One question, 4 ways of counting CO2e
Emissions of a written question to an AI, in gCO2e (logarithmic scale)
One question, 4 ways of counting CO2e
Sources
Source: Epoch AI; Ember; Google; Mistral AI. projetcelsius.com/en/ai · CC BY 4.0- Epoch AIHow much energy does ChatGPT use?February 2025 (new tab)
- EmberGlobal Electricity Review 2026, and annual data via Our World in Data2026 (new tab)
- GoogleElsworth et al. (Google), Measuring the environmental impact of delivering AI at Google scaleAugust 2025 (new tab)
- Mistral AIwith Carbone 4 and ADEME, Our contribution to a global environmental standard for AIJuly 2025 (new tab)
The analysis
- More than 90 times between the lowest and the highest: it comes from the electricity grid, purchase contracts and scope.
- Google counts its electricity according to its contracts (94 gCO2e/kWh in 2024), not according to the grid of its data centres (345 g).
- Mistral counts the whole life cycle, server manufacturing included, for a response of about half a page (400 tokens).
The figures
- 0.3 Wh, in France (calculation, 41.4 g/kWh): 0.012 g.
- Google, Gemini (according to its contracts): 0.03 g.
- 0.3 Wh, world (calculation, 458 g/kWh): 0.14 g.
- Mistral AI, Le Chat (life cycle, half a page): 1.14 g.
Assumptions
- One written question, 4 calculation methods.
- 0.3 Wh is an independent estimate (Epoch AI); Google measures 0.24 Wh for Gemini.
- None of these values counts the user's device, often the largest item in digital.
More on climate
The whole Climate topic →Charts and maps
Chart
ClimateServer electricity, country by country14times as muchin Poland than in France, for the same calculationChartClimateAI emissions, compared99MtCO2eAI in 2025, as much as BelgiumChartClimateThe digital sector in global emissions1.4 to 3.9%of global emissions, according to estimatesChartClimateGoogle and Microsoft emissions58 to 81%rise in emissions since the baseline year
ClimateServer electricity, country by country14times as muchin Poland than in France, for the same calculationChartClimateAI emissions, compared99MtCO2eAI in 2025, as much as BelgiumChartClimateThe digital sector in global emissions1.4 to 3.9%of global emissions, according to estimatesChartClimateGoogle and Microsoft emissions58 to 81%rise in emissions since the baseline yearAnalyses
AnalysisClimate12 November 2026Why do estimates of AI's impact differ?Calculated with French electricity, one question to an AI emits 0.012 gram of CO2e; Mistral AI publishes 1.14 gram, more than 90 times as much. Its water use ranges from 0.26 to 45 millilitres depending on the published measurements. These figures are accurate; they do not cover the same scope.
Where does this figure come from?Climate20 October 2026"Meta's electricity emits only 1,358 tonnes of CO2e"In 2024, Meta reports 1,358 tonnes of CO2e for the electricity it buys, the annual footprint of 170 people in France. The same report also gives 5.97 million tonnes. Both of the 2 figures are accurate.Tools
On the Projet Celsius blog ↗
ElectricityAcross the hub
AI Impact DatabaseAll the figures, their methodology and their sources.
AI Ranking10 assistants rated out of 100: performance, ethics and environment.
AnalysesThe signed pieces: analyses, viewpoints, "Where does this figure come from?".
Charts and mapsYou are hereThe reference visuals, with their data and sources.
ToolsOpen calculations, your assumptions on the left, the result on the right.
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® ↗
