Why AI Data Centers Use So Much Electricity and Water
AI data centers are on track to use as much power as Japan by 2030. Here is why the chips draw so much electricity, how much water cooling takes, and what it means for your bill.
By 2030 the world's data centers are on track to use about as much electricity as the entire country of Japan burns today. That is the headline figure from the International Energy Agency, and it reframes a question people usually ask with a shrug. Why does a building full of computers need that much power, and why does it also drink water? The answer is the same for both, and it comes down to one stubborn fact of physics: almost every watt that goes into a chip comes back out as heat.
The IEA's Energy and AI report projects that global data-center electricity demand will more than double by 2030 to around 945 terawatt-hours. Artificial intelligence is the engine. Demand from AI-optimized data centers is set to more than quadruple over the same period. This is not a rounding error in the energy system; it is a new, fast-growing customer showing up all at once.
Why do AI data centers use so much electricity?
An AI model is trained and run on racks of specialized processors doing trillions of calculations a second. Those chips are dense, they run flat out, and they convert electricity into computation and heat in roughly equal measure. Keeping them from cooking themselves takes yet more electricity for cooling. So the power bill has two halves: the computing itself, and the enormous effort of removing the heat that computing produces. Push both up with a wave of AI demand and the totals climb fast.
The scale is easier to feel in local terms. The IEA notes that a single large AI data center can draw as much electricity as 100,000 households. String enough of them together and you get the agency's blunt projection for the United States, where the numbers get genuinely strange.
"Global electricity demand from data centres is set to more than double over the next five years, consuming as much electricity by 2030 as the whole of Japan does today."
Fatih Birol, IEA Executive Director
How much water does an AI data center use?
Water is the cooling story. Many large facilities shed heat by evaporating water in cooling towers, and the volumes are not trivial: reporting on the sector describes big data centers using up to five million gallons of water a day. Researchers at the University of California, Riverside, estimated that training a single model, GPT-3, in Microsoft's US data centers could directly evaporate around 700,000 liters of clean freshwater. And that is only the water you can see. More is spent indirectly at the power plants that feed the servers, since most electricity generation uses water for steam and cooling too.
Will data centers raise my electricity bill?
This is where the abstract terawatt-hours land on a household. In the United States, the IEA expects data centers to account for almost half of the growth in electricity demand between now and 2030. The agency goes further: the US is projected to use more power in 2030 for processing data than for making all its energy-intensive goods combined, including aluminum, steel, cement and chemicals. When a region's biggest new source of demand is a cluster of server halls, utilities build new supply and new lines, and those costs have a way of reaching ratepayers. It is the same pressure already showing up as higher electric bills near data-center hubs and as the Energy Department moving to pull big campuses onto backup power when the grid runs hot.
Can data centers run on renewable energy?
Partly, and increasingly. The IEA expects a mix of sources to feed the boom, with renewables and natural gas taking the lead because they are cheap and available in the right places. The catch is timing. Solar and wind are intermittent, while an AI cluster wants steady power around the clock, which keeps gas, and in some plans nuclear, in the picture. The report also flags a quieter risk: cyberattacks on energy utilities have tripled in four years, partly sharpened by AI itself.
None of this makes AI's appetite a verdict, but it does make it a choice with a meter attached. Every new model and every faster answer runs on electricity and water drawn from the same grid and the same reservoirs that a summer heat dome is already straining. The technology feels weightless on the screen. The infrastructure behind it is anything but.
