China Gains Access to Nvidia Chips via Foreign Data Centers

The U.S. government has heightened scrutiny on a method that enables Chinese AI companies to utilize advanced Nvidia chips while evading direct export restrictions. Instead of importing processors into China, these companies are renting computing power from data centers located abroad, according to CNBC.

The physical hardware remains in countries like Malaysia or Thailand, while Chinese developers access it remotely to train their models.

Understanding the Loophole

Since 2022, the U.S. has progressively restricted the supply of high-performance AI chips to China. The regulations primarily focus on the export of the physical equipment, which has allowed remote access to processors located outside the country to remain a separate issue for some time.

As a result, Chinese firms have begun renting Nvidia servers in Southeast Asia. Research from the Carnegie Endowment indicates that companies like Alibaba and ByteDance have benefited from this approach.

Growth of AI infrastructure in Southeast Asia. Source: CNBC/DC Byte.

The U.S. Bureau of Industry and Security is now examining these schemes more comprehensively. The agency is compiling lists of countries where banned chips are believed to physically enter China and those from which Chinese companies gain remote access. The latter scenario does not always violate existing export regulations.

An example of this is Alibaba's collaboration with Megaspeed, a Singapore-registered firm. The Chinese company accesses Nvidia chips located in Malaysia, but the agreement is not direct: it involves a Singaporean entity connected to a Cayman Islands company, which is ultimately owned by Alibaba.

Kimi K3 Sparks Debate

The issue of China circumventing restrictions gained traction following the launch of Kimi K3 by Moonshot AI. In July, Michael Kratsios, the director of the White House Office of Science and Technology Policy, stated that the startup had acquired servers with Nvidia GB300 and accessed these processors in Thailand, presumably for training its models. However, he did not provide public technical evidence to support his claims.

Kratsios also accused Moonshot AI of leveraging Anthropic Fable's capabilities in developing Kimi K3, which the Chinese side labeled as slander.

Just days earlier, the project unveiled an open model featuring 2.8 trillion parameters, achieving results comparable to leading American systems in various tests. The developers optimized GPU cores, including the use of Nvidia H200.

A similar situation arose with DeepSeek. In June 2025, a U.S. State Department representative stated that the startup attempted to use shell companies in Southeast Asia to bypass export controls and sought access to data centers in the region to work remotely with American chips.

Nvidia responded at the time that DeepSeek had utilized H800 chips acquired legally.

Congress Proposes Extended Controls on Cloud Access

The "Remote Access Security Act" aims to close this loophole. The House of Representatives approved the bill earlier this year in January.

This legislation expands export controls to include the use of technologies via the internet and cloud services. Thus, restrictions will apply not only to the physical shipment of chips but also to the ability to utilize their computational power from China. The bill still needs to pass the Senate.

Nvidia opposes the further tightening of restrictions, arguing that current regulations specifically allow for the establishment of cloud infrastructure outside controlled jurisdictions and that its partners undergo vetting.

The company has also warned that export bans have effectively deprived it of its second-largest commercial market, allowing competitors from other countries to strengthen their positions.

In April 2025, the U.S. introduced licensing for the H20 processor, which Nvidia had specially developed for China following previous restrictions. The company estimated the costs associated with this decision to be around $5.5 billion.

Later on, Washington softened its approach several times. In January 2026, authorities permitted the supply of more powerful H200 chips to certain Chinese clients under specific conditions.