Insatiable: The truth behind AI’s thirst for water

Overview

• Data centre water consumption: AI data centres use billions of gallons of water every year to keep their computer chips cool. In fact, writing a short 100-word prompt uses as much water as a 500-millilitre bottle.
• AI’s environmental impact: Beyond the water used at the data centre, power plants need huge amounts of water to make electricity. Also, making computer chips requires “ultrapure” water.
• A sustainable path forward: Canada is finding “greener” ways to run AI, like using cold winter air for cooling or using heat from servers to warm homes. But as AI grows, it puts even more pressure on the world’s water. That is why groups like World Vision Canada are working hard to help the one billion people who still do not have basic access to clean water.

What is AI?

For many people, the idea of artificial intelligence (AI) first took shape through the lens of science fiction. I’m a big fan of Star Trek: The Next Generation, and the show shaped how I first understood AI. I grew up thinking of AI as “holodecks”—rooms that create lifelike simulations—warp drives and human-like androids.

More accurately, AI is an umbrella term for technologies that process information to copy human thinking. Simple forms of AI have been around since the 1950s, but the technology has improved incredibly fast in recent years. This rapid growth is partly because computers have become much more powerful, and because we now have huge amounts of data. All that data helps train AI systems to learn and get better, faster than ever before.

What is an AI data centre?

Think of a data centre as a massive, air-conditioned warehouse. Inside are thousands of powerful computers and storage drives that run AI tools like ChatGPT.

These buildings affect the environment because they use huge amounts of electricity and water. Amazon, the world’s largest internet and cloud company, owns over 100 of these buildings and each one holds about 50,000 computers.

Data centres have been around since the 1940s. However, because we use AI so much now, companies are building them faster than ever to keep up with the demand.

What are the environmental impacts of AI?

It feels like you can’t go anywhere now without seeing AI. We can use it to learn a new language or find recipes for ingredients that don’t seem to go together. It helps us understand difficult topics and even edit an “unwanted” person—like an ex—out of a photo to make it better. This kind of convenience is powerful. AI is here to stay, and it promises to bring us one step closer to a future that once only seemed possible in Star Trek.

But AI comes with a hidden cost: the data centres that house AI equipment drain a huge amount of energy and resources from the environment.

Here’s what we know about data centres:

• They are power-hungry: To train just one AI model (like GPT-3), it takes as much electricity as 120 average homes use in a whole year.
• Their thirst is real: These buildings run incredibly hot, so they need massive amounts of water to keep their equipment from melting. For every single kilowatt-hour of energy used, they need about two litres of water for cooling. This can really drain the water supply for the towns nearby.
• There’s nothing fluffy about “the cloud”: AI sounds like it only exists in code, but it actually lives in massive buildings. These data centres are made of tonnes of steel and concrete, which create a lot of pollution to build. The intelligence might be artificial, but the environmental cost of these buildings is very real.
• They run on “super-chips”: Inside these data centres are millions of high-powered Graphics Processing Units (GPUs). Making these chips is a dirty process. It requires mining raw materials and using toxic chemicals. This adds another hidden layer to AI’s environmental footprint.

AI is thirstier than you think

When someone is thinking deeply or processing information, it’s said that their ‘gears are turning.” When you ask AI a question, it does the same thing—not in the cloud, but in a physical computer in a data centre. And when AI’s gears are turning, the computer it lives in starts to heat up, getting hotter and hotter the harder it works on your request.

These computers need cool water to keep from breaking— and they need a lot of it. According to Environmental and Energy Study Institute (EESI), AI "drinks" a massive amount of water. It uses this water both right at the data centre and indirectly through the power plants that make its electricity.

On-site cooling (Direct use)

These are some of the ways that data centres use water to keep the computers from overheating.

• Evaporative cooling: Like humans, AI data centres need to sweat to cool down. They do this by pulling in hot air from the servers and blowing it over wet pads or spraying a fine mist of water. As the water evaporates, it cools the air. About 80 per cent of the water used this way is lost to evaporation.
• Water chillers and cooling towers: Many centres use large chillers to keep computer rooms cool. They also use cooling towers to move water that absorbs and carries heat away from the building.
• Direct-to-chip cooling: This is a high-tech way to keep computers cool. Instead of using fans and air, liquid or water flows directly over the most important parts of the computer, like the Central Processing Units (CPUs) and GPUs. This “liquid cooling” is much faster and better at pulling heat away than just blowing air on them.
• Immersion cooling: This is a new method where computer parts are “bathed” in a special liquid. Unlike water, this liquid won’t damage the electronics. It takes in the heat and carries it away to be cooled back down. Because the liquid stays trapped inside pipes and tanks in a “closed-loop” system, it uses much less water than older methods.

Indirect water use (off-site)

This is the “hidden” water that isn’t used at the data centre itself but is used to support it from far away. This usually adds up to much more water than the direct use methods.

