AI at the office all year: what's the real footprint?
3.6 kWh of electricity, 3.6 liters of water and 2.4 kg of CO₂ a year: that's what heavy AI use adds up to for one office employee, based on this site's working assumption of 12,000 prompts a year. Worth a proper look at a footprint that gets pointed at a lot.
Even under a heavy-use assumption (12,000 prompts a year, about 33 a day), one employee's AI use adds up to 3.6 kWh of electricity a year: barely 14% of what a laptop draws if simply left running all year. Water and CO₂ tell the same story: a few liters, and the equivalent of 14 km driven by car, over a full working year.
Figures updated on June 30, 2026
How many AI prompts does an employee send in a year?
12,000 prompts a year is the heavy-use assumption this site works with for an active user, about 33 requests per working day. That's not a measured per-desk average: globally, ChatGPT's 700 million weekly users send about 18 billion messages a week, roughly 26 per person per week across all use. And the work-related share of those messages actually fell, from 47% in June 2024 to 27% in June 2025 (NBER / OpenAI): most public AI use is personal, not professional.
What electricity does that use over a year?
3.6 kWh a year is the electricity behind those 12,000 prompts, using this site's per-prompt cost for a standard text prompt (0.3 Wh, sourced from Epoch AI and Google Cloud). That's about 14% of the electricity a laptop draws if simply left running all year (~25 kWh), and a fraction of what any other office device uses over twelve months.
What about water and CO₂?
3.6 liters of water and 2.4 kg of CO₂ a year is what those same 12,000 prompts add up to (0.3 mL and 0.2 g per prompt, same sources). That water figure is less than a tenth of a single 5-minute shower (~50 liters). That CO₂ figure is roughly equal to 14 km driven by car, over an entire working year.
Does every employee use AI this much?
No, most use far less: in France, only 21% of executives (cadres) use generative AI daily in 2026, up from 12% a year earlier, and 50% use it at least weekly, per Apec's barometer. The calculation above is a high-end assumption; most employees fall well below it. On the other end, heavier use (reasoning, code analysis) can multiply the footprint by 65 to 70 times: a reasoning prompt costs about 20 Wh, versus 0.3 Wh for a standard text prompt.
Should you feel guilty about using AI at work?
No: for a single employee, even under the high-end assumption, the footprint stays tiny next to plenty of other everyday digital and physical habits. The real question isn't the individual prompt, it's the sum across the millions of employees using it every day: scale is what changes the picture, not the isolated action.
Related equivalents
Sources
- Epoch AI : énergie par requête ChatGPT
- Google Cloud : impact environnemental de l'inférence IA
- ADEME : consommation des appareils ménagers
- « How Hungry is AI? » (arXiv 2505.09598)
- Apec : baromètre « Les cadres et l'IA » (mars 2026) : usage quotidien 12 % → 21 %, usage hebdomadaire 35 % → 50 %
- NBER / OpenAI : « How People Use ChatGPT » (Chatterji et al., 2025) : 700 M d'utilisateurs hebdomadaires, ~18 Md de messages/semaine, part professionnelle 47 % → 27 % (juin 2024 → juin 2025)
Frequently asked
How many AI prompts does an employee send per year?+
This site uses a heavy-use assumption of 12,000 prompts a year (~33/day) for an active user. In practice, most employees fall well below that: only 21% of French executives use AI daily in 2026 (Apec).
Does an employee's AI use emit a lot of CO₂?+
No: even under the high-end assumption, those 12,000 prompts emit about 2.4 kg of CO₂ a year, the equivalent of barely 14 km driven by car.
Does the footprint change by request type?+
Yes, a lot: a reasoning prompt costs about 20 Wh versus 0.3 Wh for a standard text prompt, 65 to 70 times more. The site's “Convert” mode lets you simulate your own usage.