AI Impact Database
Environmental database for artificial intelligence · by Projet Celsius
My year with AI
740 gCO2e
over one year, equivalent to 5.2 km by petrol or diesel car
Deep research tasks, 270 gCO2e, 37% of the year
Run the full calculation →Photo Shixart1985, CC BY 2.0
Main uses
Emissions from one use, from the lightest request to the heaviest.
Text and conversation
- Write an email0.11 gCO2e
- Quick question0.14 gCO2e
- Write a page0.21 gCO2e
- Conversation of 10 exchanges1.6 gCO2e
Search
- AI overview of a search0.061 gCO2e
- Web search with an AI1.3 gCO2e
- Literature search13 gCO2e
- Deep research45 gCO2e
Image
- Create an image1.5 gCO2e
- Retouch a photo1.6 gCO2e
- Create a large image5.9 gCO2e
- Visual in 4 attempts5.9 gCO2e
Video
- Second of video7.6 gCO2e
- Clip of 5 seconds45 gCO2e
- Clip of 10 seconds160 gCO2e
- Video of 3 minutes8.4 kgCO2e
Voice
- Transcribe one minute0.17 gCO2e
- Voice mode, 1 minute0.29 gCO2e
- Synthesise one minute0.53 gCO2e
- Summarise a meeting14 gCO2e
Code and agents
- Code suggestion0.031 gCO2e
- Generate a function0.54 gCO2e
- Browser agent5.3 gCO2e
- Coding agent hour310 gCO2e
Other topics
Occupations20 entries
Personal use and subscriptions11 entries
Organisations and services18 entries
Cities, territories, populations15 entries
Major AI services8 entries
Electricity and data centres11 entries
Water8 entries
Tech companies4 entries
Hardware and training5 entries
Benchmarks for comparison6 entries
132 entries in 6 levels, plus 18 infrastructure entries kept separate
All entries
Columns
| Unit | |||||
|---|---|---|---|---|---|
| Re-read from cache1,000 tokens | 0.015 gCO2e | 1,000 tokens | 0.032 Wh | convention | |
| Read 1,000 tokens1,000 tokens | 0.15 gCO2e | 1,000 tokens | 0.32 Wh | Watershed | published |
| Write 1,000 tokens1,000 tokens | 0.44 gCO2e | 1,000 tokens | 0.96 Wh | Watershed | published |
| Write, with reasoning1,000 tokens | 0.84 gCO2e | 1,000 tokens | 1.8 Wh | convention | |
| Reasoningfactor | × 1.92 | factor | PEReN and 3 others | measured | |
| Write an email1 email | 0.11 gCO2e | 1 email | 0.23 Wh | estimated | |
| Quick question1 question | 0.14 gCO2e | 1 question | 0.3 Wh | Google and 3 others | estimated |
| Write a page1 page | 0.21 gCO2e | 1 page | 0.45 Wh | estimated | |
| Proofread a page1 page | 0.27 gCO2e | 1 page | 0.6 Wh | estimated | |
| Translate a page1 page | 0.27 gCO2e | 1 page | 0.6 Wh | estimated | |
| Article of 1,000 words1 article | 0.51 gCO2e | 1 article | 1.1 Wh | estimated | |
| Conversation of 10 exchanges1 conversation | 1.6 gCO2e | 1 conversation | 3.5 Wh | estimated | |
| Analyse a spreadsheet1 analysis | 1.3 gCO2e | 1 analysis | 2.8 Wh | estimated | |
| Query a PDF1 question | 2.2 gCO2e | 1 question | 4.7 Wh | estimated | |
| Summarise 30 pages1 summary | 2.2 gCO2e | 1 summary | 4.9 Wh | estimated | |
| Handle a customer request1 request | 0.56 gCO2e | 1 request | 1.2 Wh | estimated | |
| Write meeting minutes1 set of minutes | 0.66 gCO2e | 1 set of minutes | 1.4 Wh | estimated | |
| Review a contract1 contract | 2.7 gCO2e | 1 contract | 5.8 Wh | estimated | |
