How much electricity does making an AI picture or video use?
In lab tests on specific open models, one AI picture used about 0.63 to 1.22 Wh, a few short text questions’ worth (MIT Technology Review estimates). One short video clip used about 90 Wh on one open model (Hugging Face), about 5 phone charges, and about 940 Wh on another (MIT Technology Review). The models and settings differ, so they can’t be ranked. Apps like Sora and Veo don’t publish per-video figures.
The bigger pictureNobody publishes how many AI pictures and videos are made, or what they use in total. All the world’s data centres, of which AI is one part, used about 485 TWh in 2025, and the IEA expects about 950 TWh by 2030.
- 0.63 Wh One standard picture, Stable Diffusion 3 Medium: estimated by MIT Technology Review
- 1.22 Wh One higher-quality picture, same model: estimated by MIT Technology Review
- 90 Wh One short clip, WAN2.1–T2V–1.3B: measured by Hugging Face on one graphics chip
- 940 Wh One 5-second clip, the newer CogVideoX: estimated by MIT Technology Review
Unit: Wh per picture or clip.
How big is that?
- One short AI question0.34 WhOpenAI says 0.34 Wh; other published figures differ
- Charging a phone once19 WhUS EPA, about 19 Whabout 56× the line above
- One short AI video clip90 Whabout 90 Wh, measured by Hugging Faceabout 4.7× the line above
- Boiling a litre of water99 Whabout 99 Wh from 15 °C, by Hum’s sums from physicsabout the same as the line above
- A US home for a day28 kWhabout 28 kWh, Epoch AI from US dataabout 283× the line above
- A US home for a year11 MWhabout 10,500 kWh, Epoch AI from US dataabout 375× the line above
- AI in the world’s data centres for a year156 TWhabout 156 TWh by Hum’s sums: the middle of the US share of AI, 26.5 to 38%, used as a world estimateabout 14.9 million times the line above
- All the world’s data centres for a year485 TWhabout 485 TWh in 2025, IEAabout 3.1× the line above
The dots sit on a log scale: each step to the right is ten times bigger. It’s the only way a phone charge and all the world’s data centres fit on one line.
What nobody knows yet
- OpenAI, Google and the other companies behind popular picture and video apps haven’t published how much electricity each picture or clip uses.
- The lab figures are for specific open models and settings. Hugging Face measured the computer itself (graphics chip, processor and memory), not the data centre’s cooling, and MIT Technology Review doubled the chip’s energy to estimate the rest.
Sources (8)
- MIT Technology Review (one open image model, Stable Diffusion 3 Medium, standard 1024 x 1024 picture, estimated total) Estimated
“That means an estimated 2,282 joules total.”
Quote checked on the page, 25 Oct 2026 - MIT Technology Review (same open image model, higher quality, estimated total) Estimated
“Improving the image quality by doubling the number diffusion steps to 50 just about doubles the energy required, to about 4,402 joules.”
Quote checked on the page, 25 Oct 2026 - MIT Technology Review (one open video model, the newer CogVideoX, estimated total) Estimated
“The new model uses more than 30 times more energy on each 5-second video: about 3.4 million joules”
Quote checked on the page, 25 Oct 2026 - Delavande, Pierrard & Luccioni (Hugging Face), Video Killed the Energy Budget Measured
“generating a single short video with WAN2.1–T2V–1.3B consumes nearly ∼90 Wh”
Quote checked on the page, 3 Oct 2026 - OpenAI (Sam Altman; average query, method not published) Company says so
“the average query uses about 0.34 watt-hours, about what an oven would use in a little over one second”
Quote checked on the page, 2 Oct 2026 - US EPA Estimated
“[28.446 Wh – (22 hours x 0.411 Watts)] x 1 kWh/1,000 Wh = 0.019 kWh/smartphone charged”
Quote checked on the page, 18 Oct 2026 - NIST (the calorie: heating 1 g of water by 1 °C) Measured
“a calorie is 4.184 joules”
Quote checked on the page, 18 Oct 2026 - IEA, Key Questions on Energy and AI (2026) Estimated
“electricity consumption from data centres roughly doubling from 485 TWh in 2025 to 950 TWh in 2030”
Quote checked on the page, 3 Oct 2026
Hum’s sums
- 2,282 J ÷ 3,600 ≈ 0.63 Wh
- 4,402 J ÷ 3,600 ≈ 1.22 Wh
- 3.4 million J ÷ 3,600 ≈ 944 Wh, about 940 Wh
- 1.22 Wh ÷ 0.34 Wh ≈ 3.6, a few short questions at OpenAI’s figure
- 90 Wh ÷ 19 Wh ≈ 4.7, about 5 phone charges
- 0.019 kWh = 19 Wh
- 1,000 g × (100 − 15) °C × 4.184 J ÷ 3,600 ≈ 99 Wh to heat a litre of water from 15 °C to boiling
Cite this
MIT Technology Review (one open image model, Stable Diffusion 3 Medium, standard 1024 x 1024 picture, estimated total), as quoted in Hum, “How much electricity does making an AI picture or video use?”, https://humearth.org/q/pictures-and-video (quote checked 25 Oct 2026).
Cite the original source first; Hum is where you found it.
What do Wh, kWh and TWh mean?
Wh is a watt-hour: a 1-watt light left on for an hour. 1 kWh = 1,000 Wh, and 1 TWh = 1 billion kWh. Hum compares each number with something everyday so you don’t need the units.
Send this answer
One short AI video clip used about 90 Wh in a lab test on an open model (Hugging Face), about as much as boiling a litre of water. One AI picture: about 0.63 to 1.22 Wh (MIT Technology Review).