The AI ​​photo trend of the 1980s may never be remembered as a major environmental event, because we only see retro photos. It's all about processors, servers, electricity, cooling equipment, and in many cases, water.

How much electricity does one AI image consume?

These days, a trend of AI photos featuring a retro 1980s look is gaining momentum on social media. People are submitting their selfies to AI tools and transforming them into portraits with big hair, vintage clothing, old studio backdrops, and film-like textures. The process seems simple: upload a photo, type in a prompt, and the image is ready in seconds. The physical infrastructure behind this process consumes electricity, generates heat, and consumes water to cool down...

How much electricity does one AI image consume?

There's no fixed number. Dr. Srinivas Padmanabhuni, CTO of AiEnsured, estimates that generating a single AI image can consume approximately 1.2 watt-hours of electricity. Ritwik Batabyal, CTO of MaskTech Group, says this number can range from fractions of a watt-hour to several watt-hours, depending on the model and deployment.

Jaspreet Bindra, co-founder of AI&Beyond, estimates the energy footprint at 0.1 to 4 watt-hours per image. This variation is due to the model size, hardware, and workload affecting the energy footprint.

According to the International Energy Agency (IEA), image generation is much more energy-intensive than simple text-based tasks. A 2026 UN university assessment estimated that a typical AI-generated image requires approximately 1,450 times more energy than a basic text-classification task.

The real problem is not the image but the scale.

A few watt-hours is a small amount for the average person, but the problem begins when it becomes a mass activity. People don't usually create the same image. The first version is rejected because the face doesn't look right, the second prompt changes the clothing, and the third changes the lighting. After creating several versions, the final image is shared.

A viral trend therefore translates into millions of inference tasks. The United Nations University (UNU) estimates that text-to-image systems will generate over 15 billion images in 2022 and 2023. In 2023, approximately 34 million AI images will be created daily. There is no reliable public data on the number of images generated during the current 1980s trend, so estimating its total energy or water footprint would be speculative.

The cost of water: From two teaspoons to a major crisis

Electricity is only half the story. Data centers require water to cool them and to generate electricity. According to UNU's 2026 assessment, by 2030, data centers running AI could require 945 terawatt-hours of electricity annually.

The associated water footprint could reach 9.3 trillion liters, equivalent to the annual domestic water needs of 1.3 billion people in sub-Saharan Africa. The land footprint could exceed 14,500 square kilometers.

What is the water footprint of a single AI image?

UNU estimates the electricity-related water footprint of a single AI image to be approximately 29 milliliters, or about two teaspoons. This number depends on the data center's location, cooling technology, local climate, power mix, and infrastructure efficiency.

According to the IEA, it takes about 23 milliliters of water to create a single AI image. This may seem small for a single image, but when millions of images are created daily, the annual cost could be equivalent to filling 200 Olympic-sized swimming pools.

Besides energy and water, where does the selfie data go?

There's another less-discussed aspect of this trend: what happens to your photos. To achieve the retro look, you have to give the AI ​​tool a clean, well-lit selfie, or sometimes multiple photos.

Security researchers have been warning for months that facial geometry, skin texture, and even fingerprints in close-up shots can be extracted from high-resolution selfies. Unlike passwords, faces cannot be reset.

The biggest concern is that many platforms keep uploaded images. Very few users read the terms and conditions to see if their photos can be used to train the next model.

The regulation on this is very weak; that is, except for some specific laws, there are no strict general rules that prevent the company from reusing the selfie uploaded by you for fun.

The AI ​​boom isn't just a software story.

With the rise of AI, demand for data center capacity is also rapidly increasing. According to the IEA's April 2026 projections, global data center power consumption grew rapidly in 2025 and is projected to grow even faster with the expansion of AI. Capital expenditures by the five largest tech companies exceeded $400 billion in 2025 and are expected to increase by another 75% in 2026.

According to the IEA report, electricity consumption from data centers could increase from 485 TWh in 2025 to 950 TWh by 2030, representing approximately 3% of global electricity demand. Electricity consumption from AI-focused data centers increased by 50% in 2025.

UNU estimates that 80 to 90 percent of AI energy use occurs during inference, i.e., when deployed models are actually being used, not during training. Approximately 2.5 billion ChatGPT prompts are executed daily, demonstrating how everyday AI use translates into a significant infrastructure load.

So should AI stop creating images?

The environmental footprint of a single AI-generated image is significantly smaller than many everyday emissions and resource use. The real question is what happens when AI image generation becomes so easy that people create billions of images. This is where efficiency gains become crucial.

AI models can be made more efficient, hardware can be improved, data centers can use different cooling technologies, and operators can choose where to build facilities and which power sources to use.

Ritwik Batabyal of MasTech Group says, “Generative AI has made image creation extremely easy, but we shouldn’t mistake digital convenience for zero physical cost.” According to Ritwik, the efficiency of the entire AI stack needs to be continuously improved.