Skip to content
Where does this figure come from?Uses20 October 2026

"A generated video uses 1 kWh"

Guillaume PakulaClément ReynaudSébastien PierfedericiAlexandre Veilly By Projet Celsius
Published on 20 October 2026 · 3 min read
An Nvidia server tray with 8 GPUs, pulled out of its chassis and placed on a table

The figure has circulated since 2025, often attached to Sora. It comes from a real measurement, but one taken on a different, open model, in a particularly heavy version. Measurements of 7 generators range from 0.14 to 415 Wh.

Contents
1.

Where the figure comes from

In May 2025, MIT Technology Review measured an open video generator, CogVideoX: about 30 Wh for 5 seconds with the older version of the model, 944 Wh with the new one; the graphics card was measured and the result doubled to allow for the rest of the server. Blogs turned this into "1 kWh per Sora video". Neither OpenAI for Sora, nor Google for Veo, nor Kuaishou for Kling has published a measurement.

2.

Researchers' measurements

In September 2025, Julien Delavande, Régis Pierrard and Sasha Luccioni measured 7 open generators with their default settings. The most economical uses 0.14 Wh, the most power-hungry more than 415 Wh, a factor of 3,000. The most downloaded open generator, WAN 2.1 in its 1.3-billion-parameter version, uses 90 Wh for a 5.4-second video in 720p. The International Energy Agency (IEA), for its part, takes 115 Wh for 6 seconds, GPU only.

For the same model, CogVideoX in its 5-billion-parameter version, the 2 studies give about 25 Wh and 944 Wh. The gap is not explained: settings, number of steps and hardware differ, and none of the 2 publishes everything that would be needed to settle it.

AI Impact, by Projet CelsiusUses

Electricity used by a generated video, 7 models measured

Electricity per video, default settings, in Wh, logarithmic scale · Measurements published in September 2025

AnimateDiff1.6 s, 512p0.14Wh
LTX-Video5 s, 512p3.7Wh
CogVideoX 2B6.1 s, 480p9.7Wh
CogVideoX 5B6.1 s, 480p25Wh
Mochi 12.8 s, 480p52Wh
WAN 2.1 1.3 Md5.4 s, 720p90Wh
WAN 2.1 14 Md5.4 s, 720pmore than 415Wh
0.11101001,000

Source: Delavande, Pierrard and Luccioni, September 2025 · projetcelsius.com/en/ai · CC BY 4.0

3.

Why the gaps are so large

The energy of a video grows as the square of its duration and as the square of its number of pixels: doubling the duration multiplies it by about 4, moving from 720p to 1080p by 5. By extrapolation, 10 seconds in 1080p with the reference model would exceed 1.5 kWh, but the relationship has been verified only up to 6.6 seconds. A kilowatt-hour is therefore not absurd for a long high-definition video; it is absurd for "a video" in general.

4.

Key points

The hub takes 90 Wh for a 5-second video, naming the model. That is 300 written questions, 31 images, and 41 grams of CO2e with the world average electricity mix, the equivalent of about 290 metres in a petrol or diesel car. Over a year of use, one video a week weighs more than 10 questions a day.

Electricity used by a generated 5-second video90Wh

Most downloaded open generator · 5.4 seconds in 720p · measured in September 2025

Electricity used by a generated 5-second video→
Point to note

A figure for a video means nothing without the model name, the duration and the resolution: depending on these 3 parameters, it varies from 1 to more than 3,000.

Guillaume PakulaClément ReynaudSébastien PierfedericiAlexandre Veilly
By Projet CelsiusAI Impact, Projet Celsius · 20 October 2026
Write to us →

Sources

3 documents

Every figure was read in its source; the hub's methodology is described in Methodology and sources.

Cite this analysis

AI Impact, Projet Celsius, "A generated video uses 1 kWh", 20 October 2026. https://projetcelsius.com/en/ai/analyses/video-equals-one-kilowatt-hour/

Charts and maps

A question about a figure?

An error, a source to add, a methodology question, a press request: write to us. Guillaume Pakula, Clément Reynaud, Sébastien Pierfederici and Alexandre Veilly will reply.

Write to us →

Projet Celsius, which runs this hub, also helps companies measure and reduce the footprint of their AI use.

Accounting for AI in your Bilan Carbone® ↗
Projet Celsius, environmental strategy consultancy

We support companies that account for AI in their Bilan Carbone®

Visiter Projet Celsius ↗