NVIDIA has warned customers that allowing Chinese companies to remotely access its artificial intelligence (AI) chips through overseas data centres could expose both the chipmaker and its buyers to legal action under proposed US legislation. The warning, reported by CNBC on November 17, 2026, comes as Washington moves to extend export-control oversight beyond physical semiconductor shipments to computing capacity delivered through cloud services.
According to two people familiar with NVIDIA’s private communications, the company has cautioned customers about the proposed Remote Access Security Act (RASA), which would give the US government broader authority to regulate remote access to advanced American computing technology. NVIDIA has also reportedly told customers that it will not sell chips to buyers that rent more than 50% of their graphics processing units (GPUs) to Chinese companies.
The development signals a significant change in how the AI semiconductor industry manages international customers. Instead of examining only where chips are physically delivered, manufacturers and regulators are increasingly concerned about which companies ultimately use the computing power.
Key takeaways
- NVIDIA reportedly warned chip buyers that proposed US legislation could expose both customers and the company to legal action over Chinese access to AI computing resources.
- The proposed Remote Access Security Act seeks to extend export-control authority to remote access through cloud computing services.
- NVIDIA has reportedly introduced a commercial restriction against selling chips to customers that rent more than 50% of their GPUs to Chinese companies.
- A separate Financial Times report said NVIDIA instructed data-centre partners to direct at least 51% of their services to non-Chinese customers under future contracts.
- The US House of Representatives passed RASA in January 2026, but the legislation was subsequently referred to the Senate.
- On November 16, the US Commerce Department’s Bureau of Industry and Security announced a separate interim final rule addressing remote access to advanced American semiconductors.
Why NVIDIA is warning customers about China
The central issue is the difference between exporting a physical AI chip and providing access to its computing capabilities through an overseas data centre.
US export restrictions already limit the shipment of certain advanced American semiconductors to China. However, cloud computing creates a more complicated situation.
A Chinese AI company does not necessarily need to own or physically import an advanced GPU to use its processing power. It can potentially rent computing capacity from a cloud provider operating in another country.
For example, a data-centre operator in Europe or Southeast Asia could purchase NVIDIA GPUs, install them in its facilities and offer computing services to customers internationally.
A Chinese AI developer could then rent access to those GPUs remotely to train or operate AI models.
The physical hardware would remain outside China, but the computing resources could still be used by a Chinese company.
This distinction has become increasingly important as AI development requires large quantities of specialised computing infrastructure.
NVIDIA’s reported warning reflects concerns that this arrangement could create legal exposure if US authorities determine that remote access violates applicable export-control requirements.
What is the Remote Access Security Act?
The Remote Access Security Act, or RASA, is proposed US legislation designed to address potential gaps in existing export-control regulations.
The legislation would expand the government’s ability to regulate remote access to controlled technologies under the Export Control Reform Act of 2018.
The bill is intended to prevent foreign entities from accessing restricted American technology through overseas infrastructure when direct physical exports would otherwise be prohibited.
The US House passed H.R. 2683, the Remote Access Security Act, on January 12, 2026.
According to the official House voting record, the legislation received 369 votes in favour and 22 against.
It was received in the Senate on January 13, read twice and referred to the Senate Committee on Banking, Housing, and Urban Affairs. The official legislative record does not establish that it has subsequently become law.
This distinction matters because the proposed legislation and the existing regulatory framework are not the same thing.
NVIDIA’s reported customer warnings concern the potential consequences of the proposed legislation. They should not be interpreted as confirmation that RASA has already been enacted.
How the proposed restrictions could affect cloud computing
The legislation could change the commercial relationship between semiconductor manufacturers, cloud infrastructure providers and AI developers.
Under a conventional hardware export-control framework, authorities primarily examine the destination, recipient and intended use of controlled products.
Remote computing introduces another layer.
A GPU can remain inside a data centre in a country permitted to receive the hardware while its computing resources are accessed by a company in a restricted market.
The proposed approach would give US authorities broader tools to examine this type of access.
This could affect several participants.
1. NVIDIA and semiconductor manufacturers
NVIDIA supplies GPUs that power many of the world’s advanced AI data centres.
The company therefore has an interest in ensuring that its products are not used in ways that could violate US export-control requirements.
If customers or infrastructure partners become subject to additional restrictions, NVIDIA could face increased compliance obligations.
The company may also need more detailed information about the ultimate users of its chips and the customers accessing GPU computing capacity.
2. Cloud computing providers
Cloud providers could face greater pressure to identify their customers and monitor how computing resources are being used.
A company purchasing thousands of GPUs might previously have focused on building infrastructure and securing customers.
