China’s plan to increase renewable power use in AI data centres is facing challenges as unpredictable electricity demand makes green energy integration difficult.
China’s push to power its fast-growing AI data centre sector with renewable energy is running into challenges, as unpredictable peak power demand continues to strain grid planning and energy supply.
According to Reuters, ensuring reliable electricity for AI-focused data centres has become a strategic priority under China’s 2026 government roadmap, which aims to strengthen links between computing infrastructure and power networks.
A major goal is to increase renewable energy’s share in powering data centres to 80% by 2030, up sharply from 11% in 2023.
Power demand from China’s data centres is projected to rise by 300 billion to 500 billion kilowatt-hours between 2026 and 2030, accounting for nearly 18% of the country’s total electricity demand growth during the period, according to Pei Shanpeng of State Power Investment Corp.
However, industry experts say AI data centres remain difficult to match with renewable energy because their peak power usage is highly unpredictable.
“At least for now, they do not appear to be very flexible in managing power demand,” Pei said, according to Reuters.
He added that data centres cannot significantly adjust electricity usage because GPUs are expensive and operators want to run them as intensively as possible.
Experts also warned that direct green-power links to data centres could face resistance from grid operators, who fear lower electricity sales and higher infrastructure risks if demand weakens.
Wang Zelin of the State Grid Jibei Electric Power Research Institute said improving load flexibility could ease pressure on the grid. “If 15% of power consumption loads can be adjusted, it will significantly reduce capacity expansion pressure on the grid over the next three to five years,” he said.
The challenge highlights the growing tension between China’s clean energy goals and the rising power demands of AI infrastructure.
