Deep in the Yangtze River basin, one of China’s most prosperous economic regions, the country is betting big on rice fields scattered across a 760-acre island, located in the municipality of Wuhu. According to the Financial Times, this agricultural land will soon host some of Beijing's biggest servers dedicated to serving wealthy areas nearby, which will represent one part of the plan to build the "Stargate of China," as described by an executive tied to one of the project's suppliers.
Currently, China only accounts for about 15% of the global AI compute power, far less than the ~75% held by the United States, according to a study by Epoch AI. In an effort to better compete with America, China is taking a different, measured approach that will see it consolidating already-existing but dispersed data centers into one unified network... to be linked together using Huawei's UB-Mesh tech.
On the other hand, the aforementioned farmland will be converted into a "data island" housing massive data centers for four companies: Huawei, China Mobile, China Telecom, and China Unicom. These servers are strategically placed near major cities like Shanghai, Hangzhou, and Nanjing — all of which will benefit from this upgrade, whose goal is to speed up inference for people living in the area. Think faster responses in agentic AI applications, for instance.
The Wuhu project is worth a reported $37 billion on its own — significantly less than the $500 billion Stargate project that serves as its aspiration — and while Wuhu's "mega-cluster" is the focal point, there's actually a lot more in motion. China has designated Ulanqab in Inner Mongolia to supply Beijing and Tianjin, Guizhou in the south to feed Guangzhou, and Qingyang in Gansu to serve Chengdu and Chongqing; these new data centers are intentionally in close proximity to large civilian populations, while existing data centers in remote areas are being rewired to focus on training LLMs.
That narrative further extends to AI chips sitting idle in faraway facilities located in remote but energy-dense provinces. From 2022, China catalyzed the construction of server farms in said regions, but due to low demand in the vicinity, a lot of hardware ultimately was never used, and just sat underutilized while demand was soaring elsewhere (because local governments were reluctant to give them up). Now, the plan is to link these data centers using the aforementioned UB-Mesh tech to serve as redundancy, and to sell surplus compute.
Interestingly, a stockpile like that isn't worrying Beijing anymore, as Wuhu is offering subsidies that cover up to 30% of the cost of procurement for these AI chips. Clearly, China wants to transform its underdeveloped terrain fast. Especially when the country is not allowed to receive most high-end AI GPUs made by Nvidia, they rely on local options that can't match capacity, which is part of the reason smuggling these accelerators is a huge prospect — even if Nvidia is not worried, given the practicality.
At the moment, China and the U.S. are locked in a ruthless battle for chip dominance, with the former playing catch-up as the latter works to limit any counter. Export restrictions have forced Beijing to develop homegrown AI stacks, which can ironically threaten U.S. leadership more by reducing foreign reliance. Coupled with these ambitions to massively scale AI compute near both urban centers and rural areas, China is emerging as a competitor that won’t back down.
Follow Tom's Hardware on Google News, or add us as a preferred source, to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button!
