How much water does AI use per day? The daily numbers explained simply (2026)

Picture a single day of chatbot traffic: hundreds of millions of people asking for recipes, code fixes and email rewrites, one prompt after another. So how much water does AI use per day? For one person, very little: a typical day of text prompting uses somewhere between a few millilitres and a few litres, depending on whose estimate you trust. For ChatGPT alone, our worked calculation puts the daily total at roughly 0.8 million to over 100 million litres, and all the world's data centers together use about 12 billion litres a day.

That is a wide range, and we think the range itself is the story. This page sticks to the daily view: how the per-day numbers are built, what they look like for a whole service and for one user, and where the guesswork creeps in. If you want the per-prompt studies and yearly totals in depth, our main guide to how much water AI uses covers them.

Short answer: how much water does AI use per day?

Globally, data centers (which run AI along with everything else online) consume around 12 billion litres of water a day, based on a UN University estimate of 4.5 trillion litres for 2025. ChatGPT's text prompts alone account for somewhere between about 800,000 and 112 million litres a day, depending on the method. One heavy user's daily share is a few litres at most.

How it works: where the daily numbers come from

Almost nobody measures AI water use by the day. Daily figures are built in one of two ways.

The first is top-down: take a yearly total for data centers and divide by 365. It is grounded in real reporting, but mixes AI with streaming, cloud storage and everything else.

The second is bottom-up: multiply the number of prompts a service handles each day by an estimate of water per prompt. This isolates AI, but inherits the weaknesses of per-prompt estimates, which differ by more than a hundredfold.

Why so different? Some figures count only water evaporated to cool chips on site; others add the water power plants use to make the electricity, often the larger share. Our explainer on why AI uses water covers the cooling side.

Real examples: AI water use per day, worked out

The global daily picture

Start with the big totals. The UN University Institute for Water, Environment and Health says data centers consumed 4.5 trillion litres in 2025 and could reach about 9.3 trillion by 2030. That is roughly 12.3 billion litres a day today, rising towards 25 billion a day. For AI specifically, Morgan Stanley's estimate of about 1,068 billion litres a year by 2028 (cooling plus electricity) works out to about 2.9 billion litres a day.

In the US, Lawrence Berkeley National Laboratory figures cited by researchers writing in The Conversation put 2023 direct cooling water at 17 billion gallons and indirect water from electricity at about 211 billion gallons. Per day, that is roughly 176 million litres on site and about 2.2 billion litres at power plants. The daily ratio is the same as the yearly one: the water you can't see is around 12 times the water you can.

One service, one day: a ChatGPT calculation

Now the bottom-up version. OpenAI told Axios in July 2025 that ChatGPT users send more than 2.5 billion prompts a day, with more than 330 million of those in the US. It is the latest official daily count we found, and usage has likely grown since, so treat the results below as a floor.

Multiply 2.5 billion by four published per-prompt figures and you get this original comparison:

The calculations are ours, not the companies'. Google's figure is for Gemini, not ChatGPT, and is included only to show what an on-site-only method gives at the same volume.

To give those numbers a shape, an Olympic swimming pool holds about 2.5 million litres. On Altman's figure, a day of ChatGPT uses about a third of one pool. On Mistral's method, it is about 45 pools. For another reference point, Google's planned data center site in Botetourt County, Virginia, is contracted for 2 million gallons a day, about 7.6 million litres. On the low estimate, one large campus could use roughly ten times what all of ChatGPT's daily prompts use. On the high one, ChatGPT's day equals about 15 such campuses.

Scale keeps climbing on the Google side too. At I/O 2026, Google said its AI products were processing more than 3.2 quadrillion tokens a month, which is over 100 trillion tokens a day. Tokens don't convert neatly into water, but tiny per-prompt figures clearly add up.

One person, one day

Here is the same maths for a single user, using three made-up but realistic usage levels:

Even the heaviest case is less than half a bucket of water, which puts one person's daily share in proportion.

Benefits and limitations of per-day estimates

Daily figures match how people think. "A third of a swimming pool a day" is easier to picture than "0.32 millilitres per query", and easier to compare with a town's water supply.

But we should be honest about what these numbers can't tell us. First, the prompt counts are snapshots: the 2.5 billion figure is over a year old. Second, all four per-prompt estimates are for text. Image generation, video and long reasoning modes use far more compute, and none of the daily totals above include them. Third, Google's own paper leaves out training and power-plant water, while Mistral's audit includes both, so they are not like-for-like. Fourth, a national or global daily average hides location. A million litres a day matters far more drawn from an aquifer in Arizona than from a wet region with plenty of rain.

Separating what we know from what we are guessing, then: we know the order of magnitude of global data center water use, and we know ChatGPT handles billions of prompts a day. We are guessing about the exact water per prompt, today's real prompt count, and how much of the global total is AI rather than other computing.

What it means for you

If your question is personal, the evidence suggests your daily AI water footprint is small, from a few drops to a few litres even if you use AI constantly. A few habits keep it lower and save you time:

• Ask once, clearly. Fewer retries mean fewer prompts. Our guide to writing better AI prompts helps.
• Pick the lighter mode. Use quick text answers when you don't need deep reasoning, images or video.
• Look beyond your own use. The bigger questions are about where large AI data centers are built and where their water comes from.

For energy and emissions as well as water, see our broader look at whether AI is bad for the environment.

What happens next

Two forces pull against each other. Per-prompt efficiency is improving fast, and companies now publish more data than they did a couple of years ago. At the same time, the daily volume of prompts and tokens keeps rising, and the Morgan Stanley projection implies AI's daily water use could grow about elevenfold between 2024 and 2028.

Where do we land? The daily figure will probably keep rising for a while, even as each prompt gets cheaper. The most useful change would be regular official usage counts and site-level water data, so calculations like ours could rest on audited numbers.

FAQ

Does AI use a lot of water?

Per prompt, no: estimates range from about 0.26 mL to 50 mL for a text answer. In total, yes. A single day of ChatGPT prompts may use anywhere from about 800,000 to over 100 million litres, and all data centers together consume around 12 billion litres a day, based on UN University figures.

Why does AI use so much water?

AI chips turn electricity into heat, and many data centers remove that heat by evaporating water in cooling towers. Power plants supplying the electricity also evaporate water, often more than the data center itself. Multiply that by billions of prompts a day and the small per-prompt amounts become very large daily totals.

How much water is used for AI?

No one publishes an exact AI-only figure. Morgan Stanley estimates AI data centers could use about 1,068 billion litres a year by 2028, around 2.9 billion litres a day, counting both cooling and electricity. All data centers, which also run non-AI services, used about 4.5 trillion litres in 2025.

How much water does AI use per prompt?

It depends on what is counted. Google says a median Gemini text prompt uses 0.26 mL, and Sam Altman put an average ChatGPT query at about 0.32 mL. Methods that include power-plant water, such as Mistral's audited study (45 mL) and the Li and Ren research (10 to 50 mL), give far higher figures.

How much water is AI using?

Right now, roughly: ChatGPT's daily text prompts use somewhere between about 0.8 million and 112 million litres a day on our calculation, and AI's share of data center water is growing quickly. Google alone reported processing over 3.2 quadrillion tokens a month in 2026, which shows how fast overall AI use is climbing.