| Prepare a lesson1 lesson | 2.8 gCO2e | 1 lesson | 6.2 Wh | estimated | |
| Presentation of 10 slides1 presentation | 5.2 gCO2e | 1 presentation | 11 Wh | estimated | |
| Question to Gemini1 question | 0.11 gCO2e | 1 question | 0.24 Wh | measured | |
| Question to Claude1 question | 0.14 gCO2e | 1 question | 0.3 Wh | estimated | |
| Question to ChatGPT1 question | 0.16 gCO2e | 1 question | 0.34 Wh | OpenAI | published |
| Le Chat, LCA1 answer | 1.14 gCO2e | 1 answer | Mistral AI | published | |
| Index 1,000 pages1,000 pages | 1.4 gCO2e | 1,000 pages | 3 Wh | estimated | |
| Deep research1 search | 45 gCO2e | 1 search | 99 Wh | estimated | |
| AI overview of a search1 search | 0.061 gCO2e | 1 search | 0.13 Wh | estimated | |
| Web search (benchmark)1 search | 0.14 gCO2e | 1 search | 0.3 Wh | published | |
| Web search with an AI1 search | 1.3 gCO2e | 1 search | 2.7 Wh | estimated | |
| Literature search1 search | 13 gCO2e | 1 search | 29 Wh | estimated | |
| Analyse an image1 image | 0.16 gCO2e | 1 image | 0.35 Wh | estimated | |
| Generate an image1 image | 1.3 gCO2e | 1 image | 2.91 Wh | Luccioni et al. | measured |
| Create an image1 image | 1.5 gCO2e | 1 image | 3.2 Wh | measured | |
| Retouch a photo1 photo | 1.6 gCO2e | 1 photo | 3.5 Wh | estimated | |
| Create a large image1 image | 5.9 gCO2e | 1 image | 13 Wh | estimated | |
| Visual in 4 attempts1 visual | 5.9 gCO2e | 1 visual | 13 Wh | estimated | |
| Second of video1 second of video | 7.6 gCO2e | 1 second of video | 16.7 Wh | estimated | |
| Clip of 5 seconds1 video | 45 gCO2e | 1 video | 99 Wh | measured | |
| Clip of 10 seconds1 video | 160 gCO2e | 1 video | 340 Wh | estimated | |
| Video of 3 minutes1 project | 8.4 kgCO2e | 1 project | 18 kWh | estimated | |
| Transcribe one minute1 minute of audio | 0.17 gCO2e | 1 minute of audio | 0.38 Wh | estimated | |
| Synthesise one minute1 minute of voice | 0.53 gCO2e | 1 minute of voice | 1.2 Wh | estimated | |
| Voice mode, 1 minute1 minute | 0.29 gCO2e | 1 minute | 0.62 Wh | estimated | |
| Summarise a meeting1 meeting | 14 gCO2e | 1 meeting | 30 Wh | estimated | |
| Code suggestion1 suggestion | 0.031 gCO2e | 1 suggestion | 0.069 Wh | estimated | |
| Generate a function1 function | 0.54 gCO2e | 1 function | 1.2 Wh | estimated | |
| Have code reviewed1 review | 0.59 gCO2e | 1 review | 1.3 Wh | estimated | |
| Coding agent hour1 hour | 310 gCO2e | 1 hour | 670 Wh | estimated | |
| Specialised task1 task | 0.00092 gCO2e | 1 task | 0.002 Wh | Luccioni et al. | measured |
| Automated workflow1 run | 0.11 gCO2e | 1 run | 0.24 Wh | estimated | |
| Browser agent1 task | 5.3 gCO2e | 1 task | 12 Wh | estimated | |
| Book written with AI1 project | 340 gCO2e | 1 project | 750 Wh | estimated | |
| End-to-end study1 project | 540 gCO2e | 1 project | 1.2 kWh | estimated | |
| Automated monitoring, 1 year1 project | 7.6 kgCO2e | 1 project | 17 kWh | estimated | |
| Application with an agent1 project | 12 kgCO2e | 1 project | 27 kWh | estimated | |
| Autonomous agent, 1 month1 month | 57 kgCO2e | 1 month | 120 kWh | estimated | |
| Agents talking to each other, 1 month1 month | 210 kgCO2e | 1 month | 470 kWh | estimated | |
