Log in
The AI companies insist: we barely use water, hardly a drop!
But we won’t tell you how much water we use. And we’ll take you to court to stop you from finding out.
Google wants to build a new data centre in Roanoke in Virginia. They did the deal in June 2025. How much water will it need?
Roanoke gets its drinking water from Carvins Cove Reservoir. The locals tried to find out just how much water Google would be taking. But Google wanted the water and power numbers kept secret: [Roanoke Rambler]
Neither the power company nor the water authority would specify the amount of electricity and water the project might need. By the time of the Botetourt County announcement, project representatives had already persuaded electricity and water officials to sign non-disclosure agreements.
Henri Gendreau from the Roanoke Rambler sued the Western Virginia Water Authority for the details. The judge agreed with him: [ruling, PDF, archive]
The water usage information being sought is not a right, much less a right associated with ownership or possession. It is not a thing that is possessed. It is not information that is owned by Google. It is not an item that is made or marketed. Considering the plain meaning of the word, the court finds that WVWA has not established that the water usage information is proprietary information.
But the Water Authority said it would appeal the ruling. Why? [Roanoke Rambler]
Because Google strongly believes the redacted information is proprietary.
The details were finally released last week: [WSLS, video]
Carvins Cove here behind me is not just the main drinking source for Botetourt but most of the Roanoke Valley. Officials say they can handle that two to eight million gallons a day, but if it exceeds that demand, they’re going to have to start looking for a new water source.
That’s 7.5 million to 30 million litres of drinking water every single day. This is the reservoir’s entire remaining capacity. Google is taking absolutely the limit of all the water they can.
How about the other AI vendors, like OpenAI? Well, Sam Altman assures us that water is fake: [YouTube]
Anant Goenka, Indian Express: The amount of natural resources that are going into the data centres, the amount of water, the amount of …
Altman: Water is totally fake. It used to be true, we used to do evaporative cooling in data centres, but now that we don’t do that, you see these things on the internet where, don’t use ChatGPT, it’s 17 gallons of water for each query or whatever, this is completely untrue. Totally insane. No connection to reality. What is fair, though, is the energy consumption. Not per query, but in total, because the world is using so much AI, is real, and we need to move towards nuclear or wind and solar very quickly.
Notice what Altman did there — he started with the headline claim “water is totally fake” then he gave a made-up example ending with “or whatever.” What he did not give was anything like a number. A current number.
(Also, Altman’s actual power solution is all the gas turbines he can get.)
Last year, in June 2025, Altman said an average ChatGPT query uses: [blog post]
about 0.000085 gallons of water; roughly one fifteenth of a teaspoon.
That many US gallons is 0.32 of a millilitre. As of June 2025, ChatGPT was running about 2.6 billion queries a day. That works out to 837,000 litres a day. [Substack]
I’m not inclined to believe that’s the entire number until I see precisely how OpenAI calculated it and what they included — and didn’t include. Like ongoing training.
OpenAI did not respond to any of the press queries asking how Altman got this number. So I see no reason to trust it’s the number anyone else would calculate.
Given the secrecy, assume all the hyperscalers use a huge amount of fresh water. Until they give us official numbers somewhere they’re not allowed to lie. They’re not fighting to keep the numbers secret because they’re good.
