Chinese AI firms rush to secure overseas Nvidia chip capacity

The rush for remote capacity

Chinese artificial intelligence companies are currently moving to secure access to Nvidia-powered computing capacity located outside of China. This activity comes as the industry anticipates that Washington might tighten restrictions on how Chinese developers use advanced processors remotely. By renting capacity at overseas data centers, these firms can currently use high-end chips without importing the physical hardware into Chinese territory.

Several major Chinese cloud providers have held internal discussions regarding the necessity of reserving this overseas infrastructure. Smaller AI developers are also participating in this trend by contacting major providers of rented computing power in Southeast Asia. These companies are trying to lock in their needs before potential regulatory changes make such remote access more difficult or expensive.

The primary goal for these developers is to maintain the ability to train and operate sophisticated AI models. If the current loophole—where hardware stays abroad but the processing power is accessed remotely—is closed, these companies could face significant hurdles. The uncertainty surrounding US technology policy is already forcing these commercial entities to make preemptive decisions about their infrastructure.

Regulatory gaps and RASA

A significant focus of this regulatory tension is the proposed Remote Access Security Act, commonly referred to as RASA. This legislation would grant the US Commerce Department the authority to regulate remote, cloud-based access to restricted technologies. While the House of Representatives passed the bill in January, it is currently awaiting action in the Senate.

Chinese companies are particularly concerned that RASA will close the gap between physical exports and digital access. Currently, Washington restricts the direct sale of Nvidia’s most advanced AI chips to China, but it has not fully addressed the use of those chips via third-party cloud services. If RASA is enacted, it could effectively block Chinese developers from using high-end processors regardless of where the hardware is physically located.

The broader objective for the United States is to preserve its technological lead in advanced semiconductors. By limiting China's access to computing resources, Washington hopes to curb the development of AI systems with potential military or strategic applications. This shift suggests that export controls are evolving to monitor who can access computing power and where that access takes place.

The Jensen Huang controversy

Recent reports involving Nvidia CEO Jensen Huang have added a layer of political pressure to the debate over export controls. Reports surfaced regarding Huang and Alice Huang, a former executive at a Singapore-based cloud-computing company called Megaspeed. Huang admitted to being unfaithful to his wife, and the New York Times reported that authorities concluded Alice Huang was a Chinese intelligence asset.

Alice Huang reportedly had business ties and investments with major Chinese technology companies, including Alibaba and Tencent. These connections have drawn scrutiny because Megaspeed has faced suspicions of helping Chinese customers obtain Nvidia processors that are subject to US export controls. While the company denies illegally diverting chips, the situation has highlighted the complexities of the international supply chain.

The controversy has renewed attention on how advanced American technology reaches Chinese customers. It has provided a politically sensitive dimension to the ongoing discussion about technology leaks. For policymakers, these allegations may provide additional support for stricter enforcement of existing rules and the passage of new legislation like RASA.

Grey markets and enforcement

Beyond the cloud, there is a growing concern regarding the grey market for restricted Nvidia AI processors. Chinese traders and intermediaries have reportedly begun stockpiling these chips in anticipation of a potential crackdown. Buyers are worried that the US will intensify efforts to stop advanced chips and AI servers from entering China through unofficial channels.

Maintaining these restrictions across international supply chains remains a difficult task for officials. For example, procurement records showed that a Chinese university with ties to the People’s Liberation Army bought Nvidia H200 AI chips last month, even though they were on a US export blacklist. This case shows how restricted technology can still reach Chinese institutions through procurement channels that are difficult for authorities to monitor.

Another example involves a case where US prosecutors charged the co-founder of Super Micro Computer with conspiring to divert servers to China. Prosecutors alleged the company purchased approximately $2.5 billion worth of servers between 2024 and 2025, with a substantial portion diverted to China. These incidents highlight the difficulty of controlling the international networks through which advanced computing equipment is redirected.

The future of AI access

The immediate next step for the industry will be to see if the US Senate advances the Remote Access Security Act. The outcome will determine whether Washington introduces new measures governing overseas cloud access to restricted processors. For Chinese AI firms, the current uncertainty is already creating pressure to secure capacity before new rules take effect.

Overseas data centers, especially those in Southeast Asia, are expected to become increasingly important for Chinese computing strategies. The US faces a difficult task in ensuring restricted technology does not reach Chinese users while managing the complexities of global cloud computing. This requires a delicate balance between national security and the global nature of cloud computing.

The dispute highlights a broader evolution in the US-China technology competition. Export controls are no longer just about where a chip can be physically sold. They are increasingly about deciding who can access computing power, where that access happens, and how that access is monitored. This could have consequences for the entire global AI industry and international cloud providers.