How much water does AI use per day?
Globally, AI queries consume an estimated 20 million liters of water per day at our Mid scenario: about 230 liters every second, or eight Olympic pools a day. That covers chat queries only; data centers as a whole (AI and everything else) consumed about 46 million gallons per day directly in the US alone in 2023.
The home-page ticker runs on exactly this estimate. Here is the arithmetic, so you can argue with it:
The ticker math, shown in full
- OpenAI disclosed roughly 2.5 billion ChatGPT queries per day in 2025.
- We scale to ~5 billion AI queries/day across all assistants, a deliberately round estimate for Gemini, Claude, Copilot, Meta AI, DeepSeek and the rest combined.
- We charge each query ~0.8 Wh, a blended site assumption: a ~700-token exchange comes to ~0.56 Wh on a standard-class model and ~1.4 Wh on a frontier-class one, while operators' own per-prompt disclosures are lower (0.24–0.34 Wh).
- At Mid (cooling + the water behind electricity, ~5 L/kWh): ~4 mL per query.
- 5B × 4 mL ≈ 20 million liters/day ≈ 230 L/second.
Every step is an estimate, and the label says so. Step 2 is the softest; if you think total volume is 3B or 8B queries, scale accordingly; the answer stays "millions of liters, not billions."
ChatGPT's day, three ways
ChatGPT is the one service with both a public query volume and a public per-query figure, so its daily total can be computed rather than guessed, once per boundary:
- Low (cooling only): 2.5B queries × 0.32 mL (Altman, 2025) ≈ 800,000 L/day: a third of an Olympic pool, for the most-used AI product on Earth.
- Mid (+ electricity): between Altman's 0.34 Wh and our frontier-class ~1.4 Wh per exchange, 1.7–7 mL per query ≈ 4–18 million L/day: two to seven pools.
- High (full lifecycle): the same two energy figures at the High factor (30 L/kWh, derived from Mistral's LCA), 10–42 mL per query ≈ 25–105 million L/day: ten to forty pools.
The boundary moves the answer by two orders of magnitude: the single most important thing to know about any AI-water number.
Per day, per person
Your personal day is smaller than the ticker makes it feel. Twenty-five ChatGPT prompts (the calculator's Daily preset) come to about 0.18 liters at Mid: about a third of a water bottle. Your morning coffee embodied roughly 750 times that before you opened a chatbot.
The same day, across all three boundaries:
| Scenario | Your day (25 prompts) | Looks like |
|---|---|---|
| Low (cooling only) | ~10 mL | a couple of teaspoons |
| Mid (+ electricity) | ~0.18 L | a third of a water bottle |
| High (full lifecycle) | ~1 L | a large water bottle |
Run High for a full year and the total is roughly 380 liters: a little over one day of your home plumbing (EPA: 82 gallons, about 310 liters, per person per day). One skipped hamburger (~2,500 L embedded, Water Footprint Network) buys more than six years of this daily habit at full-lifecycle accounting.
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
When "per day" is the right question
Daily aggregates matter where the water is drawn: a single hyperscale campus can consume 1–5 million gallons on a hot day, which is the scale a municipal water utility actually plans around. That's the question our data center pages track, state by state, with xAI's Memphis cluster (~1M gal/day) being the current reference case.