| Human resources, 1 year1 person, 1 year | 330 gCO2e | 1 person, 1 year | 720 Wh | estimated | |
| Manager or executive, 1 year1 person, 1 year | 370 gCO2e | 1 person, 1 year | 810 Wh | estimated | |
| Accountant, 1 year1 person, 1 year | 380 gCO2e | 1 person, 1 year | 830 Wh | estimated | |
| Office worker, 1 year1 person, 1 year | 590 gCO2e | 1 person, 1 year | 1.3 kWh | estimated | |
| Executive assistant, 1 year1 person, 1 year | 630 gCO2e | 1 person, 1 year | 1.4 kWh | estimated | |
| Lawyer, 1 year1 person, 1 year | 520 gCO2e | 1 person, 1 year | 1.1 kWh | estimated | |
| Researcher, 1 year1 person, 1 year | 680 gCO2e | 1 person, 1 year | 1.5 kWh | estimated | |
| Communications officer, 1 year1 person, 1 year | 800 gCO2e | 1 person, 1 year | 1.7 kWh | estimated | |
| Consultant, 1 year1 person, 1 year | 970 gCO2e | 1 person, 1 year | 2.1 kWh | estimated | |
| Journalist, 1 year1 person, 1 year | 2.1 kgCO2e | 1 person, 1 year | 4.6 kWh | estimated | |
| Data scientist, 1 year1 person, 1 year | 1.5 kgCO2e | 1 person, 1 year | 3.2 kWh | estimated | |
| Developer, 1 year1 person, 1 year | 2 kgCO2e | 1 person, 1 year | 4.3 kWh | estimated | |
| Graphic designer, 1 year1 person, 1 year | 1.4 kgCO2e | 1 person, 1 year | 3 kWh | estimated | |
| Videographer, 1 year1 person, 1 year | 3.4 kgCO2e | 1 person, 1 year | 7.4 kWh | estimated | |
| Field-based occupation, 1 year1 person, 1 year | 220 gCO2e | 1 person, 1 year | 480 Wh | estimated | |
| Salesperson, 1 year1 person, 1 year | 660 gCO2e | 1 person, 1 year | 1.4 kWh | estimated | |
| Customer service officer, 1 year1 person, 1 year | 960 gCO2e | 1 person, 1 year | 2.1 kWh | estimated | |
| Secondary school pupil, 1 year1 pupil, 1 year | 150 gCO2e | 1 pupil, 1 year | 340 Wh | estimated | |
| Teacher, 1 year1 person, 1 year | 260 gCO2e | 1 person, 1 year | 570 Wh | estimated | |
| Student, 1 year1 student, 1 year | 490 gCO2e | 1 student, 1 year | 1.1 kWh | estimated | |
| Free personal use, 1 year1 person, 1 year | 340 gCO2e | 1 person, 1 year | 750 Wh | estimated | |
| ChatGPT Business, 1 year1 subscription, 1 year | 1 kgCO2e | 1 subscription, 1 year | 2.2 kWh | estimated | |
| Vibe Pro, 1 year1 subscription, 1 year | 1.2 kgCO2e | 1 subscription, 1 year | 2.6 kWh | estimated | |
| Claude Pro, 1 year1 subscription, 1 year | 1.3 kgCO2e | 1 subscription, 1 year | 2.8 kWh | estimated | |
| Microsoft 365 Copilot, 1 year1 subscription, 1 year | 1.3 kgCO2e | 1 subscription, 1 year | 2.9 kWh | estimated | |
| ChatGPT Plus, 1 year1 subscription, 1 year | 1.4 kgCO2e | 1 subscription, 1 year | 3 kWh | estimated | |
| Google AI Pro, 1 year1 subscription, 1 year | 2.1 kgCO2e | 1 subscription, 1 year | 4.6 kWh | estimated | |
| Perplexity Pro, 1 year1 subscription, 1 year | 6.3 kgCO2e | 1 subscription, 1 year | 14 kWh | estimated | |
| GitHub Copilot, 1 year1 subscription, 1 year | 29 kgCO2e | 1 subscription, 1 year | 63 kWh | estimated | |
| ChatGPT Pro, 1 year1 subscription, 1 year | 91 kgCO2e | 1 subscription, 1 year | 200 kWh | estimated | |
| Claude Max, 1 year1 subscription, 1 year | 120 kgCO2e | 1 subscription, 1 year | 270 kWh | estimated | |
