AI's Energy and Water Appetite

How the data centres behind AI use growing amounts of electricity and water, the effects on power grids and climate goals, and how companies are responding.

AI runs in data centres: large buildings full of servers that need electricity to run and cooling to prevent overheating. As AI use has grown, so has concern about its energy and water needs.

Electricity

The International Energy Agency estimated in 2025 that data centres used around 1.5 percent of global electricity in 2024, and projected that their electricity use could roughly double by 2030, driven largely by AI. In some places, such as parts of the United States and Ireland, data centres already consume a significant share of local electricity.

Effects on power systems

Grid strain: new data centres can require as much power as a small city, sometimes faster than grids can expand.
Power prices: rising demand can push up electricity prices in some regions.
Energy sources: technology companies have signed deals for renewable energy and nuclear power. In 2024, Microsoft agreed to buy power from a planned restart of a reactor at the Three Mile Island nuclear plant in the United States.

Emissions

If new demand is met by fossil fuels, AI could increase emissions. Some large technology firms have reported rising emissions linked to data centre growth, making their climate targets harder to meet.

Water

Many data centres use water for cooling. In dry regions, this can compete with local needs. Companies are exploring cooling methods that use less water.

Efficiency gains

Efficiency is improving rapidly: newer chips perform more calculations per unit of energy, and models are becoming more efficient. Some economists point to the Jevons paradox: when technology becomes more efficient, cheaper use can lead to more total use, so total energy demand may still rise.

A data centre comes to town

A technology company plans a large data centre in a region with cheap land and power. It brings construction jobs, tax revenue and some permanent jobs. But the local utility must build new power plants and transmission lines, and residents worry about higher electricity bills and water use. Local governments must weigh these costs and benefits.

Thinking AI uses only a trivial amount of energy

Individual AI queries use modest energy, but billions of queries and the training of large models add up. Data centre electricity use is growing fast and matters for grids and climate goals.

Key takeaways

Data centres used around 1.5 percent of global electricity in 2024, according to the IEA.
Their electricity use could roughly double by 2030, driven largely by AI.
AI growth strains grids, can raise emissions and uses water for cooling.
Efficiency is improving, but the Jevons paradox means total demand may still rise.