---
title: "How much water does a ChatGPT query use?"
description: "Water per ChatGPT or Gemini query: 0.26 mL measured by Google for Gemini, 45 mL in Mistral's LCA. The IEA projects data centre water use at x2.1 by 2030."
canonical: https://projetcelsius.com/en/blog/ai-water-footprint/
author: "Guillaume Pakula"
datePublished: 2026-06-08
dateModified: 2026-11-11
language: en
translationOf: https://projetcelsius.com/blog/ia-empreinte-eau/
---

# How much water does a ChatGPT query use?

**Author**: Guillaume Pakula
**Published**: 2026-06-08
**Updated**: 2026-11-11
**Section**: Explained
**Canonical URL**: https://projetcelsius.com/en/blog/ai-water-footprint/
**French original**: https://projetcelsius.com/blog/ia-empreinte-eau/
**Source**: Projet Celsius (https://projetcelsius.com/en/), a climate strategy consultancy of engineers and PhDs based in Paris and Marseille.

> An AI query evaporates a few drops of water, not the "half a litre" often quoted: Google measured **0.26 mL** for a Gemini query and Sam Altman cites ~0.3 mL for ChatGPT. Both figures count only cooling; Mistral's life cycle assessment (LCA), which covers the full life cycle, reaches 45 mL per response. The issue becomes serious at the scale of the infrastructure: [worldwide data centre water consumption is projected to double by 2030, according to the International Energy Agency (IEA)](https://www.iea.org/reports/energy-and-ai).

## Key takeaways

- **A query uses a few drops of water**: 0.26 mL measured for Gemini, ~0.3 mL cited for ChatGPT.
- **Published figures range from 0.26 to 519 mL** because they do not count the same water.
- **"Half a litre per query" is a misleading shortcut**, taken from studies that counted total water or whole sessions.
- **Location decides**: the same query matters little in Sweden but a great deal in Arizona, where water is already scarce.

Searching Google for "ChatGPT water usage" quickly turns up a striking claim: each AI query supposedly "drinks" a bottle of water. The published figures are far lower. Servers are cooled with water, and the only 2 figures that AI providers have put forward for a single query come to a few drops: **0.26 mL** measured by Google for Gemini and ~0.3 mL cited by Sam Altman for ChatGPT.

These volumes cover only the **cooling water evaporated on site**; they exclude the water used to generate the servers' electricity and the manufacture of the hardware. Volume does not tell the whole story: the same query has a small footprint in Sweden and a large one in Arizona, where data centres draw on water basins that are already under strain.

This leaves open where the "half a litre" figure comes from, why published figures range from 0.26 to 519 mL, and at what scale AI's water use becomes a real issue. On the energy side, [a query uses around 0.3 Wh](https://projetcelsius.com/en/blog/chatgpt-query-energy-consumption/); this article is about water, which is quantified far less often.

## How much water does a ChatGPT or Gemini query use?

Only one figure comes from a measurement in production: Google puts [a median Gemini query (in French)](https://projetcelsius.com/ia/chiffres/eau-question-gemini/) at **0.26 mL of water**, about 5 drops, in its [August 2025 study](https://arxiv.org/abs/2508.15734). For ChatGPT there is no measurement: in a [June 2025 blog post](https://blog.samaltman.com/the-gentle-singularity), Sam Altman cited ~0.3 mL for an average query, with no stated method or scope. The 2 figures converge. Google's counts only the data centre's cooling water, not the water used to generate its electricity; Altman's does not say what it counts.

- A Gemini query: 0.26 mL, or 5 drops of water. This is the only figure that has been measured and published, by Google, and it leaves out the water tied to electricity.
- A ChatGPT query: ~0.3 mL according to Sam Altman, one fifteenth of a teaspoon, taken from a statement on his blog rather than an audited measurement.
- A teaspoon (5 mL): about twenty queries.
- A glass of water (25 cl): around 1,000 queries.

