than in one sweeping document. The reporting describes this sequence:

January 2026: Authorities allowed more powerful H200 chips to certain Chinese clients under specific conditions, according to an AI news roundup [3].
May 31, 2026: A Bureau of Industry and Security (BIS) guidance closed the loophole for direct chip purchases by Chinese-headquartered subsidiaries. It did not require data centers already running prior arrangements to stop [6].
August 7, 2026: TechTimes reported that BIS was targeting legal cloud compute as Chinese AI firms bypass export controls [6]. Bloomberg also reported that BIS was reviewing Chinese firms' access to overseas compute.
August 2026: A draft Commerce rule would require data-center operators or major chip buyers to block remote access by Chinese companies when exporting advanced chips to countries such as Thailand and Singapore [5]. Whether that draft has been finalized is unconfirmed.
November 5, 2026: CNBC and Digitimes reported that the remote-access channel is still open and that Washington is widening chip controls beyond US borders [1][2].

An AI news roundup dated November 5 described the development as an extension of chip export controls to overseas data centers [3]. USA 24 could not open that article, so the exact scope of any rule remains unclear. Treat the November 5 coverage as a clear signal of direction, not a final rulebook.

The administration has defended its record. In August, an unnamed White House official told CNBC, "The Trump administration has implemented the most rigorous export control regime in modern history" [4]. The gaps described in this reporting are the test of that claim, and they are still open.

Congress wants the rule written into law

Lawmakers are moving on a parallel track. Tom's Hardware reported that the House passed the Remote Access Security Act, a bill that would extend export controls on advanced AI chips from AMD and Nvidia to cloud computing [7]. That source's date is unclear, and USA 24 could not confirm any Senate action.

John Moolenaar, a cosponsor of the bill, put the argument bluntly: "The CCP's AI ambitions are being fueled by its access to American chips" [7]. The logic is that a chip control means little if the same hardware can be rented from a building abroad. The bill would make controls follow the compute rather than the shipment.

The difference between a rule and a statute matters here. A Commerce rule can move quickly, but a future administration can narrow or rescind it. A law would last longer, but it has to clear the Senate and the president first.

Who gains and who loses

Nvidia and AMD face the most direct demand risk. CNBC reported that US chip designers could see demand from data-center operators serving Chinese clients shrink if the rules tighten [4]. The reporting does not put a number on that exposure, so investors should treat it as a direction, not a forecast.

Cloud and data-center operators in the US, Southeast Asia and Europe may face new compliance duties for customers from China [2][5]. A European neocloud that rents Blackwell capacity could be asked to screen customers by ownership and location, not only by the name on the contract. Data-center expansion is also drawing political pushback at home, a pressure that OpenAI's pledge to cover full data-center power costs after midterm backlash illustrates.

Western AI labs compete in the same race. CNBC's coverage ties these rules to a pricing and training contest in which OpenAI and Anthropic are competing with Chinese developers [1]. Limits on Chinese access to compute could narrow that gap, though the reporting does not say by how much.

Chinese labs and Southeast Asian operators are the most exposed. Digitimes reports that Beijing is pushing its own tech giants toward domestic hardware, which means the squeeze is coming from both Washington and Beijing [2]. Ordinary consumers face little direct effect, because the rules govern who can rent advanced compute rather than which apps people can use.

Enforcement is the hard part

Rules only work if someone checks them. BIS has reportedly stepped up investigations into leasing at overseas data centers and into smuggling cases [5]. The enforcement problem is that a rented GPU leaves no shipping record, so regulators have to trace access rather than cargo.

The Justice Department has its own effort on AI misuse, a task force that USA 24 covered in a separate story. The reporting for this story does not tie that effort to chip leasing, so it should not be read as part of the same action.

What businesses renting offshore GPUs should check

Individuals have no official government guidance on this issue, as far as the reporting shows. Companies that rent GPU capacity abroad have more to do:

Ask providers in writing whether they screen customers by headquarters and ownership, including Chinese-headquartered subsidiaries.
Read the customer-screening and compliance terms in your contract, especially clauses on end users, subleasing and remote access.
Ask whether the provider has a plan to block remote access by Chinese companies in case the August draft rule takes effect in that form.
Keep records of where your workloads run and who can reach them.
Check the Federal Register and the BIS website for final text before signing multi-year commitments.

A provider that cannot answer these questions is a compliance risk, whatever the final rule says.

What to watch next

Several questions remain open, and each one moves on its own timeline. The first is the text of the November 5 action, which should appear in the Federal Register and on the BIS website. The second is whether the August draft is final, and whether it covers all remote access or only designated countries. Then comes the Senate, where USA 24 could not confirm any action on the Remote Access Security Act. Moonshot's European talks are also unresolved, since CNBC's sources say they are ongoing. Finally, no reporting quantifies how much Chinese demand has already moved to Europe, Southeast Asia or the Middle East.

The next concrete signal will likely come from the cloud side. If a European neocloud or Southeast Asian operator changes its customer terms in public, the rules will have a visible price. If nothing visible changes, the remote-access channel stays open, and the policy fight will continue on paper while the rentals continue in data halls.

Frequently asked questions

Are Nvidia Blackwell chips allowed in China?

Blackwell chips are currently barred from import into China under US export controls, according to CNBC's November 5 reporting. The new tightening targets a different route, which is renting compute on Blackwell hardware in data centers outside China. CNBC reported that Moonshot AI sought Blackwell capacity in Europe, and the White House said in July that the company had reached Blackwell chips through a facility in Thailand.

Is the remote access loophole for Chinese AI labs closed?

Not fully. BIS guidance dated May 31, 2026 closed the route for direct chip purchases by Chinese-headquartered subsidiaries, but it did not require data centers already running prior arrangements to stop. CNBC and Digitimes reported on November 5 that the remote-access channel is still open. A draft Commerce rule from August would require operators to block remote access by Chinese companies, but its final status is unconfirmed.

What is the Remote Access Security Act?

Tom's Hardware reported that the House passed the Remote Access Security Act, which would extend export controls on advanced AI chips from AMD and Nvidia to cloud computing. The date of that report is unclear, and USA 24 could not confirm any Senate action on the bill. Cosponsor John Moolenaar has argued that China's AI ambitions depend on access to American chips.

Does this affect ordinary consumers?

The direct effect on everyday users appears small, because the rules govern who can rent advanced computing capacity rather than which apps people use. The heavier impact falls on chip designers, cloud operators and businesses that rent offshore GPU capacity. Companies in that last group should review their providers' customer-screening and compliance terms.

Sources

#Nvidia #Export Controls #AI Chips #China #Data Centers

About this article. Reported and produced by the USA 24 newsroom using the sources cited above. Read our editorial policy. Spotted an error? Tell us and we will correct it.