The US government is preparing new trade rules to stop Chinese AI companies from renting computing capacity in overseas data centers. According to reports from The Information and CNBC, the policy targets Chinese firms accessing restricted Nvidia GB300 chips through cloud providers in Southeast Asia and Japan. Closing this cloud rental loophole marks a major shift from tracking physical hardware shipments to regulating remote computing services.
The loophole is simple. Earlier export control policies limited physical chip shipments exceeding 600 TOPS to China. However, those rules tracked the physical destination of the silicon rather than where the computing capacity traveled. Chinese developers quickly adapted. Companies began renting servers abroad to train frontier models without importing the physical hardware.
Federal officials recently identified Moonshot AI for running workloads on Nvidia GB300 chips located inside a Thailand data center. Other firms have established similar pathways. ByteDance partnered with Singapore cloud vendor Aolani to access computing clusters in Malaysia. Similar arrangements have appeared across Alibaba and Tencent. Because the hardware stays outside Chinese borders, these setups skirted existing export restrictions.
Lawmakers in Washington are moving to turn these restrictions into law. The House of Representatives passed the Remote Access Security Act, which currently sits with the Senate. This legislation would grant the federal government explicit authority to control remote access to restricted software and chips. It also introduces mandatory Know Your Customer rules for cloud operators.
Policy experts point out that stopping remote access is central to the core goals of American export policy. Michelle Nie, a visiting scholar at the Center for a New American Security, highlighted the urgency of the problem.
The purpose of chip export controls is to prevent China from using advanced American silicon to train frontier AI, and remote access models directly threaten that goal.
Regulating cloud access across international borders creates major technical hurdles. Cassia King, a senior researcher at the AI Policy and Strategy Institute, noted that the true difficulty lies in crafting rules that officials can practically enforce without disrupting lawful enterprise traffic.
The upcoming rules will create new compliance burdens for cloud providers and server operators worldwide. Global data center capacity is expanding rapidly, with commercial real estate firm JLL projecting total capacity will double to 200GW by 2030. Much of that construction is happening across Southeast Asia.
In Malaysia, Indonesia, and Thailand, developers have planned 31 data center projects with capacities exceeding 100MW. Only 2 of those massive facilities are currently operating. If Washington restricts cross border computing rentals, developers who designed business models around Chinese demand will need to rethink their investments.
