How Much Water Does AI Really Use? And How Does It Compare to Our Food?

You've probably seen the headlines: Every time you ask ChatGPT a question, you're using a bottle of water.

Wait, WHAT?

With artificial intelligence becoming part of our everyday lives, there's been a LOT of conversation about its environmental footprint. And one question keeps popping up: How much water does AI use, really?

The answer? Well, it's a little more complicated than a water bottle.

From massive AI data centers to the electricity needed to power them, artificial intelligence relies on more natural resources than most of us realize. And with AI use growing rapidly, so are concerns about water consumption, energy efficiency, and the impact on local communities.

But here's something else worth talking about: the food we eat has a water footprint, too.

And depending on what's on your plate, those numbers can be pretty surprising.

So, let's break down AI's water consumption, where all that water actually goes, and what it can teach us about the everyday choices we make. Including what's for dinner!

How Much Water Does AI Use?

First things first. Does asking ChatGPT a question really consume an entire bottle of water?

Not exactly.

That popular claim comes from a research paper titled Making AI Less Thirsty, first published in 2023, which explored the water footprint of artificial intelligence.

The researchers estimated that training GPT-3, one of OpenAI's earlier large models, consumed approximately 700,000 liters of freshwater through on-site cooling.

That's a LOT of water.

But there's an important difference between training large models and using AI services to answer everyday questions.

The same research estimated that approximately 500 milliliters of water, or about one standard water bottle, could be consumed across 10 to 50 medium-length AI responses.

Not one bottle for every question.

And even that estimate isn't a fixed number.

AI's water consumption depends on several factors, including the type of AI computing involved, the efficiency of the hardware, the location of the data centers, and the cooling systems used to dissipate heat.

Some newer AI models can process requests more efficiently. Others require more energy, especially when handling complex tasks.

So when we ask how much water AI uses, we're really asking about a whole system of electricity, cooling, computing, and infrastructure.

And as more people turn to artificial intelligence for everything from writing emails to planning dinner, the total water use can add up quickly.

How Much Water Could AI Use in the Future?

Here's where the numbers get even more interesting.

Researchers behind Making AI Less Thirsty projected that global AI demand could account for 4.2 to 6.6 billion cubic meters of water withdrawal annually by 2027.

That's trillions of liters of water!

Now, water withdrawal isn't quite the same as water consumption. Withdrawal measures water taken from a source, while consumption refers to water that isn't returned to that source for immediate reuse.

Still, it's a pretty eye-opening projection.

And remember, AI's total water footprint isn't just about training large models like GPT-3. It also includes operating those models every time someone submits a question, generates an image, or uses another AI-powered service.

Each AI query has an associated water cost, even if that amount is relatively small.

But across billions of interactions? Those numbers can add up quickly.

According to the International Energy Agency, data centers accounted for approximately 1.5% of global electricity consumption in 2024. Its projections suggested that electricity demand from data centers could more than double by 2030, with AI playing a major role in that growth.

More AI computing means more electricity consumption. And more electricity often means more heat, more cooling, and potentially more water consumed.

Which brings us to the next question.

AI Data Center Water Usage: Why Does Cooling Require So Much Water?

Ever notice how your laptop gets warm when you're running a bunch of programs?

Now imagine that, but on a MUCH bigger scale.

That's essentially what's happening inside data centers.

These enormous facilities house thousands of powerful servers that process information around the clock. And all that computing generates waste heat.

To keep equipment from overheating, data center operators rely on sophisticated cooling systems to dissipate heat and maintain safe operating temperatures.

Some of those cooling systems use a significant amount of water.

Evaporative Cooling

One common approach is evaporative cooling.

Here's how it works: heated water circulates through a cooling system, and a portion of that water evaporates, carrying heat away.

In facilities that use a cooling tower, water evaporation helps cool the remaining water before it circulates through the system again.

Think of it like sweating. When moisture evaporates from your skin, it helps cool you down.

Pretty clever, right?

The downside is that evaporative cooling requires a continuing supply of cooling water. Some of that water is consumed through evaporation, while additional water may need to be discharged to manage mineral buildup.

And depending on the facility, that water may come from municipal drinking water supplies, groundwater, or other fresh water sources.

Here's another interesting detail: not all cooling water is used just once.

Many data centers circulate cooling water through their systems multiple times before it needs to be replaced or discharged.

But even when water is reused, some is still lost through evaporation.

That's why data center water usage can vary so dramatically depending on the cooling technology, location, climate, and overall infrastructure.

Air Cooling

Other data centers use air cooling to manage heat.

Instead of relying heavily on water evaporation, these systems use fans, air circulation, and refrigeration equipment to remove heat from servers.

Air cooling can dramatically reduce direct water usage, but there's a trade-off.

Under certain conditions, air cooling requires more energy than evaporative cooling. That can increase electricity demand and the indirect water footprint associated with power generation.

So while air cooling may conserve water at the data center itself, it doesn't necessarily eliminate the environmental impact.

Liquid Cooling and Closed-Loop Systems

As AI data centers become more powerful, tech companies are exploring additional ways to improve cooling efficiency.

