American regulators are seriously concerned about a new scheme that allows Chinese artificial intelligence developers to use advanced Nvidia chips without formally violating export bans. This is not about smuggling silicon, but a more sophisticated approach: renting computing power in foreign data centers.
Physically, servers with GPUs remain on the territory of third countries—for example, in Malaysia or Thailand. Chinese engineers connect to them remotely and calmly train their models, staying within the legal framework, since formally the equipment does not cross the PRC border.
The mechanics of circumvention and first examples
Since 2022, Washington has been consistently tightening rules on the supply of the most powerful AI chips to China. However, the restrictions primarily concern physical exports. Remote access to processors remained outside the spotlight for a long time, which Chinese businesses took advantage of.
According to my data, a network of data centers focused on renting AI capacity is actively growing in Southeast Asia. Among the clients of such services are Alibaba and ByteDance. The scheme looks quite legal: a Singaporean intermediary rents capacity in Malaysia, and a Chinese company gains access to it through a complex corporate chain.
The example of Alibaba and Singapore's Megaspeed is telling. The agreement is not structured directly: a Cayman Islands entity is involved in the chain, with the Chinese giant as the ultimate beneficiary. Formally, everything is legal, but in essence, it is a classic sanctions circumvention scheme.
Political resonance and new initiatives
The topic gained new momentum after the release of the open-source model Kimi K3 from Moonshot AI. In July, White House Office of Science and Technology Policy Director Michael Kratios stated that the startup had rented servers with Nvidia GB300 in Thailand, presumably for training. He provided no public evidence, but the signal was sent.
A similar story arose around DeepSeek. In June 2025, the State Department accused the startup of attempting to use shell companies in Southeast Asia. Nvidia then countered that DeepSeek had used legally purchased H800 chips.
To close the loophole, the House of Representatives has already approved the "Remote Access Security Act." The document extends export controls to cloud services and remote use of technologies. Now restrictions will apply not only to the physical delivery of a chip but also to the ability to utilize its power from China. However, the bill still needs to pass the Senate.
My assessment of the situation
Nvidia opposes the tightening, rightly pointing to financial losses. The company estimated losses from the H20 ban at approximately $5.5 billion, and in January 2026, Washington even allowed H200 shipments to select clients. However, in my view, the administrative fight against capacity rental is a game of cat and mouse. Technology knows no borders, and as long as demand exists, ways to satisfy it will be found. The question is only how harsh Washington's response will be and whether it will lead to a final technological split between the US and China.