How China gets the Nvidia chips it needs to develop AI
Chinese AI developers are reportedly accessing high-end Nvidia graphics processing units (GPUs) despite US export bans on top-tier Blackwell and Rubin chips. This access primarily occurs through renting computing power at data centers in third nations, often in Southeast Asia, or, less substantiated, through smuggling. Washington cites security concerns over advanced AI's military applications.
Despite US export controls on advanced Nvidia GPUs, Chinese AI developers are reportedly circumventing restrictions by renting computing power from data centers located in third countries, particularly in Southeast Asia. This method leverages a loophole in US policy that restricts where a chip is shipped, but not who ultimately uses its processing capabilities. For instance, Nvidia GPUs can be situated in regions like Malaysia's Johor, allowing customers in Shenzhen to log in remotely, a practice described as compliant with current regulations.
This approach has been observed in practice, with a White House adviser stating that Chinese startup Moonshot AI accessed high-end Nvidia chips through facilities in Thailand. While less confirmed, reports have also suggested instances of smuggling, with one tech outlet claiming DeepSeek developed a new model using thousands of chips dismantled overseas and brought into China. However, Nvidia has reportedly expressed skepticism about such claims, stating it had not seen evidence and found the idea "far-fetched."
The ongoing access to Nvidia chips remains crucial for top-level AI development, as they enable faster and easier training. However, the costs associated with these access methods, particularly for training on Chinese-made alternatives like those from Huawei, often involve more time and engineering effort. As China's domestic chip industry gains momentum, with Huawei planning an earlier release for its next-generation AI chip, Chinese AI companies are increasingly exploring home-grown solutions, potentially involving more extensive use of these improving local alternatives.
Share this article
Related reading
6 stories
How Nvidia CEO’s troubles, US midterms may tighten China chip curbs: ‘adding fuel to fire’
We previously explored how Nvidia CEO's challenges and US politics could tighten China chip curbs.

Anthropic's IPO bank closed position meant to stabilize stock, removing source of support
Apple develops Mac AI agent, plans third-party control safeguards - TechGig

Gemini 4 Neon Costs More Than Carbon Until December 1. Here’s Which Gemini Model You Should Actually Use

How to deploy an app you built with AI to your own server