Some facilities use closed-loop cooling systems that circulate the same liquid repeatedly, helping reduce the need for additional freshwater.

Others use immersion cooling, where computing equipment is submerged in a specialized non-conductive liquid that transfers heat away from sensitive electronics.

One important distinction? A closed-loop system can still transfer heat to a separate cooling tower that consumes water through evaporation.

So the overall water savings depend on how the entire cooling system is designed.

The goal is to keep servers cool while using less water and energy.

Because when you're operating thousands of servers around the clock, even small improvements in energy efficiency can make a big difference.

AI's Water Consumption Goes Beyond the Data Center

Here's something that doesn't always make it into the headlines.

The water consumed directly by AI data centers is only part of AI's total water footprint.

There's also indirect water consumption.

And that's where electricity generation comes into the picture.

Data centers use enormous amounts of electricity to power servers, networking equipment, and cooling systems.

But that electricity has to come from somewhere.

Depending on the energy source, electricity generation can require substantial amounts of water.

For example, many power plants that rely on fossil fuels use water to cool equipment during power generation. Some nuclear power plants also require substantial cooling water.

That means a data center can have an indirect water footprint even if its own cooling systems consume very little water.

A 2024 report from Lawrence Berkeley National Laboratory estimated that U.S. data centers directly consumed approximately 66 billion liters of water in 2023.

The indirect water consumption associated with their electricity use? Nearly 800 billion liters.

That's roughly twelve times the direct water consumption!

And it's why looking at data center water usage alone doesn't tell the whole story.

The total energy required for AI computing matters, too.

How Much Electricity Do Data Centers Use?

The numbers are pretty eye-opening.

According to the U.S. Department of Energy, data centers accounted for approximately 4.4% of total U.S. electricity consumption in 2023.

A 2024 government report projected that figure could climb to between 6.7% and 12% by 2028.

That's a significant share of the country's total electricity demand.

And with AI requiring increasingly powerful computing equipment, electricity consumption is becoming a major part of the conversation about artificial intelligence and sustainability.

Because the more electricity data centers need, the more important it becomes to consider how that power is generated, how much water it requires, and whether cleaner, more efficient alternatives are available.

What About Renewable Energy?

This is where renewable energy can make a difference.

Solar and wind power generally require much less operational water than many conventional thermoelectric power plants.

Increasing the use of renewable energy can help reduce both the water footprint and greenhouse gas emissions associated with electricity generation.

Of course, the environmental footprint of AI goes beyond water and electricity.

Manufacturing servers and computing equipment requires raw materials, including critical minerals.

And there's another hidden water cost: producing the semiconductor chips that power AI systems requires large quantities of ultra-pure water for cleaning and processing.

So the water footprint of artificial intelligence begins before a server is ever switched on.

As technology evolves, replacing equipment also creates electronic waste.

And when electricity comes from fossil fuels, increased energy use can contribute to carbon emissions.

Making data centers more energy efficient isn't just about reducing electricity bills. It's part of a much bigger conversation about sustainability goals, natural resources, and the long-term environmental impact of AI.

What Happens When Data Centers Are Built in Water-Stressed Regions?

Here's where location becomes especially important.

Not all water is created equal. Or, more accurately, not all water use has the same impact.

Using water in a region with abundant rainfall and reliable water supplies is very different from consuming water in an area already experiencing drought.

In water-stressed regions, local communities may already be competing for limited water resources needed for agriculture, drinking water, and everyday life.

And when new data centers enter the picture, those demands can become even more complicated.

Northern Virginia, for example, is one of the largest data center hubs in the world. As the industry expands, questions about electricity demand, infrastructure, and water usage have become increasingly important.

Other areas, including parts of New Mexico and South Dakota, have also faced discussions about how proposed data center developments could affect local resources.

For communities already dealing with water stress, the prospect of additional industrial water consumption can raise understandable concerns.

After all, potable water is essential for human well-being.

And when scarce water resources are involved, protecting local water supplies should be a priority.

Some tech companies have announced water positive sustainability goals, aiming to replenish more water than their operations consume.

But here's the important part: replenishing water in one location doesn't necessarily eliminate water stress in another.

Water conservation is most meaningful when it considers the needs of the specific communities and ecosystems involved.

As the rapid expansion of artificial intelligence continues, decisions about where new data centers are built and how they operate will matter more than ever.

How Can We Reduce AI's Water Footprint?

The good news? There are already ways to make artificial intelligence a little less thirsty.

And no, they don't all involve asking ChatGPT fewer questions.

More efficient computer chips: Newer, energy-efficient hardware can perform certain AI tasks using less electricity, potentially reducing the water needed for power generation and cooling.

Smarter scheduling: Some AI computing tasks, especially training large models, can be scheduled for times or locations where cooling is more efficient and water resources are less strained.

Better cooling technology: Closed-loop cooling systems, air cooling, and other advanced technologies can help reduce direct freshwater consumption.

Cleaner electricity: Using renewable energy sources with lower operational water requirements can reduce the indirect water footprint of AI.

Thoughtful locations: Building new data centers in places with suitable climates and reliable water resources can help minimize pressure on water-stressed regions.

