When people talk about the environmental cost of AI, they usually mean electricity. That's fair. AI is power-hungry, and I've written about what that does to your power bill and to America's energy-efficiency gains.
But there's a second resource that gets far less attention, and in many places it's the scarcer one.
Water.
Why Computers Need Water
Servers running AI models generate a tremendous amount of heat. Too much heat, and the hardware slows down or fails. So data centers need serious cooling.
Many facilities use evaporative cooling, which works a lot like sweating. Water absorbs heat and evaporates, carrying that heat away. It's efficient for energy, but the evaporated water is gone. It doesn't flow back into the local supply.
There's also indirect water use. Many power plants that supply data centers use water for cooling too. So AI's water footprint includes both the water used on-site and the water used to generate its electricity.
AI Makes the Problem Bigger
Data centers have used water for years. AI intensifies it.
Training and running large AI models takes far more computing power than traditional web services, and more computing means more heat, which means more cooling. As tech companies race to build bigger AI facilities, their total water use grows with them.
Several big tech companies have reported rising water consumption in their sustainability reports as their AI ambitions expanded. And researchers have tried to estimate the water behind everyday AI use, concluding that even ordinary chatbot use adds up to a meaningful amount once you multiply by millions of users.
Translation: your casual AI question has a physical cost. It just happens to be paid somewhere you can't see.
The Location Problem
Here's where it gets uncomfortable.
Data centers are often built where land is cheap, power is available and taxes are favorable. Some of those places are hot, dry regions already managing water stress. Facilities in desert areas can end up competing with farms, homes and ecosystems for the same limited supply.
Local communities are often told the project will bring jobs and tax revenue. Those benefits can be real. But data centers typically employ relatively few people once construction ends, and the water commitments can last for decades.
Communities rarely get the full picture before approval. Water use agreements can be negotiated quietly, and residents sometimes learn the scale only after the facility is running.
Who Pays for It
This follows a pattern you'll recognize if you've read my other pieces on the AI boom.
The profits are concentrated in a handful of companies. The costs are spread out: higher electricity rates for households, strained local water systems, environmental pressure on communities that never asked to host a server farm. It's the same imbalance I described in The AI Gold Rush Is Making a Few People Rich and Why Is Everything Getting More Expensive?
Is There a Better Way?
Yes, and some companies are working on it.
Different cooling technology. Closed-loop and liquid cooling systems can recirculate water instead of evaporating it, dramatically reducing consumption, though they can raise costs or energy use.
Smarter siting. Building in cooler climates, or near sources of recycled or non-potable water, reduces pressure on drinking supplies.
Recycled water. Some facilities use treated wastewater rather than fresh water for cooling.
Transparency. The simplest fix of all: require data centers to publicly report how much water they use, where it comes from and how it changes over time.
What Should Change
Let's be clear: the problem isn't that data centers use water. Hospitals, farms and factories do too. The problem is scale, location and secrecy.
Disclosure should be mandatory. Communities deserve to know how much water a facility will use before it's approved, not after.
Cost should follow the cause. If a data center strains local water infrastructure, it should pay for the upgrades, not residents.
Drought rules should apply equally. When households face watering restrictions, large industrial users shouldn't get a pass.
Local voices should count. Residents should have a real say in whether their region trades scarce water for server capacity.
The Bottom Line
AI feels weightless. It lives in "the cloud," a word that sounds soft and endless. But the cloud is made of concrete buildings full of hot machines, and those machines drink.
If AI is going to be part of everyday life, its water footprint deserves the same scrutiny as any other industry's. The least we can ask is that the companies profiting from it tell us how much they're using, and pay their own way.