This figure varies widely with the model and the task, as energy use does: [the gap between models reaches a factor of 100](https://projetcelsius.com/en/blog/chatgpt-vs-claude-vs-gemini-environmental-impact/). The "half a litre per query" that circulates is a genuine figure, but misquoted: a 2023 study counted **500 mL for 10 to 50 responses** from GPT-3 ([UC Riverside](https://arxiv.org/abs/2304.03271)), not for a single one.

The same team then estimated **~519 mL for a 100-word email** generated by GPT-4, a figure that counts total water, electricity included ([Washington Post](https://www.washingtonpost.com/technology/2024/09/18/energy-ai-use-electricity-water-data-centers/)). Like [the "10x" claim in the comparison with a Google search](https://projetcelsius.com/en/blog/chatgpt-vs-google-search-energy/), the "half a litre" keeps circulating without the scope that gave it meaning.

## Why AI uses water, and when it becomes a problem

A few drops per query are not a problem in themselves. The difficulty lies in the [cooling mechanism](https://projetcelsius.com/en/ai/analyses/does-ai-really-use-water/), through which water is lost to evaporation, and above all in where your queries are processed.

> **Why AI "drinks" water** Servers heat up as they compute, and a data centre houses tens of thousands of them. To remove that heat, many sites circulate water that carries it away as it evaporates, much as sweating cools us down. Evaporated water is lost: it does not return to the river or to the groundwater.

### Data centres use water to cool their servers

What matters is the difference between water **withdrawn** and water **consumed**. A data centre withdraws water, returns part of it to the environment and evaporates the rest: Google states that **80% of the water withdrawn** by its sites is consumed, and therefore lost to the local basin. There is also a hidden water cost: the water used to **generate the electricity** that powers the servers. In some countries this indirect water use outweighs cooling, and it is precisely this water that Google's 0.26 mL leaves out.

### Everything depends on where your queries are processed

Location decides almost everything. A data centre in **Sweden** is cooled with cold air for much of the year, whereas a site in **Arizona** or the **Persian Gulf** depends far more on evaporation, in [regions that are already short of water](https://projetcelsius.com/en/ai/analyses/can-data-centres-dry-up-a-region/). With the same model and the same query, the footprint differs completely from one site to another, and you do not choose which site handles yours. Location matters in the same way [for carbon](https://projetcelsius.com/en/blog/chatgpt-carbon-footprint/): electricity in Sweden emits about 10 times less CO2 per kWh than the average in the United States.

## Which AI water figures can you trust, and how to read them

Only 3 AI providers have published a figure, and each counts something different. Google publishes a measurement made in production: 0.26 mL, cooling only. OpenAI relies on a statement by its chief executive in a blog post, with no method given. Mistral has published a full [life cycle assessment (LCA)](https://projetcelsius.com/en/blog/lca-vs-carbon-footprint/) with Carbone 4 and ADEME (the French Agency for Ecological Transition): **45 mL per response of 400 tokens** (tokens are the word fragments a model processes), server manufacturing included. Anthropic and Microsoft publish no figure per query. For Claude, our [estimates for Fable 5.2 and Opus 5.5](https://projetcelsius.com/en/blog/fable-5-2-opus-5-5-energy-water/) start from the electricity and water indicators of its hosting providers.

### Why published figures per query range from 0.26 to 519 mL

Set side by side, the published figures run from 0.26 to 519 mL, a **factor of 2,000**. None of them is wrong: the first counts the cooling water of a median query, the last the total water of a long output from GPT-4, the largest model in service at the time. In between, the standard academic estimate puts a GPT-3 response at **10 to 50 mL of total water**, and the [Mistral LCA](https://mistral.ai/news/our-contribution-to-a-global-environmental-standard-for-ai/) puts a response from Le Chat, Mistral's chatbot, at **45 mL**. So before quoting a water figure, you need to know what it counts.

