AI guzzles power – how does China feed it cheap, reliable green power
AI耗电如何用便宜可靠的绿电喂饱:算电协同首次写入政府工作报告与十五五规划纲要,乌兰察布300 MW风光+45 MW储能直供算力
As hyperscale AI data centers expand worldwide, CGTN on November 17 published an article by Meng Yao, deputy director of the Energy Governance Center at Tsinghua University's Energy Internet Research Institute, exploring how China is tackling the challenge of powering AI with affordable, reliable green electricity. A single hyperscale AI data center can draw a load comparable to that of a medium-sized city; amid surging demand, Denmark has paused new grid connection approvals and New York State has imposed a moratorium on permits for large new data centers. In 2026, China included computing-electricity coordination in its Government Work Report for the first time and wrote it into the Outline of the 15th Five-Year Plan (2026-2030), making integrated development of computing networks, next-generation power grids and advanced communications networks a core strategy. China faces three structural mismatches: over 70% of its wind, solar and hydropower resources lie in the less-developed north and west, while roughly 60% of computing demand concentrates in eastern urban clusters; wind and solar output fluctuates sharply while AI infrastructure requires round-the-clock operation; and computing capacity can be built in months while energy infrastructure takes years. Responses include the East Data, West Computing initiative, direct green-power connections and integrated generation-grid-load-storage systems. In Ulanqab, Inner Mongolia, dedicated lines link renewable farms directly to computing clusters with storage to smooth output; Centrin Data's zero-carbon computing base runs on 300 MW of wind and solar capacity with 45 MW of storage, training at full load when renewables are plentiful and drawing on storage when they fade. Shanghai Lingang completed China's first cross-provincial migration test for a 100-card AI cluster, moving workloads between nodes in under three minutes during grid peaks, and the Xinjiang Computing for Chongqing initiative runs Changan Automobile's autonomous-driving model training during green-power surpluses in Xinjiang.