Under a stricter regulatory environment, it could also need to demonstrate that its customer relationships comply with applicable restrictions.
This may increase legal expenses, customer verification requirements and operational complexity.
3. Chinese AI companies
Chinese AI developers could face fewer opportunities to access advanced American GPUs through international cloud providers.
Even when a company cannot directly purchase restricted hardware, overseas leasing arrangements may provide another route to computing capacity.
If that route becomes subject to stricter regulation, Chinese developers could find it more difficult or expensive to secure the computing resources required for advanced AI development.
4. International data-centre operators
Infrastructure companies in Europe, Singapore, Malaysia and other markets could face additional uncertainty.
They may need to reconsider customer concentration, contract structures and the geographic distribution of their computing services.
The impact could extend beyond companies physically operating in China because the regulations would concern access to computing resources rather than simply the location of the hardware.
NVIDIA’s reported 50% restriction and the 51% customer requirement
Two related reports have brought additional attention to NVIDIA’s approach to overseas cloud customers.
According to CNBC’s November 17 report, NVIDIA has told chip buyers that it would not sell GPUs to customers renting more than 50% of their computing capacity to Chinese companies.
Separately, the Financial Times reported on November 3 that NVIDIA had instructed global data-centre and cloud partners to ensure that at least 51% of their services went to non-Chinese customers.
These figures describe related but distinct reported restrictions.
The first concerns NVIDIA’s reported willingness to sell chips to particular customers. The second concerns the customer mix of cloud and data-centre partners.
The 51% requirement would leave a maximum of 49% of services for Chinese customers if applied as a straightforward allocation rule.
However, neither report establishes that every NVIDIA customer worldwide is subject to an identical publicly documented contractual requirement.
The Financial Times also reported that the instructions applied to future contracts rather than existing leases.
This suggests NVIDIA is attempting to manage potential regulatory exposure through commercial relationships even before the proposed legislation becomes law.
US government moves to regulate remote access to AI chips
The legislative debate is taking place alongside a separate regulatory development.
On November 16, 2026, the US Department of Commerce’s Bureau of Industry and Security (BIS) announced an interim final rule addressing remote access to advanced American superintelligence technology.
The agency said the measure was designed to prevent restrictions on sensitive American technology from being circumvented through remote computing arrangements involving data centres in third countries.
The accompanying regulatory document, titled “Addressing Remote Access to Advanced Semiconductors”, establishes new restrictions involving overseas data centres that remotely supply significant quantities of computing capacity to companies headquartered in certain restricted jurisdictions.
It also introduces additional licensing requirements for specified advanced semiconductor transactions.
The BIS announcement is important because it demonstrates that the US government is not relying exclusively on Congress to address remote access.
The interim final rule is a separate regulatory instrument from RASA.
Consequently, businesses need to distinguish between obligations under existing and newly issued regulations and the additional authority that the proposed legislation could provide.
For international cloud providers, this creates a more complicated compliance environment.
Why overseas data centres have become important to China’s AI industry
Advanced AI models require substantial computing resources.
Training involves processing large quantities of data and repeatedly adjusting model parameters. Inference, meanwhile, involves running a trained model to generate responses, analyse information or perform other tasks.
Both activities can require specialised GPUs, particularly when companies operate large AI models.
NVIDIA’s Blackwell architecture is among the advanced computing platforms attracting demand from AI developers and infrastructure providers.
When direct access to certain processors is restricted, overseas data centres can become commercially attractive alternatives.
A November 5 CNBC report said Chinese AI developer Moonshot AI had explored leasing NVIDIA Blackwell GPU capacity from European infrastructure providers.
The discussions reportedly involved large GPU clusters and were not evidence that a completed leasing agreement had been secured.
The report illustrates why US policymakers are increasingly interested in remote computing arrangements.
The issue is not limited to the physical movement of semiconductor products. It also concerns where the resulting computing power is used and who benefits from it.
This is particularly relevant because modern AI infrastructure is increasingly distributed across international data-centre networks.
What NVIDIA’s strategy means for the global AI infrastructure market
NVIDIA’s reported restrictions could influence how cloud providers build and commercialise their infrastructure.
Greater importance of customer verification
Cloud operators may increasingly need to identify the ultimate users of GPU services rather than relying solely on the identity of the company signing a contract.
This could involve additional checks on ownership, corporate relationships and the intended use of computing capacity.
Changes in data-centre investment decisions
Companies investing billions of dollars in GPU infrastructure need confidence that they can sell computing capacity to customers.
If a substantial portion of potential demand comes from Chinese AI developers, restrictions could affect the financial attractiveness of particular projects.
