What the evidence says about AI and climate
AI-focused data centres used an estimated 0.5% of the world's electricity in 2025, around a third of all data-centre consumption. Treating the whole sector's demand as AI demand overstates its footprint. Demand is growing quickly, though, and cheaper, more efficient queries can encourage heavier use, eating into those savings. The climate applications deserve the same attention. ECMWF's operational AI ensemble improves several measures of forecast accuracy while using roughly one-thousandth as much energy to generate forecasts as its physics-based system. It still relies on physics-based modelling for its starting conditions. The IEA also identifies ways AI can reduce industrial energy waste and help grids use existing capacity more efficiently.AI-focused data centres used an estimated 0.5% of the world's electricity in 2025, around a third of all data-centre consumption. Treating the whole sector's demand as AI demand overstates its footprint. Demand is growing quickly, though, and cheaper, more efficient queries can encourage heavier use, eating into those savings. The climate applications deserve the same attention. ECMWF's operational AI ensemble improves several measures of forecast accuracy while using roughly one-thousandth as much energy to generate forecasts as its physics-based system. It still relies on physics-based modelling for its starting conditions. The IEA also identifies ways AI can reduce industrial energy waste and help grids use existing capacity more efficiently.Expert summary pending. Showing the newcomer version.