중국 기업들은 미국 AI 칩 수출 통제를 우회하기 위해 대체 조달 경로를 개발하는 한편, 워싱턴은 규제 내 클라우드 컴퓨팅 허점을 막기 위한 조치를 동시에 추진하고 있다.

The Trump administration is considering new restrictions to prevent Chinese AI companies from remotely accessing advanced U.S. chips through overseas data centers, addressing another potential loophole in export controls designed to limit Beijing’s access to cutting-edge computing power. According to an Aug. 28 report by The Information citing four people familiar with the discussions, a small group within the Commerce Department is developing a replacement for the Biden administration’s discarded AI Diffusion Rule.

A primary target of the proposal is Chinese AI firms accessing advanced Nvidia processors housed in overseas data centers, including facilities in Thailand and Singapore. The chips never need to cross the Chinese border; instead, customers can connect remotely to rent computing power for training AI models. This presents a novel challenge for an export-control regime traditionally built around tracking physical shipments. U.S. lawmakers have increasingly argued that blocking direct GPU sales to China is ineffective if Chinese firms can secure equivalent computing power through foreign servers.

Rep. Josh Gottheimer (D-N.J.) emphasized this point in June, introducing bipartisan legislation targeting cloud access. “We can’t let our adversaries — especially China — dodge our export controls by simply renting what they can’t buy,” he said, according to the House Select Committee on the Chinese Communist Party. These proposed cloud restrictions emerge alongside U.S. and allied investigations into alleged diversion networks that use third countries, shell companies, and falsified shipping documents to route advanced computing equipment to China. The cloud-computing issue represents only the latest front in a broader struggle over China’s access to advanced American AI technology.

On Aug. 25, Taiwanese prosecutors charged nine individuals in connection with an alleged scheme to illegally ship servers equipped with Nvidia’s advanced B300 processors to China. According to the Associated Press, prosecutors stated that 74 servers successfully reached China after being routed through Indonesia, Japan, and Hong Kong, while another 56 were intercepted in Taiwan. Among those charged was an Nvidia manager and former employees of server manufacturer Super Micro Computer. Prosecutors alleged the defendants created shell companies and websites, falsifying end-use documentation to conceal the equipment’s true destination. Both Nvidia and Super Micro confirmed they are cooperating with authorities.

This case follows multiple U.S. investigations into similar diversion tactics. In March, the Justice Department charged a Chinese national and two U.S. citizens with conspiring to illegally acquire millions of dollars in export-controlled chips and route them to China via Thailand. Prosecutors alleged the defendants concealed the chips’ true destination. Assistant Attorney General for National Security John Eisenberg stated the hardware represented “the best of American ingenuity” and affirmed the department’s commitment to pursuing export control violations. In a separate March case, the Justice Department alleged that three individuals attempted to divert high-performance, U.S.-assembled AI servers to China using false paperwork, dummy hardware, and complex transshipment routes. Prosecutors described the operation as an effort to obscure China’s role as the ultimate destination through “false documents, staged dummy servers,” and layered shipping schemes. Additionally, in December 2025, U.S. authorities announced the seizure of over $50 million in Nvidia technology and cash following an operation against a separate China-linked trafficking network. One defendant and his company pleaded guilty after admitting to exporting or attempting to export at least $160 million in controlled Nvidia H100 and H200 GPUs.

In an April investigation, the House Select Committee on the Chinese Communist Party concluded that China acquires advanced AI capabilities through a mix of legitimate purchases and illicit methods. The committee’s report noted that China “utilizes sophisticated smuggling networks to acquire restricted AI chips” and recommended strengthening both physical export controls and restrictions on remote access to computing infrastructure.

The growing reliance on overseas infrastructure has compelled U.S. policymakers to redefine what constitutes an AI technology “export.” Traditionally, physically shipping an Nvidia GPU to China allows authorities to examine customs records, shipping documents, serial numbers, and end-user declarations. Under the cloud model, however, a data center in Thailand or Singapore can legally house thousands of U.S.-made processors while a remote customer accesses them via the internet. In this scenario, the customer purchases computing capacity rather than taking physical possession of the hardware. Awareness of the loophole intensified in July after White House Office of Science and Technology Policy Director Michael Kratsios raised concerns that Chinese AI developer Moonshot AI may have accessed advanced Nvidia processors hosted in Thailand, according to The Information.

In response, the Commerce Department has already begun tightening regulations governing Chinese companies operating abroad. In May, the Bureau of Industry and Security (BIS) issued guidance clarifying that advanced computing products may still require an export license when supplied to entities outside China if the company or its ultimate parent is headquartered in China or another restricted jurisdiction. Current BIS regulations also mandate that certain applicants verify data centers’ “Know Your Customer” procedures and identify remote Infrastructure-as-a-Service users based in China and other restricted nations. The forthcoming rule could expand these requirements, mandating that overseas operators of advanced American chips verify the identity of users accessing their GPU clusters and monitor how that computing power is utilized.

Janet Kim, an export-control attorney at Baker McKenzie, told The Information that it is “widely acknowledged” the Commerce Department currently lacks clear authority to regulate direct remote access. To address this gap, Congress passed the bipartisan Remote Access Security Act in January, which would explicitly extend U.S. export controls to cover remote access to controlled technology via cloud-computing services. “The CCP’s AI ambitions are being fueled by its access to American chips housed in data centers located outside of China,” said Select Committee Chairman John Moolenaar (R-Mich.) following the bill’s passage.

While the bill remains pending in Congress, the Commerce Department continues exploring enforcement options under existing regulatory authority. According to The Information, the new rule could be circulated to AI companies and industry groups for comment as early as September, though the proposal remains subject to revision or delay. Nevertheless, the broader policy trajectory is becoming clear. U.S. technology controls initially focused on preventing the most advanced chips from physically reaching China. However, alleged diversion networks have proven that hardware can be rerouted through third countries, while cloud computing introduces a fundamentally different vector: keeping the chips overseas while granting Chinese firms remote access to their processing power. For U.S. policymakers, closing this loophole means export controls will increasingly track not only the physical destination of an AI chip, but also the ultimate users of its computational capacity.