For four years China has tried to solve the power problem of computing with geography. Its Eastern Data, Western Compute programme, launched nationwide in 2022, designated eight national computing hubs and ten data centre clusters, steering power-hungry workloads to the western provinces where land and renewable energy are abundant. Latency-sensitive work stayed near users in the east.
Artificial intelligence has strained that model, and Beijing's new AI+ energy action plan adds a second tool: price. As an analysis by the China Policy team published on 30 July puts it, the plan "asks data centres to choose, and to pay for their choice."
How fast demand is growing
The numbers explain the shift. China's computing facilities used 166 billion kWh of electricity in 2024, about 1.7 per cent of national consumption, according to the Ministry of Industry and Information Technology. The ministry said their power use has grown by more than 10 per cent a year since the start of the 14th Five-Year Plan period and is expected to keep rising quickly as AI computing demand surges.
The pace has since accelerated. The China Policy analysis says national computing centre consumption reached 170 TWh in 2025, 1.6 per cent of national demand, growing at nearly 40 per cent a year. Wang Hongzhi of the National Energy Administration expects 800 TWh, around 6 per cent of demand, by 2030, which the analysis notes is double the middle scenario that the China Academy of Information and Communications Technology modelled only a year earlier.
Estimates vary widely. Carbon Brief has noted that the International Energy Agency put Chinese data centre demand at about 100 TWh in 2024, while other forecasts reach 600 TWh by 2030. It also cited the IEA's view that data centres accounted for just 3 per cent of new Chinese demand since 2022, with industry and the electrification of heat and transport doing far more to drive growth. Han Xue of the Development Research Centre of the State Council has said data centres could account for 1 per cent of China's total CO2 emissions by the end of 2025.
What the plan asks
The China Policy team describes three fronts. First, demand is to become responsive rather than fixed. Electricity market prices are to guide data centre energy management and the scheduling of workloads across regions and grids, and new facilities are pushed towards multi-year green electricity contracts with renewable generators. Centres that can shift work get cheaper, greener power; those that cannot must pay for steady supply.
The analysis points to an example already operating. At Zhongjin Data's zero-carbon computing base in Ulanqab, Inner Mongolia, AI training shifts to midday when solar output peaks and data backup moves to night when wind is strong, releasing more than 30 per cent of capacity as flexible load.
Second, infrastructure is to be planned jointly. Large renewable energy bases and national computing hubs are to be coordinated, with facilities and backbone network access points concentrated in regions whose energy, water and land can support them. Pilots will pair AI computing facilities of a million kilowatts or more with dedicated energy systems.
Third, AI is to be used on the grid itself. The plan's first 51 "AI+ energy" scenarios include renewable forecasting, grid dispatch, virtual power plant coordination, storage management and operation of direct connections.
The green mandate underneath
The plan builds on rules already in force. A special action plan for green, low-carbon data centres issued by the National Development and Reform Commission in July 2024 set a goal for new data centres at national hub nodes to source more than 80 per cent of their electricity from green power by the end of 2025. It also targeted an average PUE below 1.5 nationally by the end of 2025, a utilisation rate of at least 60 per cent and annual growth of 10 per cent in renewable energy use, and told new large data centres to locate within the national hub clusters where possible. It supported moving non-real-time computing to the western hubs. For 2030, it set goals for national average PUE and the energy and carbon efficiency of computing to reach advanced international levels, and for new large data centres in northern heating regions to make much greater use of their waste heat.
In July 2025 the NDRC turned that into a formal consumption obligation, adding a green electricity consumption ratio for new data centres at national hub nodes to the list of key energy-using industries, alongside steel, cement and polysilicon.
Results so far are mixed but improving. MIIT said facilities named as national green computing facilities for 2025 achieved an average renewable energy utilisation rate above 70 per cent through green power purchases, green certificates or their own solar plants. One China Telecom data centre in Inner Mongolia reached 89.4 per cent by buying green power in the market and working with renewable generators on direct supply.
Why it matters
China is building a market for flexible data centre demand in a power sector that has been administratively priced for most of its history, the China Policy analysis notes. Time-of-use pricing and demand response for heavy industry are long established in the United States and Britain. What is new is applying them to AI computing at scale, under a green mandate those earlier schemes never had to meet.
The approach also shows how China sees the trade-off that every grid faces. Data centres want power that never fails, is clean and is cheap. The plan does not promise all three. It prices each against the others: latency against low-cost power through siting, greenness against firmness through contracts and scheduling.
What to watch
The analysis notes that the plan has no price signals yet. The detail will come in provincial tariff rules and market designs that set what firm supply costs relative to flexible, green power. The pilots pairing large AI facilities with dedicated energy systems will show whether the model can scale beyond showcase sites like Ulanqab. And the 2030 demand forecast will be tested against actual growth: if consumption does approach 800 TWh, even a strongly renewable grid will need to manage computing as one of its largest loads.