Some operators may prioritise customers in the United States, Europe, India and other markets to reduce exposure to regulatory uncertainty.
This is an analytical possibility rather than a confirmed shift across the industry.
Higher compliance costs
Monitoring end users and customer relationships could require additional legal, technical and administrative resources.
Cloud companies may also need stronger contractual protections to address potential regulatory changes.
These expenses could affect the economics of GPU leasing, particularly for smaller infrastructure providers.
More fragmented AI infrastructure
The AI computing market could become increasingly divided along geopolitical lines.
Companies may have to select infrastructure providers based not only on price, performance and availability but also on regulatory eligibility.
That could make international AI development more complicated.
What does this mean for India?
India could become relevant to this development because it is expanding its ambitions in AI infrastructure, cloud computing and domestic technology development.
The country has an opportunity to attract data-centre investment and build computing capacity for domestic enterprises, research institutions and AI startups.
However, NVIDIA’s reported restrictions also highlight an important consideration for Indian infrastructure operators.
Operating a data centre in India would not automatically make every customer eligible to access every category of restricted American computing technology.
The regulatory treatment of the hardware, its end users, ownership structures and remote access arrangements would still matter.
Indian cloud providers and data-centre companies working with advanced US semiconductors may therefore need robust customer verification and compliance procedures.
For Indian AI startups, the development could have two opposing effects.
On one hand, stricter restrictions on Chinese access could redirect some international demand towards infrastructure providers in other countries.
On the other hand, additional compliance requirements could make it more complicated for Indian companies serving international customers.
India could benefit from the changing geography of AI infrastructure, but the opportunity will depend on investment, power availability, connectivity, hardware access and regulatory clarity.
There is currently no verified evidence that NVIDIA’s reported customer restrictions automatically create a new commercial advantage for Indian cloud companies.
The bigger picture: AI chip restrictions are becoming access restrictions
The most important development is the changing definition of access to advanced computing technology.
For years, export-control discussions largely focused on the movement of physical products across national borders.
Cloud computing has complicated that approach.
A company can now use specialised hardware located thousands of kilometres away without owning or importing it.
For regulators, the distinction between owning a chip and accessing its computing capabilities has become increasingly important.
For semiconductor manufacturers, the challenge is ensuring that their products reach legitimate customers while navigating different national security requirements.
For cloud providers, the challenge is building commercially viable infrastructure without exposing themselves to regulatory violations.
NVIDIA’s reported warnings demonstrate how these concerns are already influencing commercial decisions.
The broader implications extend beyond NVIDIA. Other semiconductor manufacturers, server companies and cloud infrastructure operators could face similar questions as governments examine how advanced computing resources are distributed internationally.
What happens next?
The next important development will be the regulatory treatment of remote access under the new BIS rule and the legislative progress of the Remote Access Security Act.
Cloud providers will also need to understand how the new requirements affect existing customer agreements, future infrastructure investments and international service delivery.
The exact scope of NVIDIA’s reported 50% restriction and the reported 51% non-Chinese customer requirement will remain important questions because the company has not publicly disclosed a comprehensive set of contractual terms covering these arrangements.
The larger issue is whether governments can regulate access to computing power as effectively as they regulate the export of physical semiconductor hardware.
As AI infrastructure becomes increasingly international, that question could influence where data centres are built, which customers they serve and how the global AI industry develops.
Frequently asked questions
1. Why has NVIDIA warned customers about selling AI chips to companies serving China?
According to CNBC, NVIDIA warned customers that proposed US legislation could expose the company and buyers to legal action if Chinese companies remotely access advanced AI chips through overseas data centres. The company has also reportedly introduced restrictions on certain customers renting GPU capacity to Chinese companies.
2. What is the Remote Access Security Act?
The Remote Access Security Act is proposed US legislation intended to expand government authority over remote access to controlled technologies. It aims to address situations where restricted computing resources are accessed through overseas cloud infrastructure rather than physically exported to a restricted country.
3. Has the Remote Access Security Act become law?
No. The US House passed H.R. 2683 on January 12, 2026, by 369 votes to 22. The bill was referred to the Senate Committee on Banking, Housing, and Urban Affairs on January 13. It should not be confused with the separate BIS interim final rule announced on November 16.
4. Could NVIDIA’s restrictions affect Indian cloud companies?
Potentially. Indian infrastructure providers using advanced American GPUs may need to strengthen customer verification and regulatory compliance. However, operating in India does not automatically exempt a company from applicable US export-control requirements. The actual impact will depend on the hardware, customers and relevant regulations.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.