Using reclaimed water: Some facilities can use treated wastewater instead of potable water for cooling, helping conserve drinking water supplies.

Of course, none of these solutions eliminates AI's environmental footprint entirely.

But together, they offer meaningful opportunities to improve efficiency, conserve water, and support more sustainable AI development.

And that brings us to something a little closer to home.

Or, more specifically, our kitchens.

Now, Let's Talk About the Water on Your Plate

Okay, so we've established that AI uses water.

Sometimes a LOT of it.

But here's something that might surprise you.

The food we eat has a water footprint, too. And depending on what we're eating, it can be pretty substantial.

We're not just talking about the water used to rinse your vegetables or boil a pot of pasta.

We're talking about the water needed to grow crops, raise livestock, and produce the ingredients that eventually make their way onto our plates.

Think rainfall, irrigation, animal feed, and the water involved in agricultural production.

And when we compare how much water it takes to produce the same amount of different foods, the numbers are pretty eye-opening.

How Much Water Does It Take to Produce 1 Kilogram of Food?

According to peer-reviewed research by Mekonnen and Hoekstra on the global water footprints of crops and animal products, here are approximate global averages:

A quick note on the numbers: Water footprints can vary depending on where food is grown, how it's produced, and what's included in the calculation. For this comparison, we're looking at the total water footprint, which includes rainfall used by crops (green water), irrigation and other surface or groundwater consumed (blue water), and an estimate of the water associated with agricultural pollution (grey water). Using the same methodology across foods helps give us a clearer picture of how they compare.

Now, let's put those numbers into perspective.

Producing one kilogram of beef has an estimated total water footprint of approximately 15,400 liters.

Producing the same weight of dry lentils? About 5,900 liters.

That's approximately 62% less water for lentils.

Pretty impressive for such a tiny legume!

And lentils aren't just a delicious addition to soup. They're packed with plant-based protein, fiber, and important nutrients.

Of course, different foods have different nutritional profiles, and the water needed to grow or raise them varies depending on farming practices, climate, and location.

But the overall takeaway is pretty compelling.

Choosing more plant-based foods can be a meaningful way to reduce the environmental impact of what we eat.

And that's something we can definitely get behind.

Why Different Types of Water Matter

Remember how we talked about direct and indirect water consumption in AI data centers?

Well, food production has its own version of that story.

When researchers calculate the water footprint of food, they often look at three different categories.

Green water: Rainwater stored in the soil and used by plants as they grow.

Blue water: Water from rivers, lakes, reservoirs, and groundwater sources that is consumed through irrigation and other agricultural activities.

Grey water: A calculated amount of freshwater associated with assimilating pollution from agricultural production.

Why does this matter?

Because a crop grown primarily with rainfall doesn't place the same demands on local water supplies as a crop that requires extensive irrigation in a water-stressed region.

And the total water footprint doesn't necessarily tell us how much potable water was consumed.

That's why context matters so much when comparing water usage.

It's also why two studies examining the same food can report different water consumption figures. They may be measuring different categories of water, using different geographic averages, or accounting for different stages of production.

The same principle applies to AI's water footprint.

Some estimates focus on cooling water consumed directly at data centers. Others include the indirect water associated with electricity generation.

Neither number tells the entire story on its own.

Understanding what's being measured helps us make more meaningful comparisons.

So, How Does AI's Water Footprint Compare to Food?

Here's the thing: comparing AI's water consumption to food production isn't exactly apples to apples.

Or apples to servers?

The water footprint of a single AI prompt is measured very differently from the water needed to produce a kilogram of food.

And different types of water use have different environmental consequences.

For example, rainfall used to grow vegetables isn't the same as fresh water consumed through evaporative cooling at a data center.

Likewise, producing a kilogram of beef involves a completely different set of resources and processes than running an AI model.

So it's not as simple as declaring one activity better or worse based on a single number.

But looking at both helps us understand something important.

Our everyday choices are connected to water resources in ways we don't always see.

Every time we use AI services, electricity and computing infrastructure are working behind the scenes.

And every time we sit down to eat, we're enjoying food that required water, energy, land, and other natural resources to produce.

Both systems have an environmental footprint.

Both can benefit from improved efficiency and more sustainable practices.

And both deserve thoughtful conversations about how we can use our resources more responsibly.

Small Choices. Bigger Impact.

At Soupergirl, we believe good food should be good for people AND the planet.

That's why we're big believers in the power of plants, reducing food waste, and making sustainability part of how we do business.

We're proud to have earned Fair Food Certification, reflecting our commitment to the people behind our ingredients.

And we're always looking for ways to reduce our environmental footprint.

Does eating a bowl of soup solve the world's water challenges?

Of course not.

Just like skipping a few AI prompts won't single-handedly solve water stress.

But being curious about where our resources go? Asking better questions? Making more thoughtful choices when we can?

We think that's a pretty great place to start.

Because sustainability isn't about doing everything perfectly.

It's about finding meaningful ways to do a little better.

And sometimes, that starts with something as simple as what's in your bowl.

Hungry for a little positive change? Explore our plant-based soups and learn more about our sustainability commitments.