AI Prompt Energy Calculator

Your use

Text prompts you send on an average day, across all chatbots.

Figures checked on October 24, 2026

Nothing is fetched live: the figures are fixed in the page, each with its source, and reviewed every month.

Estimate

Energy: 0.34 Wh (Sam Altman, average ChatGPT query (Jun 2025)) · Water: 0.322 mL (Sam Altman, average ChatGPT query (Jun 2025)) · CO₂e: 0.127 g (Altman's 0.34 Wh on the average US grid (823.1 lb CO₂ per MWh, EPA eGRID))

These are averages for one text prompt in a chatbot. Images, video, long documents and reasoning models use more, and every source measures something slightly different, so the figures are shown side by side rather than averaged.

Sources

Your footprint

Energy per year
Water per year
2.35 L
CO₂e per year
927 g
Energy per day
6.8 Wh
Water per day
6.44 mL
CO₂e per day
2.54 g

A year of prompts is about

Smartphone full charges
131
Kettles of 1 L brought to the boil
26.7
Kilometres driven in an average US car
3.73
Teaspoons of water
477
500 mL bottles of water
4.7

A phone charge is 19 Wh (US EPA). Boiling 1 L of water from 20 °C takes at least 93 Wh (water's specific heat; a real kettle uses a little more). An average US car emits about 400 g of CO₂ per mile, 249 g per km (US EPA). A US teaspoon is 4.93 mL.

Published figures side by side

Per text prompt as each source states it, and for a year at your number of prompts. A dash means the source does not publish that figure.

Published figures side by side
SourceEnergy per promptWater per promptCO₂e per promptEnergy per yearWater per yearCO₂e per year
Google, median Gemini Apps text prompt (Aug 2025)Measured in May 2025: chips, servers, idle capacity and data-centre overhead. Market-based CO₂e, water consumed by cooling. Excludes training, networks and your device.0.24 Wh0.26 mL0.03 g1.75 kWh1.9 L219 g
Google, the same prompt counting only the AI chipsThe narrow scope that many estimates use; Google calls it an optimistic scenario that underestimates the real footprint.0.1 Wh0.12 mL0.02 g730 Wh876 mL146 g
Sam Altman, average ChatGPT query (Jun 2025)Stated in a blog post without a method: 0.34 Wh and 0.000085 US gallons of water (0.32 mL).0.34 Wh0.322 mL—2.48 kWh2.35 L—
Epoch AI, typical GPT-4o query (Feb 2025)An estimate from hardware and a 500-token answer. A 10,000-token input takes about 2.5 Wh.0.3 Wh——2.19 kWh——
Jegham et al., short GPT-4o prompt (2025)A modelled estimate for 100 input and 300 output tokens, data-centre overhead included (±0.13 Wh).0.42 Wh——3.07 kWh——
Mistral, one Le Chat answer (Jul 2025)The marginal impact of a 400-token answer from Mistral's lifecycle analysis, excluding users' devices; training is reported separately.—45 mL1.14 g—329 L8.32 kg
de Vries, Joule (2023)An early top-down estimate from 2023 servers: at most 2.9 Wh per ChatGPT request. Often quoted as 3 Wh.2.9 Wh——21.2 kWh——

About this tool

How much electricity, water and CO₂ does a ChatGPT or Gemini prompt use? Enter how many prompts you send a day and the calculator multiplies it by the published per-prompt figures, per day and per year, with everyday comparisons.

The page shows the primary figures side by side: Google's measured median Gemini prompt (August 2025), Sam Altman's average ChatGPT query (June 2025), Epoch AI's estimate, an academic model and Mistral's lifecycle analysis, each with its date and a link. The Low, Central and High toggle chooses which ones the headline uses.

How it works

Type your prompts per day. The result multiplies them by one figure per prompt for energy, water and CO₂e, then by 365 for a year.

