American regulators have closely focused on a scheme that has turned export restrictions on Nvidia chips into a mere formality. This refers to the practice where Chinese artificial intelligence developers do not import advanced processors into the PRC but instead rent computing power in foreign data centers, keeping the physical equipment outside U.S. jurisdiction.
This approach has become especially popular in Southeast Asia. Servers with Nvidia chips are placed in Malaysia, Thailand, or Indonesia, while Chinese engineers connect to them remotely to train their models. In my estimation, this development vector has become a logical response to the tightening of export controls, which since 2022 has consistently restricted the supply of the most powerful AI chips to China. Formally, the rules govern physical exports, leaving a "gray zone" for cloud computing.
How the loophole works
The Carnegie Endowment think tank confirms that giants like Alibaba and ByteDance have obtained computing power this way. The Alibaba example is telling: the corporation collaborates with Singapore-based Megaspeed, which provides access to Nvidia chips installed in Malaysia. The chain is structured through a Cayman Islands entity, whose ultimate beneficiary is Alibaba itself. This is a classic scheme with intermediary jurisdictions that allows for legal distancing from a direct violation.
The U.S. Bureau of Industry and Security is currently systematizing such schemes, compiling lists of countries where Chinese companies gain remote access to prohibited equipment. However, as experts emphasize, such access in itself does not always violate existing export rules, creating a serious legal conflict.
Kimi K3 as a catalyst for discussion
The topic gained particular urgency after the release of the open-source model Kimi K3 by Moonshot AI. In July, the director of the White House Office of Science and Technology Policy, Michael Kratios, stated that the startup allegedly gained access to servers with Nvidia GB300 in Thailand to train its models. He provided no public evidence, but the very fact of such a statement from a high-ranking official signals the seriousness of Washington's intentions. Moonshot AI, in turn, called the accusations slander, which only added to the tension.
Earlier, a similar story unfolded around DeepSeek: the U.S. State Department accused the startup of attempting to use shell companies in Southeast Asia. Nvidia then countered that DeepSeek had worked on legally acquired H800 chips.
Legislative response
To close this loophole, the House of Representatives has already approved the "Remote Access Security Act." The document extends export controls to the use of technology via the internet and cloud services, which would effectively ban Chinese companies from renting capacity of American chips abroad. The bill still needs to pass the Senate, but its adoption seems only a matter of time.
Nvidia opposes further tightening, pointing out that current rules deliberately allow the creation of cloud infrastructure outside controlled jurisdictions. The manufacturer also warns that export bans have already cost it the second-largest market and play into the hands of competitors. Recall that after the introduction of H20 licensing in April 2025, Nvidia estimated its losses at $5.5 billion, although Washington later softened its stance by allowing H200 shipments to select customers.
My analysis: It is clear that the U.S. is entering an endless arms race in the AI sector. Each new restriction gives rise to more sophisticated circumvention schemes, and attempts to regulate cloud computing could lead to fragmentation of the global internet. Chinese companies have already proven their ability to adapt, and the likely next step will be the development of their own chips or the use of alternative architectures. Investors should closely monitor these processes: volatility in the AI sector will only increase.