One kilowatt‑hour of electricity can run an air‑conditioner for one hour, yet it is insufficient for AI to generate a 30‑second high‑definition video. Behind a 30‑second AI video lies enormous power consumption of 200 000 kWh per day from a ten‑thousand‑GPU cluster. The future of artificial intelligence is not only a race in computing capability, but also a contest for power supply. Why have wind turbines in Inner Mongolia become “power banks” for AI? To understand Ulanqab’s “power sponge” is to grasp the shape of the future.

In 2026, “power‑computing synergy” was written into the Government Work Report for the first time and incorporated into the Outline of the 15th Five‑Year Plan. Since July, the 15th Five‑Year Plan for the Development of the New‑type Power System and the 15th Five‑Year Plan for Renewable Energy Development have been successively issued. The eight‑character guideline — “Empower computing with power, boost power with computing” — sets the tone for China’s energy foundation for AI over the next five years. More than a partnership between energy and technology, it constitutes a foundational project underpinning every swipe and scan in daily life.

The 15th Five‑Year Plan for the Development of the New‑type Power System proposes that by 2030, a safe, reliable, green‑low‑carbon, resilient, smart and flexible new‑type power grid will take initial shape to foster coordinated development of power and computing power. Advancing power‑computing synergy from the power‑system side centers on “supply guarantee plus coordination”. On one hand, power supply capacity shall be enhanced to satisfy high‑reliability power demands of various computing‑power facilities. On the other hand, energy‑resource allocation and computing‑power facility construction shall be integrated, with coordinated planning for computing‑power and power projects.

The 15th Five‑Year Plan for Renewable Energy Development specifies that by 2030, the total installed capacity of wind and solar power will exceed 2.8 TW, accounting for over 50 percent of total installed capacity. It focuses on “scaling up green energy plus consumption accommodation”. New data centers shall be planned and deployed in coordination with renewable‑energy power generation. The temporal‑spatial flexible adjustment potential of computing‑power loads shall be tapped by category. Innovative models such as direct green‑power connection and source‑grid‑load‑storage integration will be applied to steadily raise green‑power consumption of digital infrastructure. Integrated development of offshore wind power and undersea computing power is also encouraged.

When you send an AI prompt on your mobile phone, thousands upon thousands of servers in far‑off data centers spring into action.

In the AI era, artificial intelligence is not merely cloud‑hosted code but a genuine power‑hungry consumer. A ten‑thousand‑GPU cluster consumes at least 200 000 kWh daily.

Hou Wenjie, Director of the Statistics & Digital‑Intelligence Department at the China Electricity Council, stated that as AI evolves from training‑oriented toward balanced training‑and‑inference workloads, Token call‑volumes keep rising exponentially, driving sustained rapid growth in associated power consumption.

According to Tan Hongjiang, Deputy Director‑General of the Electric Power Department of the National Energy Administration, China’s annual incremental power consumption for computing power is expected to exceed 100 billion kWh during the 15th Five‑Year Plan period, hitting 800 billion kWh by 2030, equivalent to roughly 6 percent of total national electricity consumption.

While computing power surges forward, green power provides critical support.

By the end of June, China’s installed non‑fossil‑fuel power generation capacity reached 2.52 TW, making up 62.4 percent of total installed capacity. Virtually all new power demand in China is met by green electricity.

Within the eight national hubs of the East‑Data‑West‑Computing initiative, large‑scale green computing clusters keep lifting their green‑power utilization ratios.

Newly‑built data centers in Horinger (Inner Mongolia) and Qingyang (Gansu) report green‑power consumption ratios above 80 percent.

A total of 172 national computing‑power centers have obtained the 4A green‑low‑carbon certification.

Ulanqab in Inner Mongolia is pioneering a new path where “computing power chases green power”.

Hosting the world’s largest single‑site intelligent computing center covering 120 000 square meters — equal to 20 standard football fields — the facility adopts direct green‑power connection. Wind‑driven computing generates tokens with a power supply capacity of 160 MW, sufficient to run 160 000 one‑horsepower air‑conditioners simultaneously.

Zheng Zihao, General Manager of Envision Group’s AI Computing Center, explained that each cabinet requires more than 20 kW of power, comparable to running 20 air‑conditioners at maximum cooling mode non‑stop year‑round. Multi‑level energy‑storage units, transformers and solid‑state transformers function like a sponge: absorbing surplus power during output surges and releasing stored energy when generation drops, thereby securing stable power supply.

With huge computing‑power centers acting as major green‑power buyers, upstream new‑energy development gains fresh momentum. An energy relay race converting watts into bits is gathering pace.

Ulanqab is also home to the world’s largest on‑shore single‑site wind‑power project with a total capacity of 6 GW. Construction proceeds on its second‑phase 1.2 GW wind‑farm, featuring main‑model 10‑MW wind turbines. Each turbine has 113‑meter‑long blades and generates nearly 20 kWh per rotation — enough for AI to produce roughly ten minutes of video.

Yang Huiwen, project construction lead, said full‑scale construction is underway. Five turbines have been erected, and full‑capacity grid‑connection is scheduled by year‑end. The adopted 10‑MW turbines deliver roughly 20 percent lower capital cost per kilowatt.

Jia Haoyu, Operation & Management Principal of SPIC Chahar New Energy Company, noted that computing‑power enterprises consume massive volumes of green power, while wind and solar output fluctuates with weather conditions. Source‑grid‑load‑storage and similar mechanisms can reconcile the two sides perfectly.

As a key node of the national East‑Data‑West‑Computing project, Ulanqab draws on abundant wind‑solar resources to attract computing‑power facilities. Green power accounts for 67 percent of its local power mix. Instead of passively consuming electricity, local data centers operate as flexible green virtual power plants: running heavier computing workloads when wind and solar output is high and drawing on energy storage under low‑generation conditions. Driven by East‑Data‑West‑Computing, 89 data‑center projects have settled in Ulanqab, over 90 percent of which are intelligent‑computing projects. The initiative has triggered trillion‑yuan‑level social investment and effectively mitigated challenges including high computing‑power energy consumption and new‑energy output volatility.

The green‑energy foundation is boosting the global competitiveness of China’s AI. By 2030, with nearly 6 trillion kWh of green electricity feeding into power grids, every tap on your screen will become lighter and greener.

(Reporters: Ping Fan, Zhang Hanxiao, An Yu)