Estimated water consumed by AI today: … liters(estimated, and counting · rate reviewed 2026-09)
Estimate the water, energy and carbon behind your AI use, with sources for every number.
How every number is madeType it the way you'd say it. The dials below follow.
frontier class · 2 Wh per 1,000 tokens (mid benchmark)
The same prompt can be “five drops” or “a bottle of water” depending on what you count: just data-center cooling (Low), the water behind the electricity (Mid), or training and hardware too (High).
Read the methodology16.9gallons
= 63.9 liters
ChatGPT (standard), 25 prompts/day · per year, Mid scenario
Energy
12.8 kWh
Carbon
4.5 kg CO₂e
The boundary problem
The same exchange has been reported as “five drops” and as “a shot glass” of water. Both can be defended: they draw the boundary in different places. We show all three, and every number on the site says which one it uses.
How the scenarios workOne exchange, about 700 tokens
0.42mL
Low: cooling only
Water evaporated to cool the data center while it answers.
7mL
Mid: + electricity
Plus the water consumed generating that electricity. Our default.
42mL
High: full lifecycle
Plus a share of training, chips and construction: the full lifecycle.
Drop area is proportional to volume. The energy estimate is the same in all three; only what gets counted changes.
Orders of magnitude
Per-prompt water is measured in milliliters; fleets, in billions of liters. Each tick on this ruler is ten times the one above it, so the gaps between the markers are the story.
Your personal use and the industry's footprint are different questions. Both are real, and neither answers the other.
7 mLOne ChatGPT prompt
One exchange, Mid scenario
64 LA year of daily ChatGPT use
25 prompts a day at Mid: about one shower (65 L)
2,500 LThe water behind one hamburger
Feed, farming and processing, full lifecycle
2.5M LAn Olympic swimming pool
The unit headlines reach for
64.4B LEvery US data center, 2023
17 billion gallons consumed on site, before counting electricity
Instruments
Estimated intensity per 1,000 tokens at the Mid scenario. Reasoning models rank highest: they think at length, out of sight.
mL per 1,000 tokens, Mid scenario
Full model ranking
Per prompt it's milliliters; per campus it's millions of gallons a day. Ten states hold most of the tracked capacity.
17billion galUS data centers, direct use, 2023
Open the state map
Almonds, burgers, searches, streaming: the same coefficients as the calculator, so the comparisons can't drift.
≈ 357KChatGPT prompts at Mid use as much water as one burger
All comparisons
Every estimate is the same chain. The water factor is the boundary you choose.
Read the methodology
The calculator's numbers as JSON. CORS-open, no key.
GET /api/v1/estimate
?model=claude-sonnet
&prompts=25&period=yearAPI docs
One iframe for your article. No tracking inside it.
<iframe
src="aiwateruse.org/embed/frame"
loading="lazy"></iframe>Get the snippet
Receipts
Each figure on this site traces to a public source registry: versioned, dated and reviewed quarterly. Where studies disagree, we publish the disagreement as a range instead of picking the most quotable end.
32sources in the registry
v1.3.1estimates version, sources reviewed 2026-09
By evidence tier
Latest change · v1.3.1 · 2026-09-26
Corrected Google’s water figure: 6.1B gallons was its data centers’ consumption in 2023. For 2024 Google reported ~7.7B gallons for its data centers (~8.1B company-wide). Council Bluffs, Iowa is now ~1B gallons (2024), and the ~10,000-gallon Texas site is Pflugerville.
Cited on this page, in order of appearance.