Highlights
- One AI image can use up to 4.08 Wh of electricity.
- That can run a 10W LED bulb for about 25 minutes.
- Millions of images can add up to a bigger environmental cost.
That viral retro selfie may take only seconds to create, but it does not appear from nowhere. It takes computing power, electricity and cooling to produce it.
A 2025 study, The Hidden Cost of an Image: Quantifying the Energy Consumption of AI Image Generation, tested 17 AI image-generation models and found that energy use varied by as much as 46 times between them. The study measured between 0.086 and 4.08 watt-hours per image, depending on the model and conditions tested.
So what does one image actually cost?
The numbers may not mean much on their own. Put simply, the highest figure of 4.08 watt-hours is roughly enough electricity to keep a 10-watt LED bulb on for about 25 minutes.
That is not the study saying your selfie can power a bulb. It is an everyday comparison to show what the energy figure means.
You could also think of it in terms of charging a phone, running a laptop or keeping a light switched on. The exact comparison depends on the device and its power use, but the point is simple: an AI image uses real electricity, even if all we see is a picture appearing on our screen.
What happens when the trend goes viral?
One image does not sound like much. But a viral trend is never about one image.
People generate a picture, try another version, change the outfit, change the background and generate another. Then millions of other people do the same thing.
That is where the small numbers become more important. The 2025 study found major differences between image-generation models, showing that the amount of energy behind each image can vary greatly.
The International Energy Agency also says electricity use by data centres is rising rapidly, with AI-focused data centres growing particularly quickly.
The hidden water cost
There is another part of the picture we cannot see.
The computers that generate AI images run in data centres, which need cooling to keep their equipment working. The IEA says cooling can account for around 7 per cent of electricity use in efficient hyperscale data centres and more than 30 per cent in less-efficient facilities. Some cooling systems also use water.
None of this means we should stop using AI. The technology can do important work in research, medicine, education and many other areas.
But when powerful computing is being used to create millions of pictures simply because a retro selfie trend is popular for a few days, it is worth thinking about what sits behind that seemingly harmless image.
The picture may be digital. The resources used to make it are not.
