Prompts → Water

Each AI prompt uses about 0.32 ml of cooling water. What does yours add up to?

Based on OpenAI's reported average of ≈ 0.32 ml per prompt.

Total water

0 ml

Type a number of prompts to see what it would fill.

How far up the ladder are you?

    The honest bit: prompts aren't the thirsty part

    Your prompts are a rounding error. At ≈ 0.32 ml each, even a year of heavy daily use adds up to roughly one glass of water. The serious water cost of AI is training the models and running the data centres that host them.

    Researchers at UC Riverside estimated that training GPT-3 alone consumed around 700,000 litres of clean, fresh water for data-centre cooling — about the same as filling a 25-metre swimming pool, or one person's drinking water for nearly 1,000 years. Newer, larger models are thirstier still, and the same study projected that global AI demand could withdraw 4–6 billion cubic metres of water a year by 2027.

    So the fair takeaway isn't "stop prompting" — once a model is trained, your individual prompt barely registers. It's that the build-out of AI infrastructure has a real and growing water footprint, especially in regions where data centres compete with people for fresh water.

    This calculator is an illustrative estimate based on 0.32 ml of cooling water per prompt. Actual water use varies with model choice, region, cooling methods, energy mix, and upstream supply chain impacts (including building and operating data centres), so treat this as a rough guide rather than a precise footprint.

    FAQ