https://j.people.com.cn/n3/2026/0528/c95952-20461412.html
China to Establish 42 AI Computing Clusters of 10,000-Card Scale by 2025
People’s Daily Online (Japanese Edition) | May 28, 2026, 13:38
The *China “AI+” Energy Development Report 2026*—compiled under the leadership of the National Energy Administration—was recently released. This marks the first annual report dedicated to the integrated development of AI and energy within China’s energy sector, as reported by *People’s Daily*.
Driven by the rapid advancement of AI, global electricity demand for computing infrastructure continues to rise. According to projections by the International Energy Agency (IEA), global electricity consumption by data centers is expected to nearly double by 2030 compared to 2025 levels.
According to a report under the leadership of the National Energy Administration, by 2025, China had already established 42 AI computing clusters—each comprising a scale of 10,000 computing cards—and the total electricity consumption of computing centers nationwide reached 170 billion kilowatt-hours (kWh). Electricity consumption for computing purposes within the eight central hubs of the National Integrated Computing Network emerged as the primary driver of this growth; over the past three years, the average annual growth rate stood at approximately 39.5%, significantly outpacing the average growth rate of electricity consumption across society as a whole. Notably, the central hubs in the Beijing-Tianjin-Hebei region and Inner Mongolia recorded average electricity consumption growth rates of 33.3% and 66.5%, respectively, over the same three-year period. This trend demonstrates a rapid concentration of computing resources toward key central hubs and regions rich in energy resources.
Concurrently, for newly established data centers located within national computing central hubs, the target requirement of sourcing over 80% of their electricity from green energy sources is being rapidly implemented. This objective is being achieved through a diverse array of methods, including green electricity trading, direct green power supply connections, inter-regional trading, and the integrated management of power generation, power grids, electricity loads, and energy storage systems. Regarding the advancement of large language models (LLMs) for industrial applications, dozens of LLMs specifically tailored to the energy sector have already been deployed. Covering fields such as power grids, new energy sources, hydroelectric power, thermal power, nuclear power, coal, and oil and natural gas, the scenario-specific adaptability of these LLMs is continuously being enhanced.