The Invisible Cost of AI

The Invisible Cost of AI: Visualizing Water Behind Intelligence is a speculative documentary photo book that makes the water footprint of generative AI visible.

The work follows one person through a single ordinary day — twenty AI tasks, from a morning weather check to a night of video rendering. Each task is given two photographs: one of the act, one of the water it consumed. A running total carries forward across the series, measured in graduated cylinders and, by nightfall, in counted one-litre bottles.

AI’s environmental cost stays invisible not because it is concealed, but because at the scale of a single interaction there is almost nothing to see. Four millilitres is unremarkable; forty-one litres is not. Only accumulation makes the cost legible.

Estimates of AI water use vary widely depending on where the system boundary is drawn. The book presents its figures as reference points rather than measurements, treating that uncertainty as part of what remains invisible.

On Measuring What Will Not Hold Still

I began this project expecting to illustrate a number. I ended up photographing a problem I could not measure.

The first difficulty was that the data would not hold still. Estimates of water per AI query range across orders of magnitude, and the disagreement is not really about technology — it is about where measurement begins. Cooling water only? The electricity behind it? Training, chips, construction? Each boundary is defensible and produces a different book. I decided to state this openly rather than choose a convenient figure, because hiding it would have reproduced exactly the opacity I was criticising.

The second difficulty was scale. Four millilitres cannot be photographed dramatically without lying about it. So I stopped trying to make a single quantity impressive and built the series around accumulation instead.

The third was uncomfortable: I generated these images with the systems I am questioning. I have not resolved that contradiction. I have left it visible.