Yves here. Keep in mind that what is blandly depicted as “date centers'” use, which creates the impression that it is due to all sorts of cloud users, such as for smartphones and small businesses, is increasingly due to AI, particularly the growth in consumption. And while most critics of AI resource-hogging focus on energy use, they often ignore another big need, which is for water.
One issue that is not often acknowledged is the use of water for cooling…which becomes much less effective and/or environmentally dangerous when ambient temperatures are high. This sort of thing has been happening with some nuclear operations. Yours truly has no idea if and when it becomes an issue with other energy sources. But many data-center operators have been depicting nuclear plants as a “least bad” solution to their burgeoning energy needs.
For instance, in 2024, Nuclear Newwire ran the story French nuclear plant lowers output due to hot river water. And from the 2025 New York Times account, Extreme Heat Shuts Down Some Nuclear Reactors in Europe:
A record-breaking heat wave in Europe is warming up the river water that some nuclear power plants use for cooling, prompting operators to shut down at least three reactors at two separate sites.
Late on Sunday, operators shut down one of the two reactors at the Golfech Nuclear Power Plant in southern France after forecasts that the Garonne River, from which it draws water, could top 28 degrees Celsius, or roughly 82 degrees Fahrenheit.
The Beznau Nuclear Power Plant in Switzerland, built along the Aare River near the country’s northern border, followed suit, shutting down one of its reactors on Tuesday and the other on Wednesday.
Both plants are designed to keep their reactors at safe temperatures by cooling them with river water, which is then pumped back out at higher temperatures. Regulations in both countries require operators to reduce energy production when the rivers get too hot, in order to protect downstream environments.
The Gray Lady depicts these temperatures as extreme, but they are set to become part of the new normal in more and more places.
By Peyton McCauley, Water Policy Specialist, Sea Grant UW Water Science-Policy Fellow, University of Wisconsin-Milwaukee and Melissa Scanlan, Professor and Director of the Center for Water Policy, School of Freshwater Sciences, University of Wisconsin-Milwaukee. Originally published at The Conversation
As demand for artificial intelligence technology boosts construction and proposed construction of data centers around the world, those computers require not just electricity and land, but also a significant amount of water. Data centers use water directly, with cooling water pumped through pipes in and around the computer equipment. They also use water indirectly, through the water required to produce the electricity to power the facility. The amount of water used to produce electricity increases dramatically when the source is fossil fuels compared with solar or wind.
A 2024 report from the Lawrence Berkeley National Laboratory estimated that in 2023, U.S. data centers consumed 17 billion gallons (64 billion liters) of water, and projects that by 2028, those figures could double – or even quadruple. The same report estimated that in 2023, U.S. data centers consumed an additional 211 billion gallons (800 billion liters) of water indirectly through the electricity that powers them. But that is just an estimate in a fast-changing industry.
We are researchers in water law and policy based on the shores of Lake Michigan. Technology companies are eyeing the Great Lakes region to host data centers, including one proposed for Port Washington, Wisconsin, which could be one of the largest in the country. The Great Lakes region offers a relatively cool climate and an abundance of water, making the region an attractive location for hot and thirsty data centers.
The Great Lakes are an important, binational resource that more than 40 million people depend on for their drinking water and supports a US$6 trillion regional economy. Data centers compete with these existing uses and may deplete local groundwater aquifers.
Our analysis of public records, government documents and sustainability reports compiled by top data center companies has found that technology companies don’t always reveal how much water their data centers use. In a forthcoming Rutgers Computer and Technology Law Journal article, we walk through our methods and findings using these resources to uncover the water demands of data centers.
In general, corporate sustainability reports offered the most access and detail – including that in 2024, one data center in Iowa consumed 1 billion (3.8 billion liters) gallons of water – enough to supply all of Iowa’s residential water for five days.
How Do Data Centers Use Water?
The servers and routers in data centers work hard and generate a lot of heat. To cool them down, data centers use large amounts of water – in some cases over 25% of local community water supplies. In 2023, Google reported consuming over 6 billion gallons of water (nearly 23 billion liters) to cool all its data centers.
