Washington’s strategy of restricting China’s access to Nvidia’s most advanced artificial intelligence chips is facing a significant loophole: Chinese companies may be able to use the computing power of those chips without ever taking physical possession of them.
Several Chinese AI companies have reportedly accessed advanced Nvidia processors through data centers and cloud providers in Southeast Asia and other parts of Asia, exposing a gap in U.S. export controls that generally focus on where chips are shipped and who owns them rather than who can remotely use their computing capacity.
The issue is becoming crucial as Washington and Beijing compete for leadership in artificial intelligence. Nvidia’s most powerful processors are among the most critical pieces of infrastructure for training frontier AI models, making control over access to them a central element of U.S. efforts to slow China’s progress.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Yet the emergence of overseas compute markets means that restricting physical shipments to China may not be enough.
Chinese AI Firms Turn To Overseas Compute
The issue came into sharper focus after Moonshot AI released its Kimi K3 model in July.
Less than a week later, White House official Michael Kratsios accused Moonshot of using Nvidia GB300 chips through a facility in Thailand. Kimi K3 is among a new generation of Chinese AI models that have made significant gains in recent months. DeepSeek and Alibaba have also released models that have performed strongly on industry benchmarks.
Industry observers say access to large amounts of advanced computing capacity has been an important factor behind the rapid improvement of Chinese AI systems.
Cassia King, a senior researcher on the Compute Policy team at the Institute for AI Policy and Strategy, said Moonshot’s reported access to computing power in Thailand could be legal under current rules if the company did not actually purchase or own the physical chips.
“The U.S. export control regime controls physical AI chips. It does not cover remote access to those chips,” King said.
That creates a potentially important gap. A company in China may be prohibited from importing an advanced Nvidia processor, while a cloud provider outside China could potentially operate the same processor and sell computing time to Chinese customers.
The customer receives the computational output rather than the chip itself.
Southeast Asia Becomes A Strategic Computing Hub
Reports indicate that Chinese technology companies including ByteDance, Alibaba and Tencent have accessed Nvidia computing capacity remotely through facilities in countries such as Thailand, Malaysia and Japan.
ByteDance was reportedly working with Singapore-headquartered cloud provider Aolani to obtain access to Nvidia-powered computing infrastructure in Malaysia, according to a source familiar with the arrangement. The Wall Street Journal previously reported the arrangement.
Aolani said it serves a “global and diversified customer base spanning customers from North America and Asia.”
“The companies we service do not have ownership, potential future claim or physical access to the chips that power our solutions,” an Aolani spokesperson said, adding that permitted access to its services and infrastructure complies with applicable regulations.
The growth of Southeast Asia’s data-center industry makes the issue harder for Washington to contain.
Companies are investing heavily in data centers across Malaysia, Indonesia and Thailand as demand for AI computing increases. JLL estimates that global data-center capacity could roughly double to 200 gigawatts by 2030.
DC Byte data show 31 planned data centers of at least 100 megawatts across Malaysia, Indonesia and Thailand, compared with only two currently operating facilities of that scale.
That expansion is creating an increasingly important pool of computing infrastructure close to China, potentially giving Chinese AI companies another route to advanced processing capacity even as direct shipments of restricted chips remain tightly controlled.
Why Remote Access Matters
The strategic importance goes beyond individual Chinese companies gaining access to particular Nvidia chips. Training advanced AI models requires enormous amounts of computing power. If Chinese developers can obtain sufficient access to high-end processors through overseas cloud infrastructure, U.S. restrictions on physical chip exports may have a weaker effect on their ability to develop competitive models.
Michelle Nie, a visiting fellow in technology and national security at the Center for a New American Security, described the loophole as a threat to U.S. national security.
“The point of chip export controls is to deny China the ability to train frontier AI using advanced U.S. chips,” Nie said.
The challenge for Washington is therefore shifting from controlling the movement of hardware to controlling access to computing capacity itself.
That is considerably more complicated.
A physical semiconductor can be tracked through export documentation, ownership and shipping records. Computing power delivered remotely through the cloud can potentially be accessed by customers in another country without the underlying hardware crossing a border.
Lawmakers are considering legislation aimed at closing the gap.
The proposed Remote Access Security Act, or RASA, would expand U.S. export controls to cover remote, cloud-based access to certain critical hardware and software. The legislation passed the House of Representatives in January but has yet to clear the Senate.
If enacted, the measure could give U.S. authorities a legal framework for regulating who can remotely access advanced computing infrastructure. But passing the legislation would not immediately resolve the problem.
Nie said the bill would give the government authority to regulate remote access, but the administration would still need to establish specific rules governing which computing resources are restricted and which customers should be prohibited from using them.
King said the Bureau of Industry and Security could potentially move quickly once it had the necessary authority, but designing an effective system would be more difficult.
“The challenge will be in making a rule that’s effective and enforceable,” she said.
Policymakers would need to determine which levels of computing capacity should be covered, which customers should be restricted, and how cloud providers could implement reliable know-your-customer and customer-verification systems.
But the compliance burden could fall on cloud providers.
Under a remote-access regime, cloud providers operating advanced Nvidia hardware could potentially become responsible for screening customers and monitoring how their computing resources are being used. That could increase compliance costs and complicate the business of serving international customers.
Nvidia’s Chips Remain At The Center Of The AI Contest
The dispute reveals that Nvidia has become so important to the geopolitical competition over AI. The company’s advanced accelerators provide much of the computing power required to train and operate the world’s most capable AI models. Washington has therefore treated access to Nvidia’s most advanced processors as a strategic issue rather than simply a commercial trade matter.
But the emergence of overseas data centers shows that controlling the hardware supply chain and controlling access to computing power are two different challenges.
China does not necessarily need to import every restricted chip if companies can obtain access to the same processors through cloud infrastructure elsewhere.