American regulators are seriously concerned about a new scheme that allows Chinese artificial intelligence developers to use advanced Nvidia graphics processors while bypassing existing export restrictions. The essence of the loophole is simple: instead of physically importing chips into the PRC, companies rent computing power in foreign data centers while maintaining full control over the model training process.
The equipment physically remains outside China—for example, in Malaysia or Thailand. Chinese engineers connect to the servers remotely, using them for the most resource-intensive tasks, including training large language models.
The Mechanics of Bypassing Restrictions
Since 2022, Washington has consistently tightened rules on supplying the most powerful AI chips to the PRC. However, the restrictions primarily concerned the export of physical equipment, which left a legal loophole for remote access to processors located outside the country.
As a result, Chinese tech giants, including Alibaba and ByteDance, are actively renting Nvidia servers in Southeast Asia. In my estimation, this practice has become so systematic that the U.S. Bureau of Industry and Security (BIS) has been forced to compile separate lists of countries where Chinese companies gain access to prohibited chips. Notably, such rental itself does not always violate existing export regulations.
A characteristic example is Alibaba's cooperation with Singapore-based Megaspeed. The Chinese giant gains access to Nvidia chips located in Malaysia through a complex chain of legal structures, including a company in the Cayman Islands. This scheme allows formal compliance with the letter of the law while circumventing its spirit.
Kimi K3 and the Escalation of the Discussion
The topic of bypassing restrictions came to the forefront after the release of the open-source model Kimi K3 by Chinese Moonshot AI. In July, White House Office of Science and Technology Policy Director Michael Kratios stated that the startup had acquired servers with Nvidia GB300 and had access to them in Thailand, presumably for model training. He provided no public technical evidence, but the very fact of such a statement at the highest level speaks to the seriousness of the problem.
Additional context: a few days earlier, Moonshot AI unveiled a model with 2.8 trillion parameters that showed results on par with leading American systems in certain tests. To optimize GPU cores, the developers used, among others, Nvidia H200.
A similar situation previously arose around DeepSeek. In June 2025, a U.S. State Department representative accused the startup of attempting to use shell companies in Southeast Asia to bypass export controls and seeking access to data centers in the region. Nvidia then responded that DeepSeek had used legally acquired H800 chips.
Legislative Initiatives and Nvidia's Position
To close the loophole, the House of Representatives approved the "Remote Access Security Act" in January. The document extends export controls to the use of technology via the internet and cloud services, which would make remote use of chips from China impossible. The bill still needs to pass the Senate.
Nvidia is categorically opposed to further expanding restrictions. The company points out that current rules specifically allow the creation of cloud infrastructure outside controlled jurisdictions and that its partners undergo rigorous vetting. The manufacturer also emphasizes that export bans have already effectively deprived it of its second-largest commercial market and are helping competitors from other countries strengthen their positions.
In April 2025, the U.S. introduced licensing for the H20—a processor specifically designed for China after previous restrictions. Nvidia estimated the costs associated with this decision at approximately $5.5 billion. Washington later softened its approach several times: in January 2026, authorities allowed the supply of more powerful H200 chips to select Chinese clients under certain conditions.
My analysis: This situation demonstrates a fundamental problem with modern export controls: in the era of cloud computing, the physical location of a chip ceases to be the determining factor. The U.S. is trying to regulate technology that is inherently cross-border, and this creates an endless game of cat and mouse. Until Washington finds a way to control access to computing power, not just its physical movement, China will continue to find workarounds. The question is merely which side adapts faster to the new realities.