The cleanest way to power a data centre is also the least predictable. The sun sets, the wind drops, and a server hall full of AI accelerators does not care; it wants the same steady draw at three in the morning that it wanted at noon. That simple mismatch, according to industry experts cited by Reuters this week, is now sitting squarely between China's climate ambitions and its computing ambitions, and it is proving far harder to solve than the official targets imply.
The targets themselves are not vague. Beijing wants renewables to supply roughly four-fifths of the AI data-centre sector's total power consumption by 2030, a steep climb from around 11 percent in 2023. The 2026 government work report named tighter integration between computing infrastructure and the power supply as a priority, and a green-data-centre action plan requires new projects in the national computing hubs to draw most of their electricity from clean sources. On paper, the direction is set. The scale gives it urgency: power demand from Chinese data centres is projected to rise by 300 to 500 billion kilowatt-hours between 2026 and 2030, close to a fifth of the country's entire growth in electricity demand, with the lower end of that range roughly equal to the United Kingdom's annual consumption.
The trouble is that the hurdles are practical rather than political. AI clusters turn out to be a poor match for renewable suppliers, worse than heavy users like aluminium smelters, because their load is so hard to forecast. "From what we understand, they cannot really adjust power consumption load much," said Pei Shanpeng, a director at State Power Investment Corp, at a Beijing conference. "GPUs are very expensive, so once they are purchased, operators want to use them as quickly and as intensively as possible." China is not short of clean electrons; it added hundreds of gigawatts of solar and wind in 2025 alone. The gap is not supply in aggregate but supply in the right place at the right moment, matched to a buyer with no tolerance for a flicker. When the choice is a clean megawatt that might not show up or a reliable one that will, operators pick reliability, which still means coal underneath.
A second obstacle is the grid companies themselves. The obvious workaround, wiring renewable generators directly to data centres and bypassing parts of the transmission network, is meeting resistance from operators who fear it will drain the electricity sales they rely on to recover their infrastructure investments. The other common fix, green electricity certificates, lets a company buy the environmental attribute of clean power without a physical line to any wind or solar farm. It satisfies the metric; whether it changes which electrons actually run the servers is a separate matter. Wang Zelin of the State Grid Jibei research institute offered a measure of the prize at stake: if just 15 percent of these loads could be made flexible, it would sharply ease the pressure to build new capacity over the next three to five years.
The tension is not unique to China. Grid operators everywhere are being asked to absorb AI demand faster than they can build for it, from Denmark pausing new connections to US utilities lining up $1.4 trillion in spending, and Europe has begun telling Big Tech to align its data centres with climate goals or stay away. What makes China worth watching is that its version of the bargain is more centrally directed and its targets are more aggressive, which is precisely why the distance between the number Beijing has written down and the grid that has to deliver it is so visible. For now the ambition is real and the mechanism is unfinished, and the unsolved part is not a shortage of panels or turbines. It is making intermittent power reliable enough for a machine that never sleeps.