2 Sources
[1]
The U.S. banned Nvidia's best chips from going to China. Now it's trying to close a crucial loophole
* Chinese AI firms are reportedly accessing advanced Nvidia computing power through data centers in Southeast Asia, despite U.S. restrictions on exporting the company's most advanced chips directly to the country. * This exploits a loophole in the U.S. export-control regime, which is more focused
[2]
Chinese AI firms accessing banned Nvidia chips via Southeast Asia
Moonshot AI, ByteDance, Alibaba, Tencent, and other Chinese AI firms have reportedly obtained access to restricted Nvidia $NVDA chip compute by routing through data centers located in Thailand, Malaysia, Japan, and other parts of Southeast Asia, according to CNBC. Cassia King, who works as a
Share
Copy Link
Chinese AI firms including Moonshot AI, ByteDance, Alibaba and Tencent are reportedly accessing restricted Nvidia computing power through data centers in Thailand, Malaysia and other Southeast Asian countries. This exploits a loophole in US export controls that regulate physical chip ownership but not remote cloud access to their computing power.
Chinese AI firms are circumventing US export restrictions by accessing banned Nvidia chips through data centers in Southeast Asia, exposing a critical gap in Washington's strategy to maintain technological superiority over Beijing. Companies including Moonshot AI, ByteDance, Alibaba and Tencent have reportedly obtained compute power from Nvidia's most advanced semiconductors via facilities in Thailand, Malaysia and Japan, despite US export controls prohibiting direct sales of these chips to China
1
2
.The arrangement is currently legal because US export controls focus on physical chip ownership rather than remote cloud access. Cassia King, senior researcher at the Institute for AI Policy and Strategy, confirmed that the US export control regime "controls physical AI chips. It does not cover remote access to those chips"
1
2
. This loophole in US export controls allows Chinese companies to train frontier AI systems using computing power they cannot legally purchase.In July, White House official Michael Kratsios accused Moonshot AI of accessing Nvidia's GB300 chips through a facility in Thailand to develop its Kimi K3 model
1
. The GB300 is part of Nvidia's Blackwell chip generation, which the US prohibits from sale to Chinese companies. Moonshot AI released Kimi K3 as one of several new Chinese AI models that have demonstrated significant performance improvements in recent months, joining systems from DeepSeek and Alibaba that have scored well on performance benchmarks1
.Industry observers identify access to advanced compute via overseas cloud providers as a key factor enabling Chinese AI models to gain capability in the intensifying US-China AI race. The ability to train models using Nvidia chips remotely provides Chinese AI firms with computational resources comparable to their Western competitors, despite physical export restrictions
1
.ByteDance was working with Singapore-headquartered Aolani, a cloud provider with Nvidia chips, to access compute capacity in Malaysia, according to sources familiar with the matter
1
2
. The Wall Street Journal first reported this arrangement in March. Aolani stated it works "with a global and diversified customer base spanning customers from North America and Asia" and emphasized that "companies we service do not have ownership, potential future claim or physical access to the chips that power our solutions"1
.The cloud provider insisted any access to its services, AI infrastructure or technology is "fully compliant with all applicable regulations"
1
. This highlights how data centers in Southeast Asia have become critical intermediaries, providing Chinese firms with legal pathways to computing power from chips they cannot directly import.Related Stories
AI infrastructure buildouts across Southeast Asia are accelerating as companies rush to meet growing demand for advanced compute capacity. Real estate company JLL estimates that global data center capacity could roughly double to 200GW by 2030
1
2
. Malaysia, Indonesia and Thailand have 31 planned data centers of at least 100MW capacity, compared to just two facilities of that scale currently operational, according to DC Byte data1
2
.This infrastructure expansion positions Southeast Asia as a strategic hub for accessing banned Nvidia chips and advanced computing resources. The region's regulatory environment and geographic proximity to China make it an attractive location for facilities that can serve Chinese clients while technically complying with US export restrictions.
Michelle Nie, visiting fellow in technology and national security at the Center for a New American Security, described the loophole as "threatening US national security" and argued that "the point of chip export controls is to deny China the ability to train frontier AI using advanced US chips"
1
2
.The proposed Remote Access Security Act (RASA) seeks to expand US export controls to include remote cloud access to critical hardware and software. The legislation passed the House of Representatives in January but has yet to clear the Senate
1
2
. Even if RASA becomes law, Nie cautioned that passage alone would not immediately close the loophole. While the act would establish legal authority to regulate remote cloud access, a separate rulemaking process would be required before that authority could be exercised over advanced chips2
.Cloud providers are expected to resist the compliance burden RASA would impose. King noted that with White House backing, the Bureau of Industry and Security could potentially produce implementing rules within days, but crafting effective regulations with real enforcement teeth would require decisions about which chips fall under the rule's scope and how to build a workable customer verification system
2
. Watch for Senate action on RASA and subsequent rulemaking efforts that will determine whether the US can effectively regulate remote access to controlled technology and close this pathway for Chinese AI firms accessing Nvidia computing power.Summarized by
Navi
[1]
1
Technology

2
Policy and Regulation

3
Technology
