As viral AI trends sweep social media platforms, a new study by the United Nations University Institute for Water, Environment and Health highlights the growing environmental cost behind quick-generated content, warning that everyday user prompts are driving an unprecedented surge in data center resource demands.

The "How would I look in the '80s?" trend has brought attention back to the power, water, and other resources used to create images in a matter of seconds by allowing users to alter a snapshot into 1980s-style hair, attire, and esthetics.

While one retro photo generated in seconds may seem negligible, it adds to a vast resource chain spanning electricity, water, land and hardware tied to billions of AI operations.

The Institute's "Environment and Health: Carbon, Water and Land Footprints," study showed the energy used to generate a typical AI image could power a 10-watt LED bulb for around 17 minutes.

The water footprint of that electricity is about 29 milliliters per image, roughly two tablespoons, but reaches 29 million liters across some 1 billion images.

Google Trends data showed searches for "80s trend" surged 5,000% last week over the previous week, while the related AI images were shared by millions on Instagram.

While generating an image requires significant electricity, it is only one facet of AI's environmental cost, which also encompasses power for data centers, server cooling, infrastructure, and hardware manufacturing.

Resource needs grow with scale

Global data centers consumed roughly 448 terawatt-hours of electricity in 2025, the study found.

International Energy Agency (IEA) expects this to rise to about 950 terawatt-hours by 2030, driven largely by AI's expansion.

By 2030, the water footprint tied to AI data centers' electricity use could reach 9.3 trillion liters, with a land footprint exceeding 14,500 square kilometers, the study projects.

Most of AI's energy use occurs after training. Processing users' daily requests accounts for roughly 80-90% of total AI energy consumption, the research found.

A typical AI image was calculated to require about 1,450 times more energy than basic text classification, consuming around 2.9 watt-hours.

A separate study by Hugging Face and Carnegie Mellon University researchers found that image generation consumes an average of about 2.9 kilowatt-hours per 1,000 prompts, with energy needs varying significantly by model and task, making a single fixed consumption figure per image misleading.

Hardware demands are also rising. AI-related electronic waste is projected to hit 2.5 million tons annually by 2030.