- China turns farmland into data centers to compete with American AI dominance
- Wuhu’s $37 billion project highlights Beijing’s urgency in artificial intelligence
- Export restrictions leave China relying heavily on less powerful local chips
China’s ambitions in artificial intelligence have gained new visibility through its plan to develop a domestic alternative to the massive Project Stargate being pursued in the United States by OpenAI and Oracle.
While the American initiative is expected to support up to two million AI chips, Beijing is advancing its own version anchored by a $37 billion project in Wuhu.
Although far smaller than the $500 billion price tag linked to Stargate, the Chinese project is designed to consolidate existing computing capacity into a more centralized network.
The Wuhu project and its scale
The site selected for this project is in Wuhu, eastern China, and it covers former rice fields along a 760-acre island in the Yangtze River basin.
This land, once devoted to food production, is being converted into a “data island” for four of the country’s largest technology operators: Huawei, China Mobile, China Telecom, and China Unicom.
By situating the new “mega-cluster” of data centers near major cities such as Shanghai, Hangzhou, and Nanjing, planners hope to deliver faster inference services to dense urban populations.
Beginning in 2022, China encouraged the construction of server farms in interior provinces with cheap power supplies.
Yet these sites often sat idle, as local governments reallocated capacity to areas where demand was higher.
The new plan attempts to fix that by linking both urban and remote data centers through Huawei’s UB-Mesh technology.
This technology can provide redundancy while allowing unused compute power to be sold.
The Wuhu project’s subsidies, which reportedly cover as much as 30% of AI chip procurement costs, further reflect Beijing’s urgency to make the new clusters operational.
China currently holds about 15% of global AI compute power, far less than the United States’ estimated 75%.
Export restrictions have blocked access to advanced GPUs from Nvidia, leaving domestic suppliers unable to fully match foreign performance.
That gap has created incentives for smuggling hardware, although officials seem intent on developing self-sufficient AI stacks to reduce dependence on overseas sources.
The long-term aim is that such infrastructure will allow both companies and individuals to deploy more sophisticated AI tools.
Whether local chips can support this ambition remains uncertain compared to Western options powering major data centers abroad.
The conversion of farmland into server space raises questions about sustainability, resource allocation, and energy demand.
Supporters view the projects as vital for narrowing the technological divide, while skeptics point out the costs of diverting agricultural land and the uncertainty of relying on less powerful local chips.
Via Toms Hardware
