Chinese artificial intelligence companies are reportedly circumventing United States export restrictions by accessing advanced Nvidia computing power through data centers located in Southeast Asia. This practice exploits a loophole in current U.S. export control regulations, which primarily focus on the physical ownership of chips rather than remote access to their computational capabilities.
Despite U.S. restrictions on exporting advanced semiconductors like GB300s directly to China, several Chinese firms have reportedly gained access to the compute power of these chips through facilities in countries such as Thailand, Malaysia, and Japan. Moonshot AI, for example, released a new model in July, with a White House official, Michael Kratsios, alleging the company used Nvidia GB300 chips via a Thai facility. Other Chinese AI developers, including DeepSeek and Alibaba, have also recently introduced new AI systems that have performed well in benchmarks, with industry observers attributing this progress partly to access to advanced compute through overseas cloud providers.
Cassia King, a senior researcher at the Institute for AI Policy and Strategy, stated that Moonshot’s reported access through a Thai facility is legal as long as the company does not directly purchase and own the physical hardware. King explained that the U.S. export control regime governs physical AI chips but does not extend to remote access.
Chinese hyperscalers, including ByteDance, Alibaba, and Tencent, have reportedly accessed Nvidia chip compute power remotely through other Asian nations. ByteDance, for instance, was reportedly working with Singapore-headquartered Aolani, a cloud provider with Nvidia chips, to access compute in Malaysia. Aolani confirmed working with a global customer base and stated that its clients do not own or have physical access to the chips powering its solutions, asserting full compliance with applicable regulations.
The U.S. government is considering new legislation to address this loophole. The proposed Remote Access Security Act (RASA) aims to expand U.S. export controls to include remote cloud-based access to critical hardware and software. This act passed the House of Representatives in January but still requires Senate approval. Michelle Nie, a visiting fellow at the Center for a New American Security, indicated that this loophole threatens U.S. national security by undermining the objective of denying China the ability to train frontier AI using advanced U.S. chips. Nie also noted that RASA could face industry pushback due to the compliance burden it would place on cloud providers for know-your-customer and customer verification requirements.
Even if RASA passes, it would grant the U.S. government the authority to regulate remote access, but a specific rule would still need to be created to export-control remote access to advanced chips. King suggested that the Bureau of Industry and Security could implement such a rule quickly with White House support, though challenges would remain in making it effective and enforceable, particularly in defining what compute is covered, who is prohibited from accessing it, and how to implement a robust know-your-customer scheme.
The booming AI infrastructure buildouts in Southeast Asia reflect a growing market for advanced compute. JLL, a real estate company, estimates that global data center capacity could nearly double to 200GW by 2030. DC Byte data indicates 31 planned 100MW+ data centers across Malaysia, Indonesia, and Thailand, compared to just two currently.
