Database structure
The database arranges 132 entries into 6 levels, from calculation factors to territories; each level is calculated from the levels below it. The 18 infrastructure entries, published by others, stand apart.
- France, inhabitants, 1 year
13 ktCO2e
An entire population (a typical city, France, the European Union, the world) or a global service, over a day or a year.
Table of the 19 territoriesTypical cities
Regions, countries and world
Populations
Global services
From level 4 to level 6, bypassing organisations : × 865,125office workers2.31 million in France, weighted at 37.5% by their AI use, and 17 other occupations, personal use and education
- SME of 50 people, 1 year
12 kgCO2e
A company, a team or an AI service, over a year, with the line to report in a carbon footprint.
Table of the 18 organisationsOrganisations by size
Teams
AI services of an organisation
- Coding agents, one team1.8 tCO2e
- Content production16 kgCO2e
- Licences for everyone130 kgCO2e
- Document assistant34 kgCO2e
- Automated customer service28 kgCO2e
- Process automation5.5 kgCO2e
- Data analysis1.4 kgCO2e
- Website translation690 gCO2e
- Meeting transcription36 kgCO2e
- Training with an AI tutor6.5 kgCO2e
- AI feature of a product3.7 tCO2e
- Hosted model4.8 tCO2e
- Retraining a model350 kgCO2e
From level 4 to level 5: × 1.875office worker5 out of 50, weighted at 37.5% by their AI use, and 13 other occupations, 3 services
- Office worker, 1 year
590 gCO2e
An occupation, a personal use or a project, over a year: a task basket, each task with its frequency and its share done with AI.
Table of the 38 occupations and activitiesOffice and administration
Writing, analysis and consulting
Development and data
Visual and video creation
Customer relations and field work
Teaching and studies
Personal use, subscriptions and licences
- Free personal use, 1 year340 gCO2e
- ChatGPT Plus, 1 year1.4 kgCO2e
- ChatGPT Pro, 1 year91 kgCO2e
- Claude Pro, 1 year1.3 kgCO2e
- Claude Max, 1 year120 kgCO2e
- Google AI Pro, 1 year2.1 kgCO2e
- Vibe Pro, 1 year1.2 kgCO2e
- Perplexity Pro, 1 year6.3 kgCO2e
- Microsoft 365 Copilot, 1 year1.3 kgCO2e
- GitHub Copilot, 1 year29 kgCO2e
- ChatGPT Business, 1 year1 kgCO2e
Projects delegated to AI
From level 3 to level 4: × 330emails written with AI per yearout of 3,300 written in the year, and 10 other uses
- Write an email
0.11 gCO2e
One request made to an AI, once: a question, an email, a page, an image, an agent hour.
Table of the 39 usesWrite and converse
Read and summarise
Search
Create images
Video and voice
Code
Let an AI act
Composite tasks
By assistant, with its own source
From level 2 to level 3: × 0.157for 157 tokens writtenand 184 read, then × 1.1 for the data centre
- Write 1,000 tokens
0.44 gCO2e
The server's operations, in their units: 1,000 tokens read or written, an image, a second of video, a minute of audio.
Table of the 11 machine operationsText
Image
Video and sound
Data
From level 1 to level 2: × 0.96 Whof electricity
- Grid electricity, world
458 gCO2e/kWh
Electricity and water per kilowatt-hour, the data centre, the GPU hour, hardware manufacturing.
Table of the 7 calculation factorsElectricity and water
Data centre
Computing and manufacturing
Outside the levels
Reproduced as they stand from their source, they overlap with the territories without adding to them.
Data centres by country, companies, model training, chip manufacturing, digital technology.
Table of the 18 infrastructure entriesData centres
Tech companies
Model training
Hardware
Digital sector benchmarks
Worked example: an email written with an AI, servers in the world. Source: AI Impact Database, version 0.4, Projet Celsius. CC BY 4.0 licence.
