Skip to content

1 glass of beer = 246,667 AI prompts

1 glass of beer equals roughly 246,667 AI prompts on the water side: 74,000 mL, while an average text prompt weighs just 0.3 mL. The math, the source and detailed equivalents are right below, with public, verifiable numbers.

246,667
AI prompts

Figures updated on June 30, 2026 · Source : Water Footprint Network : cultures (Mekonnen & Hoekstra 2011)

The water of a prompt is what evaporates to cool the data center while it answers. For food and goods we count virtual water: all the water mobilised by production, not just what comes out of the tap. Both compare as orders of magnitude, which is exactly what this page shows.

Why 246,667 prompts?

The math fits in three lines, using public, sourced values:

  1. 1.1 glass of beer: 74,000 mL (Water Footprint Network : cultures (Mekonnen & Hoekstra 2011))
  2. 2.1 text prompt: 0.3 mL (published range: 0.26 to 0.32 mL)
  3. 3.74,000 mL ÷ 0.3 mL ≈ 246,667 prompts

In other words, “1 glass of beer” weighs as much as 246,667 questions asked to a generative AI. It is an order of magnitude, not a to-the-decimal measurement: sources publish ranges, and we use a median value, explained in our methodology.

What about a generated image or a reasoning prompt?

Not every AI use weighs the same. A generated image costs about 9 mL and a “reasoning” prompt about 20 mL, versus 0.3 mL for a simple text prompt. So “1 glass of beer” is also roughly 8,222 times one generated image, or 3,700 times one reasoning prompt.

Convert your own usage

Where do these numbers come from?

The value we use for “1 glass of beer” comes from “Water Footprint Network : cultures (Mekonnen & Hoekstra 2011)”, referenced like all our data on the Sources page. No number on this page is made up: every value can be traced back to a public document.

On the prompt side, Google and OpenAI report between 0.26 and 0.32 mL of direct cooling water per request; we use 0.3 mL. Counting the upstream water of electricity generation, the figure sits closer to 1.5 mL.

Also worth reading

ChatGPT vs Google search: energy use

Nearby equivalents