AI in maps

Data centres are set up where electricity is available, the grid is strong and water is accessible. These maps show, country by country, what makes the footprint of the same computation vary.

Carbon intensity of electricity, country by country

Emissions from producing one kilowatt-hour, on a life cycle basis, 2025 or latest year published. The electricity for the same computation emits 14 times more CO2e in Poland than in France.

589Poland, gCO2e/kWh41France× 14Poland and France, same electricity

less than 100600 and above gCO2e/kWh

Sources: Ember and Our World in Data.

Open the map →

World maps

The Shuozhou coal power plant, in China: cooling towers and smoking chimneys behind high-voltage pylons, at the edge of the fields — **China**525 gCO2e/kWhThe Shuozhou coal power plant, in Shanxi (China).Photo Kleineolive, Wikimedia Commons, CC BY 3.0

MapClimate

The carbon intensity of electricity

Emissions per kWh generated, life cycle assessment, in gCO2e/kWh, 2025 or latest published year

World, 2025458gCO2e/kWh

Norway28g
France41g
Poland589g

Why this map matters for AI

It is the main determinant of the carbon footprint of an AI: for the same computation, a server emits 14 times more in Poland than in France and 21 times more than in Norway.

From lowest to highest emitting

Countries with nuclear or hydro electricity (France, Sweden, Norway, Switzerland, Brazil) are the darkest.
The United States, which hosts more than half of the world's AI data centres, is at 384 g; China at 525 g, India at 670 g.

Assumptions and limitations

Scope. Electricity generated in the country, all power plants, emissions from construction to operation.

An annual, national average: intensity varies hour by hour and from one region to another within a country (US states, for example).
A data centre can buy renewable electricity by contract: its accounting does not necessarily follow this map.
Over the life cycle (Ember), the world value is 458 g: this is the value used by the AI Impact Database, the hub and the blog, the same source for all countries. The IEA, which counts only the CO2 of power plants, gives 435 g.

Ember · Climate · 169 countriesOpen the map →

Google's data centre in Eemshaven, in the Netherlands, seen from above, surrounded by wind turbines, with industrial facilities on the horizon — **Netherlands**54 %Google's data centre in Eemshaven (Netherlands), in the middle of a wind farm.Photo Wvdp, Wikimedia Commons, CC0, cropped

MapElectricity

The share of low-carbon electricity

Share of nuclear and renewables in electricity generated, in %, 2025 or latest published year

World, 202543%

Norway99%
France95%
United States43%

Why this map matters for AI

Data centres look for abundant, low-carbon electricity. Where it exists, they compete with the electrification of transport, heating and industry.

Comparison of 4 countries

France is at 95%, Norway at 99%, the United States at 43%, China at 42%.
This map is the mirror of the carbon intensity map: the higher the low-carbon share, the lower the intensity.

Assumptions and limitations

Scope. Electricity generated in the country: nuclear, hydro, wind, solar, bioenergy.

The share of electricity generated, not consumed: trade between countries is not counted.
Low-carbon electricity can be scarce: a high share does not mean there is capacity left for new centres.

Ember · Electricity · 164 countriesOpen the map →

A centre-pivot irrigation rig watering a maize field in Dordogne, under a blue sky — **France**21 %Irrigation of a maize field in Dordogne, in 2014.Photo Cjp24, Wikimedia Commons, CC BY-SA 4.0, cropped

MapWater

Water stress, country by country

Freshwater withdrawals as a share of available resources, in %, 2023

World, 202318%

France21%
Spain43%
India66%

Why this map matters for AI

The volume of water of a data centre matters less than its location. Where withdrawals approach or exceed resources, every m³ evaporated by cooling weighs on agriculture and drinking water.

The 5 FAO levels

The FAO distinguishes 5 levels: no stress below 25%, low from 25 to 50%, medium from 50 to 75%, high from 75 to 100%, critical above that.
Above 100%, a country draws on non-renewable groundwater or desalinates seawater: this is the case for Saudi Arabia and Egypt.

Assumptions and limitations

Scope. All freshwater withdrawals of the country (agriculture, industry, domestic uses), as a share of renewable resources, after the needs of ecosystems.

A national average: stress plays out basin by basin. The United States (25%) has regions under critical stress, such as Arizona.
The indicator counts withdrawals, not consumption: water returned to the environment is included.

FAO · Water · 162 countriesOpen the map →

Compare countries

Aerial view of a data centre: 2 long light-roofed buildings in the middle of the Irish hedged farmland, alongside a motorway — **Ireland**Facebook's data centre in Clonee, near Dublin, in 2019.Photo Thomas Nugent, Wikimedia Commons, CC BY-SA 2.0

ChartCountries · 2025Server electricity, country by countryFranceWorldPoland589gCO2e/kWhPoland; France: 41.4Ember, 2026, via Our World in Data ChartCountries · 2023 to 2025Data centres' share of electricityIreland, 202523 %Netherlands4.6 %United States4.4 %World1.5 %France1 %23%of Ireland's electricityCSO; CBS; LBNL; IEA; RTE ToolClimate · 12 countries and regionsCompare the footprint of an AI use by the country where the servers are41.4 France330 Germany384 United States589 Poland× 14from France to PolandEmber; Our World in Data; Epoch AI

Across 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® ↗