Chinese AI firms rush to secure overseas cloud compute before new U.S. rules

The cloud access loophole

Current U.S. export controls are designed to prevent the physical shipment of high-end Nvidia AI processors to China. However, these rules do not explicitly stop Chinese companies from renting computing power from data centers located in other countries. This has allowed firms to maintain their AI development by accessing hardware remotely.

Chinese AI companies have reportedly utilized data centers in various Asian nations, including Thailand, Malaysia, and Japan, to get around these restrictions. By using these overseas hubs, they can run complex models on hardware that would otherwise be blocked from entering Chinese territory. This practice creates a significant gray area in international trade and technology policy.

The U.S. government is now moving to address this specific loophole. The Commerce Department is currently finalizing a rule intended to stop Chinese AI firms from obtaining advanced chips through these overseas cloud providers. This represents a shift toward regulating the digital infrastructure of AI rather than just the physical hardware.

Legislative pressure in D.C.

The House of Representatives recently approved a measure known as the Remote Access Security Act, or RASA. This bill would grant the Commerce Department the legal authority to oversee how restricted technology is accessed through the cloud. It aims to ensure that foreign entities cannot use remote connections to bypass existing export bans.

The bill passed the House in January with a 369-22 vote, though it has not yet been acted upon by the Senate. Lawmakers are also pushing to include additional chip-control measures in this year’s National Defense Authorization Act. These efforts reflect a growing urgency to tighten the net around Chinese access to high-performance computing.

Experts note that RASA alone might not be enough to fully close the loophole. While it establishes the legal authority to regulate remote access, a separate rulemaking process would still be required before that authority could be applied to specific advanced chips. This suggests that the regulatory landscape will remain in flux as the government determines the best way to enforce these limits.

Corporate scramble for capacity

In response to these looming regulations, major Chinese cloud providers are holding internal talks about reserving capacity abroad. These companies are worried that the new rules will build support for RASA and lead to a sudden loss of access to essential hardware. They are attempting to lock in their computing needs before the rules are officially implemented.

Smaller AI developers are following a similar path and are reaching out to compute-rental firms in Southeast Asia. This rush to secure resources is a direct reaction to the possibility of being cut off from the high-end Nvidia chips required for training large models. The uncertainty is already impacting the financial markets for these companies.

The market reaction was swift, with shares of Alibaba, Z.ai, and MiniMax falling in Hong Kong on Thursday. Alibaba saw its shares drop by as much as 7.1% at its low point, while the other two companies lost 7% or more. These fluctuations highlight the high stakes involved in the ongoing trade and technology dispute.

Gray markets and investigations

The situation has also triggered activity in China's gray market, where traders and intermediaries have started stockpiling restricted Nvidia chips. This suggests that even as the government tightens the rules, there is still a high demand for hardware that is officially restricted. These unofficial channels provide a way for entities to acquire technology despite the official barriers.

Simultaneously, U.S. authorities are scrutinizing companies that may have aided Chinese clients in obtaining export-controlled chips. One example is Megaspeed, a cloud company based in Singapore, which has been under investigation for its role in these transactions. The company has denied any illegal activity and has engaged a U.S. law firm to catalog its graphics processing units for inspectors.

The scrutiny extends to the individuals involved in these operations as well. Reports have highlighted connections between cloud executives and intelligence assets, adding a layer of national security concern to the trade issue. These investigations underscore the complexity of monitoring technology flow in a globalized cloud environment where data and hardware can be accessed from almost anywhere.