JupiteX Get the app
Economy & Business18 Nov 2026 · about 7 min

Nvidia warns customers of new US rule that could expose them to lawsuits for serving Chinese clients

The brief

RASA is the Remote Access Security Act, a U.S. bill aimed at closing a route around export controls: using advanced U.S.-technology chips remotely instead of importing them. It matters because the article says Chinese AI companies could obtain computing power from Nvidia GPUs in overseas data centers. The mechanism is rental access. A company places its data in a data center abroad, where controlled Nvidia chips are installed, and sends its workloads there. The chips stay in the host country; the customer receives computing services rather than shipping the hardware into China. The article identifies this as the loophole RASA addresses. The House has passed RASA, but it has remained in the Senate for more than ten months. Separately, the Commerce Department’s Bureau of Industry and Security released an interim final rule described as closing the loophole. The article does not say the Senate has enacted RASA, so the bill’s status and the interim rule should not be conflated.

01

What is the Remote Access Security Act (RASA), and what kind of access to AI chips is it meant to restrict?

RASA is the Remote Access Security Act, a U.S. bill aimed at closing a route around export controls: using advanced U.S.-technology chips remotely instead of importing them. It matters because the article says Chinese AI companies could obtain computing power from Nvidia GPUs in overseas data centers.

The mechanism is rental access. A company places its data in a data center abroad, where controlled Nvidia chips are installed, and sends its workloads there. The chips stay in the host country; the customer receives computing services rather than shipping the hardware into China. The article identifies this as the loophole RASA addresses.

The House has passed RASA, but it has remained in the Senate for more than ten months. Separately, the Commerce Department’s Bureau of Industry and Security released an interim final rule described as closing the loophole. The article does not say the Senate has enacted RASA, so the bill’s status and the interim rule should not be conflated.

02

How can a company use powerful U.S.-designed AI chips in a foreign data center without importing the chips itself?

The key distinction is between importing a physical chip and accessing its computing power. U.S. export controls restrict some Nvidia chips from being shipped to China, but the article describes companies seeking to use chips installed in data centers abroad. That can let them access powerful hardware without taking delivery of it.

The customer sends its data to the overseas facility and uses the Nvidia GPUs there to run computing tasks. The chips remain at the data center; the customer receives the service remotely. The article points to Chinese AI lab Moonshot apparently pursuing this approach in Europe, and says Alibaba and ByteDance have been alleged to train models in Southeast Asia to bypass restrictions.

The Commerce Department’s interim final rule is described as closing this loophole, while the House-passed RASA also addresses remote access. If access is restricted, companies relying on this arrangement may need to find other ways to obtain computing power. The article does not detail the technical setup of individual rentals.

03

What steps has Nvidia taken to monitor its data-center partners and reduce the risk of violating U.S. export controls?

Nvidia has introduced partner requirements aimed at reducing the risk that its chips serve restricted customers through remote data-center access. The article reports two steps: partners must direct a specified share of their services to non-Chinese clients, and they must join Nvidia’s Fleet Intelligence tracking system to keep official partner status.

The reported threshold is 51% of a partner’s services for non-Chinese clients. Fleet Intelligence is identified as a tracking system, although the article does not explain what data it collects or how Nvidia uses it. Together, the requirements give Nvidia a way to set conditions for its partners and monitor their participation in its official network.

Nvidia has also established an oversight committee for national security policies, export-control compliance, and global trade risk management. The committee is to work with outside experts in law, national security, and trade compliance. The article says it is unclear whether these changes were planned earlier or were prompted by renewed scrutiny. It does not describe the system’s enforcement details.

04

What legal risks could data-center operators and their customers face if restricted clients remotely use these chips?

The reported risk is legal exposure for companies involved in remote access to restricted chips. Nvidia has warned customers that RASA could create legal risk for both chip buyers and Nvidia itself. The article’s headline also raises the possibility of data-center operators facing lawsuits for serving Chinese clients, but it does not name specific claims or penalties.

The concern is that a restricted customer could use U.S.-technology GPUs installed in a foreign data center without importing the chips. In the article’s account, the customer sends data to the overseas facility and gets computing access remotely. RASA is intended to address that route, while the Commerce Department’s Bureau of Industry and Security has released an interim final rule described as closing the loophole.

The House passed RASA, but it had been in the Senate for more than ten months. The article says it is unclear whether the bill will be approved, and does not establish what legal consequences would apply to a particular operator or customer. Nvidia’s warning concerns possible risk, not a reported lawsuit or finding of wrongdoing.

05

What share of a partner’s services does Nvidia reportedly require to go to non-Chinese clients?

The reported figure is 51%: Nvidia has started requiring partners to direct at least that share of their services to non-Chinese clients. The requirement is one of the company’s steps to reduce risk around remote access to its AI chips, as U.S. officials and lawmakers focus on ways Chinese clients might use them.

The article places the threshold alongside a second partner condition. Partners must join Nvidia’s Fleet Intelligence tracking system if they want to retain official partner status. It does not explain how the system tracks usage, what information it collects, or whether every data-center arrangement is covered. The 51% figure refers to the share of services, not the share of chips or customers.

Nvidia’s reported requirements come as the Commerce Department has released an interim final rule described as closing a remote-access loophole. The article says it is unclear whether the partner changes were planned or were a response to renewed scrutiny. It does not say how Nvidia checks compliance with the 51% requirement.

06

If access to Nvidia GPUs becomes harder, what alternatives might Chinese AI labs use?

If access to Nvidia GPUs becomes harder, Chinese AI labs may turn to domestic chip makers. The article says China has several domestic chip makers, but it does not name them or say how their products compare with Nvidia’s. It also does not identify a particular replacement that labs could adopt.

The reason replacing Nvidia may be difficult, according to the article, is that many AI operators still prefer its chips for performance and reliability. It reports allegations that Alibaba and ByteDance train large language models in Southeast Asia to get around bans and restrictions. That example concerns where computing takes place, not a named alternative chip.

The article’s forward-looking point is that if the U.S. shuts down remote access to Nvidia GPUs through overseas data centers, Chinese companies could scramble to find replacements for the chips they can access now. It does not predict which companies would succeed or what hardware they would choose. Domestic chips are mentioned as a possible source, not a confirmed substitute.

07

Why do advanced GPUs make it possible to train large AI models, and why are they difficult to replace?

In general, GPUs are useful for AI training because they can perform many mathematical operations in parallel. Training a large model involves repeating calculations across large amounts of data, so this parallel work can speed up the process. This technical explanation is general background; the article itself does not describe how GPUs train models.

The article’s concrete point is that Chinese AI operators still prefer Nvidia chips for their performance and reliability. It says Chinese companies can access Nvidia hardware through foreign data centers, even when direct chip imports are restricted. That access matters because the chips provide computing capacity for training AI models, including large language models.

Replacing those GPUs may be difficult for operators who rely on their performance and reliability, but the article gives no technical comparison or reason beyond those stated advantages. It notes that China has several domestic chip makers, without naming products or saying whether they can match Nvidia. If remote access is closed, the article expects companies to seek replacements, but does not say what they will use.

This brief was written by AI from the original reporting and checked by other models. Names, figures and quotes come from the source; read it for full context.

Read more in the JupiteX app

Pulse is free. New stories every 4 hours, each one broken into the questions that explain it.

Or read more news on the web