Skip to content
ChartWaterUses · mL per question
Published on 20 October 2026

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)

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

0.11× 1010× 10100 mL× 10ScalelogarithmicmL (logarithmic scale)Google, Geminion-site cooling,measurement0.26 mLOpenAI, ChatGPTdeclaration, no methodology0.32 mLUC Riverside, GPT-3cooling and electricity10 to 50 mLMistral AI, Le Chatfull life cycle45 mLa factor of 170with Google, by scope
Source: Google; OpenAI; UC Riverside; Mistral AI. projetcelsius.com/en/ai · CC BY 4.0AI Impact, by Projet Celsius

The analysis

  • A factor of about 190 separates the lowest, 0.26 mL (Google), from the highest, 50 mL (UC Riverside): it comes from scope and task, not from a calculation error.
  • Google counts the water evaporated to cool its data centres; Mistral adds the water of electricity and of server manufacturing.
  • "A bottle of water per email" comes from a 2024 Washington Post article that we could not check at source (paywalled): its figure is neither plotted nor reused here.

The figures

  • Google, Gemini (on-site cooling, measured): 0.26 mL.
  • OpenAI, ChatGPT (declaration, no method): 0.32 mL.
  • UC Riverside, GPT-3 (cooling and electricity): 10 to 50 mL.
  • Mistral AI, Le Chat (full life cycle): 45 mL.

Assumptions

  • Water consumed, under 4 different scopes.
  • 4 studies, 4 different models and 4 different tasks.
  • Withdrawing and consuming water are not the same thing; the water stress of the location matters more than the volume.

Published on 20 October 2026MethodologyAll sourcesLicence CC BY 4.0

Charts and maps

ChartWaterMicrosoft'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 2025Key figureWaterWater consumed by the world's data centresWorld · per year, in 2023 · water consumed, direct and indirect · IEA

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 ↗