Done getting your 80s photos ? Now comes the less glamorous part: What does it take to create them?
The scale is striking. Inference — the day-to-day use of AI models — accounts for roughly 80% to 90% of their total energy demand, according to the United Nations University. A popular AI platform like ChatGPT is estimated to process around 2.5 billion prompts per day, consuming hundreds of gigawatt-hours of electricity each year.
Model choice, prompt length, output format and resolution all materially shape the footprint. Yet most of these decisions are made invisibly through product defaults that users never see. A study published in Water Research, titled ‘The water footprint of artificial intelligence: Emerging solutions and governance imperatives’, notes that AI infrastructure consumes freshwater through evaporative cooling, indirect water use in electricity generation and water-intensive semiconductor manufacturing. The United Nations University has put the footprint of individual AI tasks into more relatable terms.
The study projects that AI’s global water footprint could reach 4.2–6.6 billion cubic metres annually by 2027. It also notes that two-thirds of post-2022 data centres are located in water-stressed regions. The electricity required to generate a typical AI image is enough to power a 10-watt LED bulb for about 17 minutes. For a high-complexity AI video, the equivalent energy could power the same bulb for 42 hours.