| Very small business (5)1 organisation, 1 year | 480 gCO2e | 1 organisation, 1 year | 1.1 kWh | estimated | |
| SME of 50 people1 organisation, 1 year | 12 kgCO2e | 1 organisation, 1 year | 27 kWh | estimated | |
| Mid-cap company of 500 people1 organisation, 1 year | 150 kgCO2e | 1 organisation, 1 year | 320 kWh | estimated | |
| Large company (5,000)1 organisation, 1 year | 1.4 tCO2e | 1 organisation, 1 year | 3.2 MWh | estimated | |
| Team of 12, 1 year1 team, 1 year | 7.1 kgCO2e | 1 team, 1 year | 15 kWh | estimated | |
| Website translation1 service, 1 year | 690 gCO2e | 1 service, 1 year | 1.5 kWh | estimated | |
| Data analysis1 service, 1 year | 1.4 kgCO2e | 1 service, 1 year | 3.1 kWh | estimated | |
| Process automation1 service, 1 year | 5.5 kgCO2e | 1 service, 1 year | 12 kWh | estimated | |
| Training with an AI tutor1 service, 1 year | 6.5 kgCO2e | 1 service, 1 year | 14 kWh | estimated | |
| Content production1 service, 1 year | 16 kgCO2e | 1 service, 1 year | 34 kWh | estimated | |
| Automated customer service1 service, 1 year | 28 kgCO2e | 1 service, 1 year | 61 kWh | estimated | |
| Document assistant1 service, 1 year | 34 kgCO2e | 1 service, 1 year | 75 kWh | estimated | |
| Meeting transcription1 service, 1 year | 36 kgCO2e | 1 service, 1 year | 78 kWh | estimated | |
| Licences for everyone1 service, 1 year | 130 kgCO2e | 1 service, 1 year | 290 kWh | estimated | |
| Retraining a model1 service, 1 year | 350 kgCO2e | 1 service, 1 year | 760 kWh | estimated | |
| Coding agents, one team1 team, 1 year | 1.8 tCO2e | 1 team, 1 year | 4 MWh | estimated | |
| AI feature of a product1 service, 1 year | 3.7 tCO2e | 1 service, 1 year | 8.1 MWh | estimated | |
| Hosted model1 service, 1 year | 4.8 tCO2e | 1 service, 1 year | 10 MWh | estimated | |
| Village (2,000)1 territory, 1 year | 380 kgCO2e | 1 territory, 1 year | 820 kWh | estimated | |
| Town (20,000)1 territory, 1 year | 3.8 tCO2e | 1 territory, 1 year | 8.2 MWh | estimated | |
| Town (100,000)1 territory, 1 year | 19 tCO2e | 1 territory, 1 year | 41 MWh | estimated | |
| Large city (500,000)1 territory, 1 year | 94 tCO2e | 1 territory, 1 year | 210 MWh | estimated | |
| Metropolis (2 million)1 territory, 1 year | 380 tCO2e | 1 territory, 1 year | 820 MWh | estimated | |
| Île-de-France, inhabitants1 territory, 1 year | 3.2 ktCO2e | 1 territory, 1 year | 7 GWh | estimated | |
| France, inhabitants1 territory, 1 year | 13 ktCO2e | 1 territory, 1 year | 28 GWh | estimated | |
| European Union, inhabitants1 territory, 1 year | 77 ktCO2e | 1 territory, 1 year | 170 GWh | estimated | |
| World, inhabitants1 territory, 1 year | 640 ktCO2e | 1 territory, 1 year | 1.4 TWh | estimated | |
| Aged 65 and over1 population, 1 year | 270 tCO2e | 1 population, 1 year | 600 MWh | estimated | |
| Public employees1 population, 1 year | 640 tCO2e | 1 population, 1 year | 1.4 GWh | estimated | |
| Pupils and students1 population, 1 year | 1.5 ktCO2e | 1 population, 1 year | 3.4 GWh | estimated | |
| Under 251 population, 1 year | 2.2 ktCO2e | 1 population, 1 year | 4.8 GWh | estimated | |