### 3 checks before quoting an AI water figure

- Check the scope: cooling only, electricity included or full life cycle. The gap between these scopes ranges from a factor of 10 to a factor of 100.
- Be wary of fleet averages: an average indicator pools sites in Finland and Arizona, and so describes no real site.
- Distinguish water withdrawn from water consumed: Amazon Web Services (AWS) publishes 0.12 L/kWh of water withdrawn in 2025 (0.15 in 2024) and [Google 1.15 L/kWh of water consumed (in French)](https://projetcelsius.com/ia/base/unites/eau-du-refroidissement/), so the 2 indicators cannot be compared like for like.

> **What is WUE?** WUE (Water Usage Effectiveness) expresses the litres of water used per kWh of energy consumed by the IT equipment. It is the only indicator that operators publish. It stands at ~0.4 L/kWh of water consumed on site for the US data centre fleet in 2023 (Lawrence Berkeley National Laboratory, LBNL, for the US Department of Energy, 2024), 1.15 for Google's AI data centres (water consumed) and 0.12 for AWS (water withdrawn, 2025). It is always calculated for a whole fleet, never per product, so it serves as a benchmark and cannot be used to quantify your own use of ChatGPT.

In practice, this means reasoning in [orders of magnitude](https://projetcelsius.com/en/blog/company-carbon-footprint-main-emission-sources/) rather than in measurements. For an organisation, water is not handled in a Bilan Carbone® (the French carbon accounting method), which looks only at carbon, but in a **multi-criteria LCA**, with the range and the uncertainty stated openly. On the carbon side, AI use [fits into Scope 3 of your carbon footprint](https://projetcelsius.com/en/blog/ai-emissions-carbon-accounting-scope-3/).

> Published figures for AI's water use vary by a factor of 2,000 without any of them being wrong, because they do not count the same water.

## How much water does AI use per day and per year, at scale?

As an order of magnitude for a company, take 100 people who send 25 queries a day over 220 working days: that adds up to 550,000 queries a year. At 0.26 mL per query, this comes to **143 litres of water over the year**, the equivalent of a bathtub, or 1.4 litres per person. Even at the widest LCA scope (45 mL), the total reaches ~25 m³: real, but marginal in the footprint of a services company, [where equipment far outweighs online services](https://projetcelsius.com/en/blog/corporate-digital-carbon-footprint/).

The change of scale comes from the infrastructure. In 2021, an average Google data centre evaporated around 1.7 million litres a day, two thirds of an Olympic pool. In 2025, Google consumed **41 billion litres** of water across its data centres and offices, 34% more than in 2024 and more than double the 2021 level. Worldwide, [the International Energy Agency (IEA) puts water consumption linked to data centres at ~560 billion litres in 2023](https://www.iea.org/reports/energy-and-ai): 2 thirds to produce the energy they use, a quarter to cool them on site and the rest to manufacture chips. It projects 1,200 billion litres in 2030, 2.1 times as much within 7 years. The IEA attributes this increase to the rise of AI, which accounts for 5 to 15% of data centre electricity today and could account for 35 to 50% in 2030. The [report from United Nations University](https://news.un.org/fr/story/2026/06/1158934) (UNU-INWEH, the United Nations University Institute for Water, Environment and Health, June 2026) gives much higher volumes: 4,500 billion litres as early as 2025, enough to meet the water needs of more than 600 million people in sub-Saharan Africa, and 9,300 billion litres in 2030, equal to the annual needs of the region's 1.3 billion inhabitants. For 2030, the UN projection is nearly 8 times the IEA's, a sign of how uncertain these volumes are.

Volumes of this size matter little in Scandinavia and a great deal in Arizona or Texas, where conflicts over water use are documented. The French think tank [The Shift Project](https://theshiftproject.org/app/uploads/2025/10/Presentation-ppt-PIA.pdf) (October 2025) points to competing uses of water around the sites and recommends setting a ceiling trajectory for data centre electricity, listing the sites and their consumption and bringing the sector into France's national low-carbon strategy.

