Chinese companies have faced challenges in accessing US artificial intelligence chips and technologies due to trade restrictions. Despite efforts by Washington to limit Beijing’s AI advances, recent reports suggest that China is planning to establish massive data centers powered by over 115,000 high-end NVIDIA AI GPUs.
The ambitious project involves setting up a network of approximately 36 to 39 AI data centers in Western regions of China, such as Xinjiang and Qinghai. These data centers will be equipped with H100 and H200 Hopper GPUs, which are subject to US restrictions. The majority of these GPUs, estimated at around 70%, will be installed in a state-owned data center near Yiwu in Xinjiang Province. China has strategically chosen remote locations with abundant resources like wind, solar, and coal, as well as cooler climates, to support these large-scale computing infrastructures.
Several leading Chinese AI firms are involved in this initiative, with strong support from the Chinese government. The scale of this project is impressive, with the planned number of GPUs rivaling the compute capacity used by some of the most advanced AI models globally.
One question that arises is how China plans to acquire such a large quantity of restricted NVIDIA chips. While details are scarce, experts speculate that procurement may involve third-party suppliers or even black market channels. However, US officials have expressed doubts about the feasibility of obtaining chips on this scale through unorganized networks.
Despite uncertainties about sourcing methods, China’s determination to proceed with this infrastructure project indicates a strong belief in obtaining these crucial components. It also reflects China’s ongoing efforts to strengthen its domestic AI chip manufacturing capabilities. The advancement of AI in China is viewed as a national security threat by US officials, and the response from Washington remains to be seen.