A volume is only part of the story.
Cooling water consumed and sprinkler water applied are different measures. Equal liters do not mean equal environmental impact.
AI & our water
AI uses water. But how much? Compare a year of daily text prompts with one session of watering the lawn.
Make the comparisonReal sources. Visible assumptions. Room for context.
A sprinkler running for 20 minutes
100 gallons at 5 gallons per minute
A tiny amount per prompt becomes easier to understand when you put it next to something familiar.
02 / Put the water side by side
Start with this example, then make it your own. Both volumes are shown in liters.
A year of AI
9.49 liters
36,500 text prompts over 365 days
One lawn-watering session
378.5 liters
20 minutes · 100 U.S. gallons
Same scale. Different time periods, as labeled.
In this example, a year of text prompts equals about 30 seconds of sprinkler water.The 20-minute lawn session applies about 39.9× that volume.
The default repeats Google’s May 2025 median Gemini Apps text-prompt estimate. It is an illustration, not a measurement of your usage. Try a higher value to see how much the assumption matters.
0.01–100 mL. Custom values are scenarios.
0.1–100 gal / min. Use your measured flow.
AI liters / year = prompts per day × 365 × mL per prompt ÷ 1,000
Lawn liters / session = minutes × gallons per minute × 3.785411784
Equivalent sprinkler seconds = AI liters ÷ (gallons per minute × 3.785411784) × 60
What’s being compared: estimated AI cooling water consumed and sprinkler water applied. This is a volume comparison, not a full environmental footprint. Read the sources
03 / Keep the whole picture
The lawn comparison gives us a sense of scale. Responsible AI use also means asking where the water comes from and what the work is worth.
Cooling water consumed and sprinkler water applied are different measures. Equal liters do not mean equal environmental impact.
Water availability, season, and cooling choices matter. A global benchmark cannot tell you what a particular watershed can support.
Give AI a clear task and ask for the output you need. For the lawn, account for rainfall and fix leaks. Small habits matter when many people share them.
A few fair questions
Yes. Some data centers consume water to cool the computers serving AI, and generating their electricity can also use water. The default benchmark here covers on-site cooling for text inference. It does not include training, hardware manufacturing, or water used to generate electricity.
No. It is Google's estimate for the median Gemini Apps text prompt in May 2025, using fleetwide water efficiency. This calculator repeats that benchmark to illustrate a volume. It does not measure your account or predict your actual annual use. Models, response lengths, reasoning, and locations vary; images and video are outside this example.
It makes a small per-prompt volume easier to picture. The two time periods are deliberately labeled: 365 days of text prompts and one sprinkler session. Change the inputs to explore a different scenario. The comparison describes water volumes, not equal environmental harm.
Twenty minutes is an example, not a watering recommendation. Your lawn's needs depend on rainfall, soil, season, and sprinkler coverage. Check local guidance and measure your own sprinkler's flow before using these numbers to plan watering.
04 / Follow the water back
Source links checked September 4, 2026.
The AI measurement is from May 2025.
0.26 mL of cooling water per median Gemini Apps text prompt. A provider-reported estimate from May 2025, published August 2025.
A single sprinkler at medium flow: about 5 U.S. gallons per minute. Actual equipment and water pressure vary.
Google’s measurement methodology, system boundaries, and limitations. Read this for the detail behind the shorthand.