Nvidia warns chip buyers of legal risks tied to China AI access, sources tell CNBC
Nvidia warned customers that remote access by Chinese companies to its AI chips could expose both the chipmaker and its customers to U.S. legal action. The warning concerns the proposed Remote Access Security Act, or RASA. It has not yet become law: the House passed it in January, but the Senate has not passed it. Nvidia’s message, shared with chip buyers in recent months, said the proposal could give the U.S. government broader authority over cloud services. The idea is to control access to computing power remotely, much as the government controls physical hardware. That matters because chips can be used through a data center abroad without being shipped to China. Nvidia also said it would not currently sell chips to customers who rent more than half of their GPUs to Chinese companies. Separately, the Financial Times reported a 51% non-Chinese-services requirement for future contracts. The Commerce Department has announced a separate interim rule aimed at preventing remote access. Nvidia and the department did not comment to CNBC.
What legal risk did Nvidia warn customers they could face if Chinese companies remotely accessed its AI chips?
Nvidia warned customers that remote access by Chinese companies to its AI chips could expose both the chipmaker and its customers to U.S. legal action. The warning concerns the proposed Remote Access Security Act, or RASA. It has not yet become law: the House passed it in January, but the Senate has not passed it.
Nvidia’s message, shared with chip buyers in recent months, said the proposal could give the U.S. government broader authority over cloud services. The idea is to control access to computing power remotely, much as the government controls physical hardware. That matters because chips can be used through a data center abroad without being shipped to China.
Nvidia also said it would not currently sell chips to customers who rent more than half of their GPUs to Chinese companies. Separately, the Financial Times reported a 51% non-Chinese-services requirement for future contracts. The Commerce Department has announced a separate interim rule aimed at preventing remote access. Nvidia and the department did not comment to CNBC.
What is the proposed Remote Access Security Act, and how could it extend U.S. control to cloud access?
The Remote Access Security Act, or RASA, is proposed legislation addressing access to restricted AI chips through cloud services. Its significance is that chips can be used remotely, including from data centers outside China, even when export controls restrict the hardware from going to China. Regulators have had limited powers over this kind of access.
Nvidia told chip buyers that RASA could give the U.S. government broader authority to control cloud services in the same way it controls physical hardware. In practice, that could bring remotely provided computing power under government oversight. Nvidia warned customers that the proposal could expose both them and Nvidia to legal risk if Chinese companies accessed the chips remotely.
RASA passed the House of Representatives in January, but it has not passed the Senate. Separately, the Commerce Department’s Bureau of Industry and Security announced an interim final rule designed to prevent Chinese companies’ remote access to AI chips. The article does not detail how that rule will operate or what final requirements it will impose.
What share of a customer's rented GPUs did Nvidia say could not go to Chinese companies?
Nvidia told customers that, for now, it would not sell chips to customers who rent more than 50% of their graphics processing units, or GPUs, to Chinese companies. The threshold is about the share of a customer’s rented GPUs going to Chinese companies, not the share of chips physically shipped to China.
The concern is that customers can provide computing power through cloud services, so Chinese companies may use chips hosted in data centers overseas. Nvidia’s stated threshold is one way it said it would limit sales to customers whose rental activity is heavily directed to Chinese companies. The article does not explain how Nvidia would verify each customer’s GPU rentals.
There is a separate figure from a Financial Times report earlier that month: Nvidia had verbally told data-center clients to direct at least 51% of their services to non-Chinese customers under future contracts. That reported contract guidance is not identical to the more-than-50% rental threshold described by CNBC’s sources. The article does not say whether either condition has since changed.
How can a company in China use restricted chips remotely through a data center in another country?
A company in China can, in general, rent computing capacity from a data center in another country and send work to it remotely. The chips stay in that overseas facility; the user receives computing services rather than taking physical possession of the hardware. The article notes that regulators have had limited powers over access through cloud services hosted overseas.
CNBC reported on Nov. 5 that Chinese lab Moonshot AI sought access to Nvidia chips through data centers in Europe. The Trump administration had previously singled out the company, accusing it in July of accessing chips through centers in East Asia. These are reported allegations and efforts described in the article, not a technical account of exactly how a connection worked.
This remote model matters because U.S. export controls restrict advanced chips from going to China, while overseas cloud access can still connect users to computing power. RASA would give the government broader authority over cloud services, and a separate Commerce Department interim rule is designed to prevent Chinese companies’ remote access. The article does not describe the rule’s technical enforcement.
What limits do U.S. export controls place on advanced AI chips going to China, and why?
The article says U.S. export controls restrict Nvidia’s advanced chips from going to China. The stated reason for keeping the chips out is to maintain the U.S. lead in the AI race. The restrictions concern sending advanced hardware to China, while remote access through overseas cloud services has created a separate challenge for regulators.
For example, CNBC reported that Moonshot AI sought access to Nvidia chips through data centers in Europe. The Trump administration had also accused the company in July of accessing chips through centers in East Asia. In these cases, the hardware is hosted outside China, while a company may seek to use its computing power remotely. The article does not say that the alleged access was proven.
Regulators have had limited powers to control overseas cloud access to chips. RASA, which passed the House but not the Senate, could expand government authority over cloud services. The Commerce Department’s Bureau of Industry and Security has also announced an interim final rule designed to prevent Chinese companies’ remote access to AI chips. Nvidia says it seeks to comply with export controls but relies on customer and partner programs after sale.
What happens when a company rents computing power from a cloud provider instead of owning the hardware?
In general, renting computing power from a cloud provider means a company uses computing capacity without buying or owning the physical machines. The provider operates the hardware, while the customer pays to use services remotely. This differs from owning a chip because the customer need not have the hardware delivered to its own premises.
For example, a data center can host Nvidia GPUs and provide computing services to customers elsewhere. The source describes Nvidia’s concern about customers renting GPUs to Chinese companies, and reports that Moonshot AI sought Nvidia-chip access through data centers in Europe. It does not give the technical details of how customers connected or how rental arrangements were monitored.
This distinction matters because U.S. export controls restrict advanced chips from going to China, while regulators have had limited powers over access through overseas cloud services. Nvidia warned that proposed RASA could let the government control cloud access in a way similar to physical hardware. The Commerce Department has also announced an interim final rule aimed at preventing Chinese companies’ remote access to AI chips.
What is a GPU, and why is it useful for training and running AI systems?
GPU stands for graphics processing unit, as the article explains. It is a type of computer processor. In general, GPUs are designed to handle many calculations in parallel, which makes them useful for graphics and for the large amounts of computation involved in many AI tasks. This background is general computing knowledge; the article does not explain GPU design.
The article focuses on Nvidia’s advanced AI chips and refers to them as GPUs. It says customers may rent these processors to Chinese companies, and that Chinese labs have sought access through overseas data centers. A rented GPU can provide computing capacity through a cloud service without the user owning or physically holding the chip.
That distinction is central to the policy debate described. U.S. export controls restrict advanced chips from going to China, but regulators have had limited powers over remote access through overseas cloud services. Nvidia warned that proposed RASA could give the government broader control over cloud services, similar to its control over physical hardware. A separate Commerce Department rule is also designed to prevent Chinese companies’ remote access to AI chips.
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