Skip to content
Where does this figure come from?Water20 October 2026

"An email written by an AI is a bottle of water"

Guillaume PakulaClément ReynaudSébastien PierfedericiAlexandre Veilly By Projet Celsius
Published on 20 October 2026 · 3 min read
A Google data centre: a long beige building fronted by a row of cooling towers, in front of hills

The comparison comes from a paywalled Washington Post article (September 2024), produced with researchers: we could not check its figure, so we do not use it. The researchers' method is published, and providers have since measured the water used by a question.

Contents
1.

Where the figure comes from

In September 2024, the Washington Post published, with Shaolei Ren's team at the University of California, Riverside, an estimate of the water used by an email written by AI: "a bottle of water per email", says its headline. The article is paywalled and we were unable to check it: its figure is not used here. The team's method is published: its 2023 study gave 500 mL for 10 to 50 GPT-3 replies, or 10 to 50 mL per reply. It counts the water evaporated to cool the servers and the water consumed by the power stations that produce their electricity.

2.

Providers' measurements

In August 2025, Google published a production measurement, with its method: 0.26 mL for a median question to Gemini, the water evaporated on its sites only. OpenAI gives 0.32 mL for ChatGPT, with no method. Mistral AI, in its life cycle assessment, counts 45 mL for a reply of 400 tokens, about half a page, adding the water used for electricity and for chip manufacturing.

AI Impact, by Projet CelsiusWater

Water for one question, from 0.26 to 50 mL depending on scope

Water attributed to a written question to an AI, in millilitres, logarithmic scale · 2023 to 2025

Google, Geminion-site cooling, measured0.26mL
OpenAI, ChatGPTdeclaration, no methodology0.32mL
UC Riverside, GPT-3cooling and electricity10 to 50mL
Mistral AI, Le Chatfull life cycle45mL
↑a factor of 170with Google, by scope
0.1110100 mL

Source: Google; OpenAI; UC Riverside; Mistral AI · projetcelsius.com/en/ai · CC BY 4.0

The 4 figures in the chart are consistent once their scope is set out: the lowest counts only the water evaporated on sites, for a question to a recent model; the highest add the water used for electricity, and for Mistral AI the water used for server manufacturing.

3.

Withdrawing or consuming

Company reports distinguish water withdrawn, which is pumped and partly returned, from water consumed, which does not return to the environment. In its 2025 financial year, Microsoft withdrew 13.3 million m³ and consumed 8.2; half of its withdrawals take place in water-stressed areas. 1 m³ evaporated in Arizona, under water stress, does not count the same as 1 m³ evaporated in Ireland.

AI Impact, by Projet CelsiusWater

Microsoft's water use, 2020-2025

Water withdrawn and water consumed, in million m³ per financial year · Financial years 2020, 2024 and 2025

051015million m³2020202120222023202420256.793.99Withdrawn13.3Consumed8.2

Source: Microsoft, 2026 data sheet · projetcelsius.com/en/ai · CC BY 4.0

4.

Key points

At the scale of one person, the water used by a question weighs very little: 0.26 to 45 mL depending on the scope, whereas a French person uses 143 litres of drinking water a day at home, largely returned to the environment. Globally, data centres consumed about 560 billion litres in 2023 according to the IEA, counting cooling, electricity and chip manufacturing, and about 1,200 billion in 2030 in its base case. In our view, water has to be assessed site by site, within its river basin.

Water for one question to Gemini0.26mL

Median · written question · Google measurement, August 2025

Water for one question to Gemini→
Point to note

The water stress map, in "AI in maps", shows the countries where an additional withdrawal matters most.

Guillaume PakulaClément ReynaudSébastien PierfedericiAlexandre Veilly
By Projet CelsiusAI Impact, Projet Celsius · 20 October 2026
Write to us →

Sources

8 documents

Every figure was read in its source; the hub's methodology is described in Methodology and sources.

Cite this analysis

AI Impact, Projet Celsius, "An email written by an AI is a bottle of water", 20 October 2026. https://projetcelsius.com/en/ai/analyses/email-equals-bottle-of-water/

Charts and maps

ChartWaterWater for one question, from 0.26 to 50 mL depending on scope× 170from Google, 0.26 mL, to Mistral AI, 45 mLChartWaterMicrosoft's water use, 2020-202550%of withdrawals in financial year 2025 in water-stressed areasMapWaterWater stress, country by country17.6%world average in 2023Key figureWaterWater for one question to GeminiMedian · written question · Google measurement, August 2025

A question about a figure?

An error, a source to add, a methodology question, a press request: write to us. Guillaume Pakula, Clément Reynaud, Sébastien Pierfederici and Alexandre Veilly will reply.

Write to us →

Projet Celsius, which runs this hub, also helps companies measure and reduce the footprint of their AI use.

Accounting for AI in your Bilan Carbone® ↗
Projet Celsius, environmental strategy consultancy

We support companies that account for AI in their Bilan Carbone®

Visiter Projet Celsius ↗