In some data centers, the water is used up in the cooling process. In an evaporative cooling system, pumps push cold water through pipes in the data center. The cold water absorbs the heat produced by the data center servers, turning into steam that is vented out of the facility. This system requires a constant supply of cold water.
In closed-loop cooling systems, the cooling process is similar, but rather than venting steam to the air, air-cooled chillers cool down the hot water. The cooled water is then recirculated to cool the facility again. This does not require constant addition of large volumes of water, but it uses a lot more energy to run the chillers. The actual numbers showing those differences, which likely vary by the facility, are not publicly available.
One key way to evaluate water use is the amount of water that is considered “consumed,” meaning it is withdrawn from the local water supply and used up – for instance, evaporated as steam – and not returned to the ecosystem.
For information, we first looked to government data, such as that kept by municipal water systems, but the process of getting all the necessary data can be onerous and time-consuming, with some denying data access due to confidentiality concerns. So we turned to other sources to uncover data center water use.
Sustainability Reports Provide Insight
Many companies, especially those that prioritize sustainability, release publicly available reports about their environmental and sustainability practices, including water use. We focused on six top tech companies with data centers: Amazon, Google, Microsoft, Meta, Digital Realty and Equinix. Our findings revealed significant variability in both how much water the companies’ data centers used, and how much specific information the companies’ reports actually provided.
Sustainability reports offer a valuable glimpse into data center water use. But because the reports are voluntary, different companies report different statistics in ways that make them hard to combine or compare. Importantly, these disclosures do not consistently include the indirect water consumption from their electricity use, which the Lawrence Berkeley Lab estimated was 12 times greater than the direct use for cooling in 2023. Our estimates highlighting specific water consumption reports are all related to cooling.
Amazon releases annual sustainability reports, but those documents do not disclose how much water the company uses. Microsoft provides data on its water demands for its overall operations, but does not break down water use for its data centers. Meta does that breakdown, but only in a companywide aggregate figure. Google provides individual figures for each data center.
In general, the five companies we analyzed that do disclose water usage show a general trend of increasing direct water use each year. Researchers attribute this trend to data centers.
A closer look at Google and Meta
To take a deeper look, we focused on Google and Meta, as they provide some of the most detailed reports of data center water use.
Data centers make up significant proportions of both companies’ water use. In 2023, Meta consumed 813 million gallons of water globally (3.1 billion liters) – 95% of which, 776 million gallons (2.9 billion liters), was used by data centers.
For Google, the picture is similar, but with higher numbers. In 2023, Google operations worldwide consumed 6.4 billion gallons of water (24.2 billion liters), with 95%, 6.1 billion gallons (23.1 billion liters), used by data centers.
Google reports that in 2024, the company’s data center in Council Bluffs, Iowa, consumed 1 billion gallons of water(3.8 billion liters), the most of any of its data centers.
The Google data center using the least that year was in Pflugerville, Texas, which consumed 10,000 gallons (38,000 liters) – about as much as one Texas home would use in two months. That data center is air-cooled, not water-cooled, and consumes significantly less water than the 1.5 million gallons (5.7 million liters) at an air-cooled Google data center in Storey County, Nevada. Because Google’s disclosures do not pair water consumption data with the size of centers, technology used or indirect water consumption from power, these are simply partial views, with the big picture obscured.
Given society’s growing interest in AI, the data center industry will likely continue its rapid expansion. But without a consistent and transparent way to track water consumption over time, the public and government officials will be making decisions about locations, regulations and sustainability without complete information on how these massive companies’ hot and thirsty buildings will affect their communities and their environments.
Given society’s growing interest in AI,
I question whether “society” has an interest in ai. The interest in ai is from those who would control society.
I don’t want ai any more than I want a self driving car.
My reaction as well, tegnost.