Low uses Google's comprehensive median Gemini prompt: 0.24 Wh, 0.26 mL and 0.03 g CO₂e. Central uses Sam Altman's average ChatGPT query: 0.34 Wh and 0.32 mL; he gives no CO₂ figure, so the page multiplies his 0.34 Wh by the US average grid (EPA eGRID: 823.1 lb CO₂ per MWh), which gives 0.127 g. High uses the top of the published range: 2.9 Wh from a 2023 estimate, and Mistral's 45 mL and 1.14 g for one Le Chat answer.

The table below the result applies every source to your number of prompts, so you can see how far apart they are.

per year=prompts per day×figure per prompt×365\text{per year} = \text{prompts per day} \times \text{figure per prompt} \times 365

What the figures cover

All figures are per text prompt, as an average or a median. Generating an image or a video, uploading a long document or using a reasoning model takes many times more: Epoch AI puts a 10,000-token input at about 2.5 Wh and a 100,000-token input at almost 40 Wh.

The sources do not measure the same thing. Google counts the servers, idle capacity and building overhead but not training, networks or your device, and uses market-based emissions that reflect its clean-energy purchases. Mistral's figures for one answer come from a lifecycle analysis, with training reported separately and users' devices left out. Altman does not say what he counted. Water means water consumed, mostly evaporated for cooling; Google's figure does not say whether it includes water used by power plants.

Example

At 20 prompts a day on the Central estimate: 20 × 0.34 Wh = 6.8 Wh a day, and 6.8 × 365 = 2,482 Wh, about 2.48 kWh a year. That is about 131 phone charges (2,482 ÷ 19) or 27 kettles of 1 L (2,482 ÷ 93). Water: 20 × 0.32 mL × 365 = 2.35 L a year, about 4.7 bottles of 500 mL. CO₂e: 20 × 0.127 g × 365 = 0.93 kg, about 3.7 km in an average US car.

Frequently asked questions

How much water does ChatGPT use per prompt?

Sam Altman put the average ChatGPT query at about 0.000085 US gallons, 0.32 mL, roughly a fifteenth of a teaspoon (June 2025). Google measured 0.26 mL for its median Gemini text prompt (August 2025). Higher figures, such as 45 mL for one Mistral Le Chat answer or 500 mL for 10 to 50 answers in a 2023 study of GPT-3, draw wider boundaries, such as the water used to generate the electricity, or come from older hardware.

How much electricity does an AI prompt use?

Recent primary figures are 0.24 Wh for Google's median Gemini prompt, about 0.3 Wh for a typical GPT-4o query (Epoch AI) and 0.34 Wh for the average ChatGPT query (Sam Altman). The often-quoted 3 Wh comes from a 2023 estimate based on older servers and longer answers.

Why do the figures differ so much?

They count different things: the AI chips only or the whole data centre, operation only or training and manufacturing too, an average or a median, short or long answers, and hardware from different years. Google reported that the energy of its median prompt fell 33-fold between May 2024 and May 2025.

Does this include images, video or training the model?

No. The figures are for answering text prompts. Google leaves training out and Mistral reports it separately; Altman does not say.

How often are the figures updated?

They are checked every month against the original sources, and the date is shown above the sources. Nothing is fetched live.

Sources

Link with your values

Add these parameters to this page’s address to open it with the values filled in, for example to share a calculation or to link to it from an AI assistant. The result is calculated in the browser: the page HTML shows an example for the default values, and the tool sets the data-url-params attribute once the link’s values are applied.

ParameterValueIf omitted
prompts_dailyPrompts per daynumber; allowed 0–10000000required
estimateEstimateone of: low, central, highdefault central

Examples

  • ?prompts_daily=50
  • ?prompts_daily=100&estimate=high
  • ?prompts_daily=5&estimate=low

Write decimals with a point. Encode a space as %20, + as %2B, # as %23 and & as %26.

Result elements: [data-result-block="ai_prompt_energy_calculator"]