AI and electricity: how much power do data centres need?

Where AI actually runs

When you ask a chatbot a question, the work happens in a data centre: a building full of servers, storage and networking equipment. See What is a server? Training large AI models, and answering millions of questions every hour, needs powerful chips that use a lot of electricity and produce heat that must be removed by cooling systems.

What the International Energy Agency projects

In its 2025 report Energy and AI, the International Energy Agency (IEA) projected that:

• Global electricity demand from data centres will more than double by 2030, to around 945 terawatt-hours a year.
• Demand from data centres built specifically for AI will more than quadruple over the same period.
• In the United States, data centres are on course to account for almost half of the growth in electricity demand up to 2030.

A terawatt-hour is a billion kilowatt-hours. To put the projection in context, it is a significant but still modest share of total global electricity use, and growth from electric vehicles, air conditioning and industry is also large.

Why it matters

• Local pressure: data centres cluster in certain areas, where they can strain local grids and compete for new connections.
• Climate: the impact depends on how the electricity is generated. Some operators are signing deals for renewable and nuclear power; others still rely partly on fossil fuels.
• Water: some cooling systems use large amounts of water, which matters in dry regions.
• Prices and planning: energy companies and governments are planning new power stations and grid upgrades partly around data centre demand.

The other side of the ledger

AI can also help reduce energy use: optimising power grids, improving building efficiency, speeding up research into batteries and materials, and cutting waste in industry. The IEA notes these benefits could be significant if they are widely adopted. Chips and software are also becoming more efficient, so each AI task tends to use less energy over time, even as total use rises.

What you can do

• Use AI where it genuinely helps, rather than for everything by default.
• Prefer smaller or on-device AI features for simple tasks; see AI PCs explained.
• Keep devices longer and repair them; manufacturing devices also uses a lot of energy.
• For organisations: ask cloud and AI providers about their energy sources and efficiency.

Tip: Be cautious with viral claims about how much energy a single chatbot question uses. Estimates vary widely and change quickly as technology improves.

Want the technical version? Read our note on green IT.

Why estimates differ so much

Different studies measure different things: some count only AI chips, others include cooling, networking and storage; some model today's efficiency, others assume future improvements. Projections also depend on how quickly companies build new data centres and whether AI demand keeps growing. That is why it is wise to compare several sources and look at ranges rather than a single headline figure.