How Much Energy Does ChatGPT Use?

How Much Energy ChatGPT Uses per Prompt

OpenAI CEO Sam Altman puts an average ChatGPT query at about 0.34 watt-hours, and Google reports 0.24 watt-hours for a median Gemini text prompt. At 30 prompts a day, the two figures come to 2.63 to 3.72 kilowatt-hours a year, or about 0.025% to 0.036% of an average U.S. home's yearly electricity. In our calculation, we multiplied the prompts by each company's published per-prompt figure and compared the yearly total with an average U.S. home. We did not measure either figure, and the math is printed line by line in the calculator.

The calculator above runs this arithmetic. It runs in your browser, and nothing you enter is sent anywhere. It has two inputs, filled in this order:

1. Text prompts per day. This number box defaults to 10. Count each message you send.
2. Energy per prompt. This dropdown offers "OpenAI: 0.34 Wh per average ChatGPT query" (the default) or "Google: 0.24 Wh per median Gemini text prompt."

The results appear in three boxes:

• Electricity used shows watt-hours a day.
• Over a year shows kilowatt-hours a year.
• Share shows the percentage of an average U.S. home's yearly use.

The result sentence also gives the minutes of a 100-watt TV a day and the kilograms of carbon dioxide equivalent (CO2e) a year on the average U.S. grid, with the step-by-step math beneath it. The Copy result button copies the sentence, the math, the sources and a link. The Copy share link button copies a link that reopens the same inputs.

Where the Two Per-Prompt Figures Come From

The OpenAI figure comes from Sam Altman's blog post "The Gentle Singularity" from June 2025. In the post, Altman writes that "the average query uses about 0.34 watt-hours." The post does not say how the figure was measured or what it includes.

The Google figure comes from its August 21, 2025 post, "Measuring the environmental impact of AI inference", which comes with a technical paper on arXiv. Google's figure uses May 2025 data for the median Gemini Apps text prompt. Google says it measured this "comprehensively," including idle machines and data center overhead.

The two figures differ by about 40%. They measure different products, an average ChatGPT query and a median Gemini text prompt. Only Google has published its method.

The Formula and Worked Examples

The calculator runs these steps. You can repeat them by hand with the same inputs.

1. Prompts a day × watt-hours per prompt = watt-hours (Wh) a day.
2. Wh a day × 365 ÷ 1,000 = kilowatt-hours (kWh) a year.
3. kWh a year ÷ 10,356 kWh = share of an average U.S. home's yearly electricity. The 10,356 kWh is 863 kWh a month × 12, from the U.S. Energy Information Administration (EIA) Table 5A figure for 2024.
4. Wh a day ÷ 100 watts × 60 = minutes of a 100-watt TV a day.
5. (kWh a year ÷ 1,000) × 770.9 pounds of CO2e per megawatt-hour × 0.45359237 kilograms per pound = kilograms of CO2e a year. The 770.9 figure is the U.S. average from the U.S. Environmental Protection Agency (EPA) eGRID summary tables.

These are the calculator's outputs for common inputs:

• 10 prompts a day, OpenAI figure: 3.4 Wh a day; 1.24 kWh a year; 0.012% of a U.S. home; about 2.0 minutes of a 100-watt TV a day; about 0.43 kg of CO2e a year.
• 10 prompts a day, Google figure: 2.4 Wh a day; 0.88 kWh a year; 0.008% of a home.
• 30 prompts a day, OpenAI figure: 10.2 Wh a day; 3.72 kWh a year; 0.036% of a home; about 6.1 minutes of TV a day; about 1.30 kg of CO2e a year.
• 30 prompts a day, Google figure: 7.2 Wh a day; 2.63 kWh a year; 0.025% of a home; about 4.3 minutes of TV a day; about 0.92 kg of CO2e a year.
• 100 prompts a day, OpenAI figure: 34 Wh a day; 12.41 kWh a year; 0.12% of a home; about 20.4 minutes of TV a day; about 4.34 kg of CO2e a year.
• 100 prompts a day, Google figure: 24 Wh a day; 8.76 kWh a year; 0.085% of a home.

What the Figures Leave Out

Google's per-prompt figure describes a text prompt. Neither company has published a per-prompt figure for image generation, voice, or long "reasoning" or deep-research answers. Image, voice and long reasoning requests can use more energy than an average or median text prompt, but neither company says how much more, so the calculator has no setting for them.

The calculator also does not estimate training. It counts only the per-prompt figure each company gave. Google's paper says its measurement includes idle machines and data center overhead, while OpenAI's post does not say what its figure includes.

The carbon line uses the U.S. average grid figure because a prompt can be answered in a data center in another state.

Putting the Number in Context

For one person, the per-prompt figures are small. At 10 prompts a day, the OpenAI figure comes to 1.24 kWh a year, or 0.012% of an average U.S. home's yearly electricity. Even at 100 prompts a day on the OpenAI figure, the yearly total is 12.41 kWh.

Google compares its median Gemini text prompt with TV. The energy equals watching TV for less than nine seconds, based on a modern TV using about 100 watts. The calculator puts the OpenAI figure at about 6.1 minutes of that TV for 30 prompts a day.

The large electricity numbers come from whole data centers. A 100-megawatt (MW) facility running at full capacity uses 876,000 megawatt-hours (MWh) a year, as the data center power calculator shows. See the data center emissions calculator for the carbon side.

Water is a separate question. Altman's post gives about 0.000085 gallons of water per query, roughly one fifteenth of a teaspoon. Google's paper puts the median Gemini prompt at 0.26 milliliters, about five drops, and says its figures are far lower than earlier outside estimates of 45 to 50 milliliters per prompt. The AI water calculator works out the water for your own prompts.

At 10 to 30 prompts a day, the two figures put a year of text prompts at 0.88 to 3.72 kilowatt-hours, well under 1% of an average U.S. home's electricity. Your total rises with heavy use and with images or long reasoning answers, and neither company has published a figure for those. The answer would change if OpenAI or Google published per-image or per-model figures, or new per-prompt numbers. Enter your own daily prompt count in the calculator above, pick the figure you want to use, and copy the result with its math and sources.