Page language: Italiano English Français Deutsch

Why does artificial intelligence use so much water and energy?

By the FAIND newsroom · Updated on 17 October 2026

Diagram of a data center showing energy consumption on one side and the water footprint of cooling on the other

“Why does AI use water” and “why does AI pollute” are among the most searched questions. The answer lies in buildings almost nobody sees: data centres.

In brief

AI runs in data centres that use a lot of electricity.
Water is needed to cool the computers, directly or through power stations.
The impact depends on which energy powers the facilities.

Why artificial intelligence uses electricity

Every chatbot answer is the result of billions of calculations carried out by very powerful processors, GPUs, housed in data centres. Energy is needed at two stages: the training of the models, which lasts weeks, and everyday use, which consumes less per request but is repeated billions of times.

How much it consumes

According to the International Energy Agency, in 2024 data centres used about 1.5% of the world's electricity, and their consumption could roughly double by 2030, driven mainly by AI. How much a single request weighs depends on the model and on how it is measured: the figures in circulation should always be read with the source beside them.

Why AI uses water

Computers produce heat and have to be cooled. Many facilities use water-based systems, and part of that water evaporates. More water is consumed indirectly, by the power stations that generate the electricity. The amount varies widely with the local climate and the technology: this is why new data centres are often contested where water is scarce.

Does AI pollute?

It depends on the energy that powers it. A data centre connected to renewable sources has a very different impact from one running on gas or coal. The big companies are signing contracts for solar, wind and nuclear power, but they have also acknowledged that AI makes it harder to meet their own emissions targets.

How much water a question to ChatGPT or Gemini uses

Little, but not nothing. According to Google (August 2025), a typical text request to Gemini uses about 0.26 millilitres of water, five drops, and 0.24 watt-hours of electricity. According to Sam Altman (June 2025), an average ChatGPT query uses about 0.3 millilitres and 0.34 watt-hours, but OpenAI has not explained how it calculated this. A 2023 study by the University of California, Riverside estimated about half a litre for every 10-50 answers from the older GPT-3, also counting the water used by power stations. The numbers differ so much because the model, the data centre and what is counted all change. Does AI waste water? A single question does not; the problem is the sum of billions of questions, concentrated in a few places.

Where the water used by AI ends up

Not all of it is lost. Part of the water drawn by data centres goes back, warmer, to rivers and networks; the part that evaporates in cooling towers is truly consumed, and it matters most where water is scarce. That is why company reports distinguish between water withdrawn and water consumed: when you read a figure, check which of the two it is. Some new facilities use closed loops or air cooling, which use much less water but more electricity.

The role of AI in saving energy

The same technology is used to forecast solar and wind output, balance power grids and cut waste in buildings and factories. Whether the final balance will be positive has not yet been settled. FAIND follows the topic every day on the AI, climate and environment page.

Frequently asked questions

Why does artificial intelligence use water?

Because the computers in data centres produce heat and many facilities cool them with water. More water is used by the power stations that generate the electricity.

Does using ChatGPT pollute?

Each request uses little energy, but multiplied by billions of uses it becomes significant. The impact depends on the efficiency of the data centre and on the energy source.

How much water does a question to AI use?

According to the companies' own data, between 0.26 and 0.3 millilitres for a text request, a few drops. Older independent estimates went up to half a litre every 10-50 answers, including the water used by power stations.

Does AI waste water?

A single question does not: it uses a few drops. The problem is the sum of billions of requests, concentrated in a few data centres, sometimes in areas where water is scarce.