• Power plants: Most of the electricity for AI comes from power plants that burn fossil fuels like coal or natural gas. These plants use massive amounts of water to create steam to spin their turbines and to cool down their own giant engines. When you use AI, you are using electricity, which means a power plant somewhere else is “drinking” water to make that power.
• Manufacturing (before the centre opens): To build the “brains” of AI—the computer chips—factories need “ultrapure” water. This water is incredibly clean and is used to wash and carve the tiny parts of the chips. It takes about 1,500 gallons of regular water to create just 1,000 gallons of this special cleaning water. An average chip-making factory can use 10 million gallons of this water every single day. EESI notes that a single chip has already consumed thousands of gallons of water before it even arrives at the data centre.

How much water does AI use?

According to EESI, data centres use billions of gallons of water every year to keep AI from overheating. Here’s what that looks like by the numbers:

• Researchers say that it takes about 519 millilitres (roughly the size of one water bottle) to process a 100-word AI prompt.
• A medium-sized data centre can use about 110 million gallons of water per year, the same amount used by 1,000 homes annually.
• A large data centre can use up to five million gallons of water per day, or about 1.8 billion gallons every year. That’s about the same amount of water used by a town of 10,000–50,000 people.
• In the United States alone, data centres use an estimated 163.7 billion gallons of water per year.

AI’s water usage places a lot of stress on the towns where these buildings are located. In Northern Virginia, the “data centre capital of the world,” these buildings used about two billion gallons of water in 2023. That is a 63 per cent increase in just four years.

Consider this: only 0.5 per cent of Earth’s water is fresh and safe for humans to drink. With over eight billion people on the planet, AI adds a new and heavy demand for that water. Since we aren’t going to stop using AI, we have to ask a difficult question: How much water is too much to pay for this technology? As AI grows, we must find a way to ensure there is enough water for both our digital future and the billions of people who need it to survive.

How many data centres are there worldwide?

Our growing demand for AI means new data centres are popping up everywhere. Today, it’s estimated that there are over 12,000 data centres globally—with 45 per cent located in the United States.

Data from Cargoson tells us that there were 5,940 data centres in North America as of November 2025:

• In the U.S.: There were only 1,000 data centres in the United States in 2018. By late 2025, that number skyrocketed to over 5,427.
• In Canada: Experts estimated 239 data centres within Canada in 2024. Today, that number is 337.
• In Mexico: There are 173 data centres in Mexico.
• In Cayman Islands: There are three data centres in Cayman Islands.

Around the world:

• In Europe: There are 3,362 data centres across 45 countries. Germany has the most with 529, followed closely by the United Kingdom with 523.
• In Asia-Pacific: China leads the region with 449 data centres, followed by Japan with 222 and India with 122. In total, there are 1,504 data centres in Asia-Pacific.
• In Latin America: 481 data centres, including 197 in Brazil and 59 in Chile.
• In Oceania: There are 401 data centres in Oceania, which consists of 314 in Australia, 83 in New Zealand and four in Papua New Guinea.

Did you know? The China Telecom-Inner Mongolia Information Park is the world's largest data center. It is over one million square metres and uses a massive 150 megawatts of energy —athough no official figures have been released about its water usage.

What’s in the future for Canadian AI?

As the world races to build more AI, many companies are looking at Canada as a perfect place to set up shop. Canada has plenty of space and a cold climate, which helps cool down hot computers at reduced cost. We also have lots of clean energy from our rivers. However, as we build more “digital warehouses,” Canada must figure out how to handle the huge demand for our water and power.

Growing pains for Canada’s power grid

According to the Canada Energy Regulator (CER), the demand for electricity from data centres is growing fast:

• In provinces like Quebec and Ontario, data centres already use enough electricity to power hundreds of thousands of homes.
• Experts say that by 2030, the amount of power used by data centres in Canada could double or even triple. This is because AI tasks take much more energy than just sending an email or watching a video.
• While Canada uses a lot of “green” energy like wind and water power, we are building AI centres so fast that the power grid is struggling to keep up.

The “water vs. AI” problem

A big concern for the future is how much of our fresh water will be used to keep these computers cool. A report by CBC News highlighted some hidden issues:

• A single large data centre in Canada can use hundreds of millions of litres of water every year. This is water that is taken away from local rivers and lakes.
• When water is used for cooling, it often evaporates into the air. This means it can’t be used by farmers for crops or by families for drinking.
• Right now, many companies in Canada don’t have to tell the public exactly how much water they’re “drinking.” In the future, people are calling for stricter rules to protect our resources.

How can we make AI more sustainable?

To make AI “greener” in Canada, the future might look a bit different:

• Cold-climate cooling: Instead of using water, newer data centres are being built to use Canada’s freezing winter air to cool the machines.
• Recycling heat: Some new projects are trying to capture the hot air from data centres and pipe it into nearby homes or greenhouses. This helps keep people warm in the winter without using extra energy.
• Smarter rules: Canadian provinces are rethinking how they share power and water. They want to make sure that “feeding” AI data centres doesn’t take away resources from the people living nearby.

Conclusion: The final frontier?

Unlike what we see in Star Trek, we don’t have a limitless amount of water thanks to the technology of replicators. We don’t have warp drives to power starships—or AI data centres—and we don’t have the luxury of migrating to a new planet once Earth’s natural resources are gone.

Water is a precious resource that over one billion people living in low-income communities struggle to even access. But there’s hope. In 2025, your donations helped over 1.4 million people get clean water and better ways to stay healthy. This year, let’s aim for even more.