American regulators are seriously concerned about a new scheme that allows Chinese AI companies to bypass export restrictions on advanced Nvidia chips. Instead of directly shipping processors to China, developers rent computing power abroad. The equipment is physically located in Malaysia or Thailand, while Chinese engineers connect to it remotely to train their models. This creates a fundamental challenge for the entire export control system.
The Mechanics of the Gray Zone
Since 2022, the United States has consistently tightened rules on shipments of the most powerful AI chips to China. However, the restrictions primarily target the physical export of equipment, leaving a legal vacuum for remote access to computing resources outside the country. Chinese giants are actively exploiting this.
Based on my data, corporations such as Alibaba and ByteDance have been drawn into this scheme's orbit. They rent servers in Southeast Asia, where the number of large data centers is rapidly growing. In Malaysia, for example, facilities with capacities of 100 MW or more are already operating or under construction, while similar sites are being built in Indonesia and Thailand. The U.S. Bureau of Industry and Security is now systematically examining these chains, compiling lists of transit countries, but legally such remote access does not always violate current regulations.
The Alibaba case is telling: the company gains access to Nvidia chips through Singapore-based firm Megaspeed, which is linked to an offshore structure in the Cayman Islands. This multi-layered scheme is deliberately convoluted to make it difficult to trace the ultimate beneficiary.
The Kimi K3 Incident
The discussion intensified after the release of the open-source model Kimi K3 by Moonshot AI. In July, White House Office of Science and Technology Policy Director Michael Kratsios stated that the startup had rented servers with Nvidia GB300 chips in Thailand, presumably for training. He provided no public evidence but accused Moonshot of using Anthropic Fable's developments. The Chinese side dismissed these accusations as slander. Notably, the model itself, with 2.8 trillion parameters, showed results comparable to the best American counterparts, indirectly confirming access to powerful GPUs.
A similar scandal previously erupted around DeepSeek. The State Department accused the startup of attempting to use shell companies in Southeast Asia to bypass controls. Nvidia then countered that DeepSeek had used legally purchased H800 chips.
Legal Response and Nvidia's Position
To close the loophole, the House of Representatives approved the "Remote Access Security Act," which extends export controls to cloud computing. Now restrictions could apply not only to physical shipments but also to the very ability to utilize a chip from China. The bill still needs to pass the Senate, and its fate is uncertain.
Nvidia categorically opposes expanding the restrictions. The company argues that current rules deliberately allow cloud infrastructure outside controlled jurisdictions, and its partners undergo vetting. Meanwhile, the manufacturer is already suffering enormous losses: licensing the H20 alone in April 2025 cost it $5.5 billion. Washington later softened its stance, allowing shipments of more powerful H200 chips to select clients in January 2026 under certain conditions.
My comment: This struggle resembles a game of cat and mouse, where technological progress always outpaces regulation. While the U.S. tries to control physical exports, China is mastering virtual space, calling into question the entire effectiveness of the embargo. In the long term, such a policy only stimulates the development of alternative chips and cloud ecosystems outside U.S. jurisdiction.