Does Every AI Image Really Cost Gallons of Water?

A striking number has been making the rounds for months: every image an AI model generates supposedly consumes around ten gallons of water. It appears in social posts, opinion columns and the occasional headline, offered as proof that generative AI is quietly draining the planet one picture at a time. It is a memorable figure. It is also wrong by a spectacular margin.

No peer-reviewed study has ever measured ten gallons of water per generated image. When researchers work backward from the electricity a single image actually uses, roughly 0.003 to 0.012 kilowatt-hours, the associated water works out to something like 15 to 60 milliliters. That is a few tablespoons, not a bathtub. The viral claim overstates the real number by a factor of somewhere between one hundred and six hundred.

Where the numbers actually land

Text prompts are smaller still. Sam Altman has said the average ChatGPT query uses about 0.3 milliliters of water. Google, disclosing its own figures, put the median Gemini text prompt at 0.26 milliliters. These are drops, in the literal sense. A single search or chatbot reply does not carry a hidden reservoir behind it, whatever the more alarming posts suggest.

None of this means AI has no environmental cost. Data centers do use water for cooling, and the aggregate across billions of queries is real and worth tracking. The mistake is in the per-image figure, which took a genuine concern and inflated it into something the evidence cannot support.

The question worth asking instead

Researchers who study the field make a further point that gets lost in the water panic. Where the electricity comes from matters far more than how much water a query uses. A data center running on a coal-heavy grid does considerably more environmental harm than one drawing on renewable power, and that difference dwarfs the water arithmetic entirely.

The honest version is less shareable than the ten-gallon line. AI's footprint is real, it is dominated by how its electricity is generated, and it is best measured in aggregate rather than dramatized per picture. The water myth persists because a big round number feels like an argument. It is worth remembering that a made-up figure, even one pointed at a real problem, is still made up.