| Aged 25 to 641 population, 1 year | 4.1 ktCO2e | 1 population, 1 year | 8.8 GWh | estimated | |
| Office workers1 population, 1 year | 4.6 ktCO2e | 1 population, 1 year | 10 GWh | estimated | |
| Vibe, per day1 day | 210 kgCO2e | 1 day | 450 kWh | estimated | |
| Gemini app, per day1 day | 300 tCO2e | 1 day | 650 MWh | estimated | |
| ChatGPT, per day1 day | 410 tCO2e | 1 day | 880 MWh | ADEME | published |
| AI assistants worldwide, per day1 day | 1.1 ktCO2e | 1 day | 2.4 GWh | estimated | |
| Grid electricity1 kWh | 458 gCO2e | 1 kWh | Ember | published | |
| PUE of the data centrefactor | × 1.10 | factor | Watershed and 6 others | convention | |
| Data centres, world1 year | 415 TWh | 1 year | IEA | published | |
| Data centres, France1 year | 4 to 5 TWh | 1 year | Projet Celsius and 1 other | published | |
| AI in data centres1 year | 100 to 174 TWh | 1 year | Projet Celsius and 2 others | published | |
| Data centres, United States1 year | 180 TWh | 1 year | IEA | published | |
| Data centres, China1 year | 100 TWh | 1 year | IEA | published | |
| Data centres, Europe1 year | 70 TWh | 1 year | IEA | published | |
| Data centres, Ireland1 year | 7.66 TWh | 1 year | CSO | measured | |
| Water for electricity generation1 kWh | 0.004 to 17 L | 1 kWh | LBNL and 1 other | published | |
| On-site water1 kWh | 1.15 L | 1 kWh | Google and 3 others | published | |
| Meta1 year | 6 MtCO2e | 1 year | 18.4 TWh | Meta | published |
| Microsoft1 year | 12 MtCO2e | 1 year | 37 TWh | Microsoft | published |
| Google1 year | 15 MtCO2e | 1 year | 43.6 TWh | published | |
| Top four cloud providers1 year | 110 TWh | 1 year | IEA | published | |
| Manufacturing per GPU hour1 GPU hour | 35 gCO2e | 1 GPU hour | GreenIT | estimated | |
| H100 hour1 GPU hour | 410 gCO2e | 1 GPU hour | 900 Wh | Epoch AI | estimated |
| Manufacturing a server1 server | 1.31 tCO2e | 1 server | Nvidia | published | |
| Training a model1 training run | 552 tCO2e | 1 training run | 1.29 GWh | Patterson et al. | published |
| Chip manufacturing1 year | 450 ktCO2e | 1 year | 984 GWh | Greenpeace | published |
| Digital footprint of one person (benchmark)1 person, 1 year | 434 kgCO2e | 1 person, 1 year | ADEME | published | |
| Footprint of one person1 person, 1 year | 8.3 tCO2e | 1 person, 1 year | Insee and SDES | published | |
| Digital sector in France (benchmark)1 year | 29.5 MtCO2e | 1 year | 51.5 TWh | ADEME | published |
| Digital sector worldwide1 year | 567 MtCO2e | 1 year | 1,180 TWh | ITU and World Bank | published |
| Electricity use of one person1 person, 1 year | 6.7 MWh | 1 person, 1 year | RTE | estimated | |
Calculated values depend on where the servers are located; published values are reproduced as they are. An empty source means that the database calculated the value from its own factors. An entry that falls under several topics is filed here under one only, its main topic: a topic page may therefore list more entries.
Cite the database
Projet Celsius (2026). AI Impact Database, version 0.4 of 13 November 2026. https://projetcelsius.com/en/ai/database/. CC BY 4.0 licence.







4
3