Operators respond in 2 ways. Air cooling (dry cooling) exchanges heat directly with the atmosphere, with no evaporation. It can be used anywhere, at the cost of lower thermal efficiency and higher electricity costs, and it is becoming viable as modern servers tolerate higher operating temperatures. Google says it chooses the cooling method site by site, according to the water and carbon-free electricity available locally. Immersion cooling submerges the servers in a dielectric fluid.

At Projet Celsius, we therefore think the useful question to put to an AI provider concerns its sites rather than water per query: "where do your data centres run, and how are they cooled?"

## Key takeaways

- A query uses a few drops of water: 0.26 mL measured by Google for Gemini and ~0.3 mL cited by Sam Altman for ChatGPT, with no published scope. Google's figure counts only cooling.
- The "half a litre per query" figure is a misleading shortcut: the 500 mL was for 10 to 50 GPT-3 responses, and the ~519 mL for a GPT-4 email counts total water.
- From 0.26 mL (cooling) to 45 mL (Mistral's full LCA), each published figure is accurate within its own scope, and that scope must be checked before any comparison.
- Geography decides: the same query has a small footprint in Sweden and a large one in Arizona, and the real issue is the concentration of data centres in basins that are already under strain.
- For a company, the water used by digital services is assessed in a multi-criteria LCA, because the Bilan Carbone® counts only carbon.

Water completes the picture left by energy figures. This aspect is best read alongside [the energy comparison of AI models](https://projetcelsius.com/en/blog/chatgpt-vs-claude-vs-gemini-environmental-impact/) and [the carbon footprint of AI at organisation level](https://projetcelsius.com/en/blog/chatgpt-carbon-footprint/), then carried into an [LCA](https://projetcelsius.com/en/blog/lca-vs-carbon-footprint/) as soon as your analysis of digital services needs to be taken seriously.

## Frequently asked questions

## Frequently asked questions

### How much water does ChatGPT use per prompt?

There is no audited measurement: in a June 2025 blog post, Sam Altman, OpenAI's chief executive, cited ~0.3 mL for an average query (one fifteenth of a teaspoon), without publishing a method. The figure is consistent with the only measurement made in production, Google's for Gemini: 0.26 mL for a median query, which counts only the data centre's cooling and leaves out the water tied to electricity.

### Does ChatGPT use half a litre of water per prompt?

No, not for an everyday query. The figure comes from 2 studies by the University of California, Riverside, which count total water, meaning cooling plus the water used to generate the electricity. They estimated around 500 mL for 10 to 50 GPT-3 responses (2023), then ~519 mL for a 100-word email generated by GPT-4 (2024). Repeated as "half a litre per message", the figures are applied to a different scope and a different model: an everyday query on a recent model evaporates a few tenths of a millilitre in cooling.

### How much water does a Gemini, Claude or Mistral query use?

Gemini: 0.26 mL for a median query, the only measurement made in production to have been published (Google, 2025), cooling only. Mistral: 45 mL per response of 400 tokens, but from a full life cycle assessment that includes electricity and server manufacturing, so the 2 figures cannot be compared like for like. Claude: Anthropic has published no water figure per query to date.

### Why do data centres use water?

To cool the servers. Servers heat up when they compute, and many sites remove that heat through towers that evaporate water. This water is lost to the local basin: Google states that it consumes 80% of the water it withdraws in this way. In addition, water is used upstream to generate the servers' electricity, which "per query" figures often ignore.

### Is AI's water use a real environmental problem?

Not at the scale of your own query, but yes at the scale of the infrastructure. Google consumed 41 billion litres of water in 2025, 34% more than in 2024. The IEA projects water consumption linked to data centres (cooling, generating their electricity and manufacturing chips) of 1,200 billion litres in 2030, 2.1 times the 2023 level. It attributes this increase to AI, which accounts for 5 to 15% of data centre electricity today and could account for 35 to 50% in 2030. For 2030, the UN report gives a volume nearly 8 times higher. The problem is the concentration of these volumes in basins that are already water-stressed (Arizona, the Persian Gulf, Texas). Technical solutions exist, namely air cooling (dry cooling) and immersion cooling, but their roll-out will have to keep pace with the construction of data centres.

