Why Data Centers Use So Much Water and Electricity, Explained

In one small Oregon town, a single company's data centers went through about 550 million gallons of water in a year. That was close to 40 percent of all the water the town of The Dalles used. The figure only became public in 2025, after the city dropped a records lawsuit and a court ordered the numbers released.

Water and electricity are what these buildings consume, and they are what almost every local fight over a data center comes down to. Neighbors ask how much power the site will pull from the grid and how much water it will use to stay cool. This post explains, in plain words, why the answers are so large. It takes no side. It reports what the numbers are and where each one comes from.

First, the Electricity: Why Servers Need Power All Day, Every Day

A data center is a building full of servers. A server is a computer built to run all day for other people over the internet. When you stream a show, save a photo to the cloud, or ask an AI chatbot a question, a server does the work and sends the answer back.

Those servers never turn off. A website or an AI system has to be ready every second, so the computers run around the clock, and each one draws electricity the whole time. Electricity use is measured in kilowatt-hours, the same unit on your home electric bill. One kilowatt-hour runs a 1,000-watt appliance for an hour. A single server is small on its own. A building holding tens of thousands of them, running every hour of the year, is not.

That constant draw is the first reason the numbers get big. A home's power use rises and falls through the day and drops at night. A data center's does not. It runs flat out, all the time.

How Much Power, in Real Numbers

The clearest national figures come from a 2024 report by Lawrence Berkeley National Laboratory, a US Department of Energy lab.

The report found that US data centers used about 176 terawatt-hours of electricity in 2023, or roughly 4.4 percent of all the electricity the country used that year. A terawatt-hour is a billion kilowatt-hours, so this is a very large number written in a small-looking way. For comparison, the same measure stood at 58 terawatt-hours in 2014. In other words, the total roughly tripled in nine years.

The report projected that data centers could use between 325 and 580 terawatt-hours by 2028, which would be somewhere between 6.7 and 12 percent of US electricity. The wide range reflects how uncertain the future demand is.

Why AI Made the Draw Jump

The recent surge has a clear cause. The Berkeley Lab report found that data center power demand more than doubled between 2017 and 2023, largely because of the growth in AI servers.

Here is why AI is so power-hungry, in plain terms. The computer chips that train and run AI systems, called GPUs, pack far more work into the same space than an ordinary server chip, and they burn far more electricity doing it. The result shows up at the level of a single rack, the metal cabinet that holds a stack of servers.

A rack of ordinary servers might draw 5 to 10 kilowatts, about the demand of a few electric ovens running at once. A rack packed with AI chips can draw 25 to over 100 kilowatts. One recent high-end AI rack design from Nvidia, the GB200 NVL72, draws roughly 130 kilowatts by industry accounts, more than ten times a traditional rack. Fill a building with racks like that, and the total climbs into the range of a power plant's output.

The unit that describes that building-scale demand is the megawatt. A megawatt is a measure of electrical demand, roughly the draw of several hundred typical US homes at once. The largest data centers, the class the industry calls hyperscale, are built to pull 100 megawatts or more. Hyperscale simply means the biggest tier, the kind the large cloud and AI companies build.

What a Big Campus Uses Compared to a Town

Numbers like "100 megawatts" are hard to feel. A comparison helps.

A single 100-megawatt data center, running around the clock for a year, uses roughly as much electricity as 80,000 to 85,000 typical US homes. That estimate comes from straightforward arithmetic: a 100-megawatt draw held steady for all 8,760 hours in a year, divided by the roughly 10,400 kilowatt-hours an average US home used in 2024, according to the US Energy Information Administration.

Some of the largest campuses now being proposed are measured in gigawatts. A gigawatt is 1,000 megawatts. A one-gigawatt campus, on the same math, would use about as much electricity in a year as several hundred thousand homes. This is why residents at these meetings often say a proposed site would use as much power as their whole county. In terms of electricity, that can be close to literally true.

Now the Water: What Cooling Is, and Why It Drinks

The second resource is water, and to understand it you have to understand heat.

Every watt of electricity a server uses turns into heat. A building full of hard-working computers gets hot fast, and if the heat is not removed, the machines fail. So a large share of a data center's job is cooling, moving that heat out of the building. Many data centers do it with water.

The most water-intensive method is evaporative cooling. It works the way sweat cools skin. Warm water is exposed to moving air, some of it evaporates, and evaporation carries the heat away. The catch is that the evaporated water is gone. It leaves as vapor and is not returned. A large site using this method can consume millions of gallons of water on a hot day, which is why the figure in The Dalles reached hundreds of millions of gallons in a year.

The Berkeley Lab report estimated that US data centers directly consumed about 66 billion liters of water, roughly 17 billion gallons, for cooling in 2023. It projected that direct figure could rise to somewhere between 150 and 275 billion liters by 2028. "Directly" is doing important work in that sentence, as the next section explains.

The Second Water Bill Most People Miss

A data center has a water footprint even when it uses no water on site at all, because the power plants that supply its electricity often use water too.

Many power plants, especially those that burn gas or coal, make electricity by boiling water into steam, then cool that steam back into water using still more water. So every kilowatt-hour a data center pulls from the grid carries a hidden amount of water use back at the plant. Researchers call this indirect water use, and for a data center that relies on grid power it can be larger than the water used on site. This is one reason a facility can honestly say it uses little water for cooling while its total water footprint is still substantial.

The grid is the network of wires, substations, and power plants that carries electricity from where it is made to where it is used. When people talk about a data center's water and power together, the grid is the link between the two.

Where the Power Comes From, and Who Pays

A data center that needs as much electricity as a small city has to get that electricity from somewhere. That is often the hardest part.

Meeting the demand can require new high-voltage power lines, new substations, or new power generation, and building those takes years. The question residents raise most often is not whether the power exists in theory, but who pays to deliver it: the company, or the households and businesses on the same grid whose bills fund shared upgrades. That is a matter for utilities and their regulators, and the terms differ from one project and one state to the next. This post does not assume the answer for any specific project, because the arrangements are set case by case and are a matter of public record where they exist.

How Newer Designs Cut the Water

Water use is not fixed. The cooling method is a design choice, and some methods use far less water than evaporative cooling.

One alternative is closed-loop cooling. In a closed loop, the same coolant circulates through the building again and again in a sealed system, giving up its heat and coming back, rather than evaporating away. It uses a large amount of water once to fill the loop, then very little after that. Air cooling, which moves heat out using fans and outside air, can use little or no water for cooling but tends to use more electricity to run the fans. After the Tucson City Council rejected a large data center in 2025, the developer said a revised version would use a closed-loop air cooling system instead of evaporative cooling, according to local reporting.

There is a trade-off worth stating plainly. Methods that save water often use more electricity, and methods that save electricity often use more water. AI racks that run hot are increasingly cooled by piping liquid directly to the chips, which handles the heat that air cannot but adds its own plumbing. How much water and power a given site will use is a project-by-project question, set by its design, its location, and its cooling choice. It is usually the first thing a town asks, and the answer is specific to that project.

The Bottom Line

Data centers use so much electricity because they are warehouses of computers that never turn off, and the chips behind the AI boom draw more power than anything that came before them. They use so much water because all that electricity becomes heat, and the cheapest way to remove heat has long been to evaporate water. The national numbers are large and growing: about 4.4 percent of US electricity and roughly 17 billion gallons of cooling water in 2023, both projected to climb through 2028. But the amount any single project uses is not fixed. It turns on the design, the location, and the cooling choice, which is exactly what the questions at a town meeting are trying to pin down.