How much water does AI really use?

In April 2026, with Georgia in drought, people in Fayette County were asked to water their lawns only two days a week. Around the same time, a county letter came to light saying that a data centre being built nearby had used more than 100 million litres of county water without being billed.

In Santiago, Chile, in 2024, a court put a planned Google data centre on hold until its effect on the city’s aquifer was assessed again. That aquifer was already promised to more users than it could supply.

In The Dalles, Oregon, the city went to court to keep Google’s water use secret. It sued the newspaper that asked, and Google paid the legal bills. When the records came out, Google was taking about a quarter of the city’s water.

Put stories like these together and it sounds as if AI is drinking the world dry. Some people say every email a chatbot writes uses up a bottle of water. Data centres, the warehouses of computers that run AI along with the rest of the internet, are in the middle of each fight. But the three are very different: one is over very little water, one over a lot, and one over an aquifer with none to spare. To tell them apart, you need to know what a data centre does with water in the first place.

AI runs hot

Every answer a chatbot gives is worked out on computer chips in a data centre. The chips turn almost all the electricity they use into heat, and the heat has to go somewhere.

Many data centres get rid of it the way sweat cools skin: they let water evaporate. Evaporating carries a lot of heat away.

Doesn’t it just rain back down? It does, on average about nine days later and usually hundreds or thousands of kilometres away. Even a basin as big as the Mississippi’s gets back less than a third of the water that evaporates from it. So for the town, the water is gone: out of its river or aquifer, in the summer when it was taken, and an aquifer can take years, or far longer, to refill.

The electricity costs water too. Coal, gas and nuclear power stations turn water into steam to spin their turbines, and many cool that steam by evaporating more water. Wind and solar use almost none. Count both, and almost two-thirds of AI’s water is used up at power stations, wherever the power is made, which may be nowhere near the data centre.

Most data centres take their water from the local supply, the same rivers, reservoirs and wells as the town around them. At Google’s, about 81% of it is clean enough to drink.

Your share

How much of that is down to you?

The bottle came from a 2024 Washington Post graphic. It guessed that writing one email took hundreds of times more electricity than companies have since reported for their chatbots, and that one guess explains most of the gap.

The answer above is for an average conversation. The most common kind, a question or two, uses under a millilitre; the longer you chat, the more each reply costs, because the chatbot rereads everything said so far before it answers.

Video is the exception: a few seconds of it uses about half a glass. So is AI left to work through a big job on its own, as the end of this piece shows. For chatting, it’s a teaspoon here and there.

In mid-2025, ChatGPT’s weekly users sent about four messages a day on average. Over a whole year, that’s about a litre of water each, roughly the irrigation water behind the beans for one cup of coffee.

Now multiply it by everyone

But hundreds of millions of people chat with AI every week, and a teaspoon times a billion is not a teaspoon.

Everyone’s ChatGPT chats, in a day. All together, ChatGPT’s users sent about 2.6 billion messages a day. They used up roughly an Olympic swimming pool of water every day (somewhere between a third of a pool and two pools).

All of AI, in a year. By our rough estimate, ChatGPT’s messages come to less than 1% of AI’s water. The rest goes to other chatbots, to AI built into search, feeds, apps and businesses, and to training new models. Altogether, in 2025, AI used up about 127 billion litres.

Shared out across everyone on Earth, all of AI’s water comes to about 16 litres each a year, roughly about a week of drinking water. But it isn’t shared out: it lands in a few places.

Is that a lot?

Next to all the fresh water people use up in a year, AI’s cube is about one part in 8,000. Most of the rest goes to farming.

Is it worth the water? Some uses are easier to defend than others. Counting only water taken from rivers and aquifers and used up, the world’s beef cattle use about 280 times as much as AI, and they feed people. That doesn’t make AI’s water free: a bigger user somewhere else doesn’t refill a town’s aquifer. And there’s no fair way to price a litre of AI against a litre of food; money isn’t need.

Could it get bad enough to threaten food? Not worldwide. Even at the top of our range for 2030, built on the International Energy Agency’s highest forecast, AI would use about one litre in 1,200, and it would have to grow about 80 times to reach 1%. Locally, the question is who gives way in a drought. In Taiwan’s 2021 drought, the government stopped irrigating 74,000 hectares of rice to keep water flowing to homes and industry, chip makers among them. It had done the same in six earlier droughts since 2002.

A cube taller than the Eiffel Tower, and a speck beside the world’s farms: both are AI’s water, and most arguments pick one. What decides whether it matters is where it lands.

It depends where, and when

A data centre doesn’t use water a teaspoon at a time. It takes it all together, in one place, and those that cool with water take the most in summer, when everyone else needs it too. Whether that’s a problem depends on how much water the place has to spare. In none of the three places has a data centre yet caused a shortage, and they turn out very differently.

Fayette County was mostly a scare. The water was for construction, such as concrete and dust control, not for cooling computers, and it came to under 1% of what the county supplies. The lawn limits were a voluntary request under the state’s drought rules. The company paid the back bill at double the normal rate, and the county said its own letter had been “unclear”.

Santiago hit a real limit. The court agreed the aquifer was over-promised. Google has since dropped the original plan and says it will start again with a design cooled by air. No new plan has been approved.

The Dalles is where it bites. In 2024 Google took about 30% of the city’s water, for all its computing, not only AI, and it took more again in 2025. The city’s own plan projects a small shortfall by 2034, as demand from businesses grows. The squeeze comes in late summer, when demand peaks just as the city must leave water in its small main river.

In 2025 the city council in Tucson, Arizona, voted unanimously to end talks with a data centre that wanted city water. “We need to be the city that says not here, not now and not like this,” one council member said. It’s being built anyway, just outside the city, but now cooled by air and drawing on its own wells.

AI isn’t draining the world’s water. In towns like The Dalles, though, data centres take a share worth fighting over, and more are being built. For each new one, the question is whose water, and in which month.

What if this is wrong?

These numbers are estimates, and four things could move them.

• None of the big tech companies says how much water its AI uses in total. They report data centres as a whole, so AI’s share here is worked out from its share of their electricity.
• Much of the measuring is Google’s own. Its figures for the electricity a question takes haven’t been checked by anyone outside.
• Place matters a lot. The same conversation can use up to 70 times more water in Mumbai than in Stockholm, depending on the heat and on how the power is made, with the lakes behind dams left out.
• What counts as “using” water is a choice. This piece counts water that’s used up, as the researchers behind the bottle did. Count water that’s borrowed and put back as well, mostly by power stations, and the figures grow about 16 times.

All the checking for this piece was done by AI agents, not a human expert.

What this piece cost in water

I made this piece with a lot of help from AI: agents that read reports, checked claims, built the model and drafted the graphics, sometimes working on their own for an hour or more. Anthropic, which makes the AI I used, doesn’t publish how much energy its answers take, so this is a rough estimate: by the afternoon of 22 October 2026, making it had used up about 17 litres of water (somewhere between 2 and 150), about as much as 5,400 average conversations.

I can’t tell you where that water was taken from, or in which month. After all this, that’s the part that matters.