### Are there data centres that use no water?

Yes, there are 2 mature technical routes. Air cooling (dry cooling) exchanges heat directly with the atmosphere, so nothing evaporates. It can be applied anywhere, at the cost of lower thermal efficiency, and is now viable because modern servers tolerate higher operating temperatures. Google says it chooses the cooling method site by site, according to the water and carbon-free electricity available locally. Immersion cooling submerges the servers in a dielectric fluid, with higher computing density. How widely they spread depends mainly on their cost and on regulation.

## Sources

**Sources**: [Google, "Measuring the environmental impact of delivering AI at Google Scale" (median Gemini query: 0.24 Wh, 0.26 mL of on-site cooling water; WUE 1.15 L/kWh; 80% of water withdrawn is consumed), arXiv 2508.15734](https://arxiv.org/abs/2508.15734); [Sam Altman, "The Gentle Singularity" (~0.34 Wh and ~0.3 mL per ChatGPT query, no methodology), June 2025](https://blog.samaltman.com/the-gentle-singularity); ["Making AI Less Thirsty", UC Riverside (500 mL for 10-50 GPT-3 responses; projection of 4.2-6.6 billion m³ of AI water withdrawal in 2027), arXiv 2304.03271](https://arxiv.org/abs/2304.03271); [Washington Post × UC Riverside, "A bottle of water per email" (~519 mL per GPT-4 email, total water), September 2024](https://www.washingtonpost.com/technology/2024/09/18/energy-ai-use-electricity-water-data-centers/); [Mistral AI × Carbone 4 × ADEME, life cycle assessment of Mistral Large 2 (45 mL per 400-token response)](https://mistral.ai/news/our-contribution-to-a-global-environmental-standard-for-ai/); [Google, data centre cooling (450,000 gallons per day per site on average, 2021 data)](https://blog.google/company-news/outreach-and-initiatives/sustainability/our-commitment-to-climate-conscious-data-center-cooling/); [Google, Environmental Report 2026 (10.9 billion gallons, or 41 billion litres, consumed in 2025, +34% year on year)](https://sustainability.google/google-2026-environmental-report/); [AWS, water stewardship (WUE 0.12 L/kWh of water withdrawn in 2025, against 0.15 in 2024)](https://sustainability.aboutamazon.com/natural-resources/water); [LBNL for the US Department of Energy, 2024 United States Data Center Energy Usage Report (17.4 billion gallons of water consumed on site in 2023 for 176 TWh, or ~0.4 L/kWh)](https://eta-publications.lbl.gov/sites/default/files/2024-12/lbnl-2024-united-states-data-center-energy-usage-report_1.pdf); [The Shift Project, "Artificial intelligence, data, computing: what infrastructure in a decarbonised world?" (our translation of the French title)](https://theshiftproject.org/app/uploads/2025/10/Presentation-ppt-PIA.pdf) (October 2025: competing uses of water, ceiling trajectory for data centre electricity); [IEA, "Energy and AI"](https://www.iea.org/reports/energy-and-ai) (water consumption linked to data centres, 560 billion L in 2023 and 1,200 billion L projected for 2030; AI share of data centre electricity from 5-15% to 35-50%); [UN News, "AI threatens water and energy resources" (our translation of the French title)](https://news.un.org/fr/story/2026/06/1158934) (UNU-INWEH report, June 2026: 4,500 billion litres consumed in 2025, 9,300 billion litres in 2030). Apart from Google (measurement) and Mistral (LCA), no water figure per query is measured or audited: the other values are statements, third-party estimates or projections, which is why this article reasons in orders of magnitude. The UN News and Shift Project pages are in French. **Images**: Opening photo: Tony Webster, [Wikimedia Commons](https://commons.wikimedia.org/wiki/File:Google_Datacenter_-_The_Dalles,_Oregon_%2817832143871%29.jpg), CC BY 2.0, cropped.