And, after reading Krugman’s blog yesterday, on IA ‘data’ centers being energy hogs, his blog today, What Happens if AI Hits an Energy Wall?” reveals that investment in AI is now a larger component of GDP than personal consumption expenditures. What happens to the US economy if investors are, like, whoops, there is not enough energy to go around. And if we build a new data center we will have to supply our own energy. And slaves pushing around a water wheel just doesn’t hack it. And the neighbors might complain if we build our own nuclear reactors
So, just run this by me once again: these ‘data centers’ are sucking up massive amounts of energy and water, thus driving up electricity and gas prices for the populace, probably kicking off water shortages (and rising prices), and are projected to take most of the jobs. And we know who is going to reap the profits from these centers and it’s not the working class, rosy pictures of us all sitting around doing fun, creative things, while waiting for our monthly basic income check, to the contrary.
Oh, and I am watching James Galbraith’s video on Entropy Economics. He says that energy and resources for production are no longer abundant and easy to access. No Sh00t!
Why aren’t we panicking? Ummmm……
This is always a difficult topic to address, not least because its very difficult to define ‘usage’ when it comes to water. Most industrial water users return most or all of the river to the catchment following use (with or without treatment). One large chip foundry I’m aware of uses a staggering amount of water in cleaning alone, but pretty much all of it is returned to the catchment with no particular issue of contamination and ultimately a minimal impact on ecology.
By far the biggest user of water worldwide is agriculture and food processing – with animal production (directly and indirectly) being the biggest user per calorie produced. Much of it results in significant contamination.
There are lots of ‘unexpected’ uses of water in many industrial uses – for example, coal and fracked gas mining use enormous quantities of water (coal cleaning actually uses more than fracking). Most thermal industrial plants (coal, nuke, etc) require enormous amounts of water – it is increasingly common for coal and nuclear plants to have to shut down either due to drought or excessively high water temperatures (this happens around every decade or so for French inland nukes). This is one reason why these plants are situated by the sea or lakes wherever possible. Most cooling water can be returned to the catchment after use, usually without excessive contamination. Sea water can also be used for cooling although that results in the problem of mashing up lots of fish and occasionally jellyfish getting revenge on the plant.
While the use of water directly in data centres for coolant grabs attention, I suspect that the biggest real impact (taking account of pollution) is in the energy production for the data centres, which obviously depends on the primary energy source. For example, in Ireland data centres consume something like 20% of all energy from the grid, but use less than 1% of water from public or private sources. So unless the data centre is largely powered from solar and/or wind, the energy production water use is likely to dwarf the direct use.
So the environmental impact of water abstraction for data centre is likely to be highly regionalised, and depends on the nature of the plant and most importantly, the particular source of water. In global terms, its not really likely to be a big issue. If you are really concerned about water use internationally (and you should), then drink less milk, thats likely to have a far bigger impact than whatever data or AI you use.
For some reason (cheap electric rates and tax incentives) data centers like the Phoenix metro area, which is very hot and dry in the summer. At last count there are 93 data centers located there. Their water use isn’t advertised and no one that I am aware of is researching it in an extensive canvassing.
APS, one of the two major electric utilities in the region, distributes 8GW of electricity now to all its customers. They have current requests from data center builders for another 10GW, more than double the current electric generation capacity. I wonder who will pay for the new electric infrastructure needed and will the data centers wait 10 or 20 years for it to come online? .
I’ve read here and elsewhere that PG&Es infrastructure is on average 75 years old. PG&E services the Silicon Valley, as well. So, how is it that PG&E is going to be able to supply the demand for these data centers?
While I believe the author’s intent was to point out high consumption of water by these data centers, I finished the article fairly sure he was still being nice or less punitive than could/should be written. As Yves pointed out… there is a ton of water usage in electricity generation which is going up in order to powers these facilities.
The linked to video says evaporative cooling should not be used in area where water is scarce… like dry climates where evaporative cooling IS MOST EFFICIENT. I’m sure that rule is ignored for accounting reasons. On the flip side… in the eastern US where it is much more humid… what is often employed is indirect-direct evaporative cooling. Because humid air is less efficient than dry are at cooling, they basically use a 2-stage evaporative process… the first “indirect” stage passes the system air through a heat exchanger that is cooled by evaporative cooling to cool but mostly to dehumidify the air and then in the second stage the air is directly cooled by another round of evaporative cooling. This is effectively consuming around 2x the water.