# How do I reduce the water footprint of AI?

> Cut AI's water use by using less compute, choosing hosts and regions with low water stress and good WUE, and timing heavy work where it helps.

Source: https://greenai.guide/how-do-i/reduce-the-water-footprint-of-ai/  
Updated: 2026-11-19

## In short

Water use follows energy use, so the first step is the same: use smaller models, shorter outputs and fewer regenerations. Beyond that, choose hosts that publish water use and avoid water-stressed regions for heavy work, and ask suppliers for site-level water figures. Per text prompt the water is small; totals and local effects are what matter.

## Steps

1. Cut the computing first with smaller models, shorter outputs and less regeneration.
2. Ask hosts for site-level water use and WUE, and how the water is sourced.
3. Avoid water-stressed regions for heavy, flexible work such as training.
4. Prefer hosts that report water use publicly or to a regulator.
5. Include water in AI procurement questions and in your own reporting.

**Less computing means less water, and where the computing happens decides how much that water matters.** Data
centres use water mainly for cooling, and power stations use it to make the electricity.

## How much water are we talking about?

Per text prompt, very little: Google reports 0.26 mL for a median Gemini text prompt. In total, and in some places,
a lot: Li, Yang, Islam and Ren estimated that training GPT-3 in Microsoft's US data centres could have directly
evaporated about 700,000 litres, and projected that global AI demand could account for 4.2 to 6.6 billion cubic
metres of water withdrawal in 2027. Estimating your own share is covered in
[estimating the water footprint of AI](/how-do-i/estimate-the-water-footprint-of-ai/).

## What actually reduces it?

- **Less compute.** UNESCO and University College London found small task-specific models can cut energy by up to
  90%, and shorter prompts and responses by more than half; water falls with energy.
- **Location.** A litre drawn in a drought-hit area has more effect than one where water is plentiful, so move
  flexible heavy work away from water-stressed sites where you can.
- **Supplier evidence.** In the EU, operators of data centres with at least 500 kW of IT power demand must report
  energy and water use under Delegated Regulation 2024/1364. Ask for those figures.

## Common mistakes

- **Chasing the lowest WUE alone.** Cooling that saves water can use more electricity; read
  [PUE and WUE](/how-do-i/understand-pue-and-wue/) together.
- **Worrying about short questions** while generating images and video freely.
- **Forgetting the power station's water,** which on-site figures leave out.

See [hosting and data centres](/practice/hosting-and-data-centres/).

## Sources

- [Making AI less 'thirsty': uncovering and addressing the secret water footprint of AI models](https://arxiv.org/abs/2304.03271), Li, Yang, Islam and Ren (2023)
- [Measuring the environmental impact of delivering AI at Google scale](https://arxiv.org/abs/2508.15734), Elsworth et al., Google (2025)
- [Commission Delegated Regulation (EU) 2024/1364 on the first phase of the establishment of a common Union rating scheme for data centres](https://eur-lex.europa.eu/eli/reg_del/2024/1364/oj), European Commission (2024)
- [AI large language models: new report shows small changes can reduce energy use by 90%](https://www.unesco.org/en/articles/ai-large-language-models-new-report-shows-small-changes-can-reduce-energy-use-90), UNESCO (2025)
