Article 11 ← what is AI?

Plain English. The building AI lives in.

What are data centers?

You keep hearing the term, usually in the middle of an argument. Before you pick a side, here is what one actually is, why it drinks so much power and water, and where the technology is heading. Pro or con, it helps to know what you are talking about.

start reading ↓

the building

It is a giant shed full of computers.

The cloud is not a cloud. It is a warehouse, often bigger than a stadium, packed with rows of specialized computers that run every second of every day. They get blisteringly hot, so they need constant cooling. They never sleep, so they need constant power.

That building is a data center. Your apps, your photos, and every AI answer you have ever gotten live inside one.

When someone says something is in the cloud, they mean a specific building, in a specific town, with a real address and a real power bill.
power in 24/7 rows of servers, running nonstop cooling water in your answer
a data center: power in, water in, answers out, all day every day

the scale

AI needs a lot of them, and fast.

Every question you ask an AI runs on these machines, and the chips AI uses are extraordinarily power-hungry. An ordinary data center can pull as much electricity as a small town. A new AI-focused one can use as much as 100,000 homes or more.

That is why there is a building boom. Five big tech companies alone spent more than 400 billion dollars on data centers in 2025, and plan to spend far more in 2026. US data centers already use more than 4 percent of the country's electricity, about as much as all of Pakistan.

your house 1 vs one AI data center 100,000+ homes' worth of power
one building, a small city's worth of electricity

why the power

Why one building needs a city's worth of electricity.

Three reasons. First, the computing itself: tens of thousands of chips doing trillions of calculations. Second, it never stops, running 24 hours a day, every day, with no nights and no weekends off. Third, all that work makes heat, and keeping the machines cool takes a big slice of power on top.

Most of the electricity goes straight into the computers. Cooling takes much of the rest. Very little is wasted on anything else.

Roughly 6 of every 10 watts run the servers. Much of the rest fights the heat they make.
runs every hour, every day, no breaks where each watt goes ~60% ~30% servers (the computing) cooling the machines everything else
most of the power is the computing; cooling takes much of the rest

why the water

And why it drinks so much water.

The cheapest way to cool hot machines is to evaporate water, the same way sweating cools you down. The catch is that evaporated water does not come back. A big data center can drink up to 5 million gallons a day, about what a town of 50,000 people uses.

It adds up in small sips too. A single 100-word AI answer takes roughly one bottle of water to cool. And about two-thirds of new US data centers are being planned on land that has recently been in drought, which is where the worry comes from.

One short AI reply is about a bottle of water. The real concern is where that water is being drawn from.
evaporates, gone for good hot servers water in 1 reply ≈ 1 bottle
evaporative cooling works, but the water does not come back

the friction

This is why towns are arguing about them.

The concerns are real. Neighbors worry about their power bills, their water in a drought, and the constant hum of an industrial site. They also notice that these buildings get big tax breaks but employ very few people once they are built.

The other side is real too. Construction brings jobs and tax revenue, and this is the infrastructure behind the AI that hundreds of millions of people now use every day. So it is a genuine tradeoff, not a villain. The country wants the AI; the town next door carries a concentrated share of the cost.

Most Americans say they would not want one next door. The honest answer is not banning them, it is building them better, which is exactly where the technology is going.
billswaternoise local costs jobstaxesthe AI the upside a real tradeoff not a villain
the country wants it; the neighbors carry the cost

where it is heading

Here is the part worth knowing: it is getting better, fast.

The water problem is being engineered out. New closed loop cooling fills with water once, then recirculates the same water over and over with almost none evaporating. Microsoft says the approach can cut a facility's water use by tens of millions of gallons a year, and is piloting zero-water designs starting in 2026, including in Arizona. The honest tradeoff is that using less water often means using a bit more electricity.

The backup diesel generators are going too, replaced by hydrogen fuel cells and grid batteries that produce little or no pollution. And to feed the steady power these sites need, companies are turning to clean sources, including a wave of nuclear deals. That includes micro nuclear: small, factory-built reactors meant to sit next to a data center and power it cleanly around the clock. The big tech companies have already signed more than a dozen nuclear deals, and some of these reactors are designed to drop in where a diesel yard used to be.

TODAY power and water heavy water evaporates diesel backup strains the grid tech improves TOMORROW cleaner, water-smart water recycled micro nuclear solar heat reused
same building, a cleaner recipe: recycled water, clean power, waste heat put to use

the honest caveat

None of this is finished yet.

Most of the small reactors are still being licensed and built, with real deployments mostly landing between 2028 and the early 2030s. Waterless cooling is rolling out now but is not yet everywhere. Today's data centers still mostly run on the regular grid and still mostly use water.

So the direction is clear, even though the timeline is not. That is the honest picture: real costs today, real fixes arriving, and a few years in between.

the short version

Just the building AI lives in.

A data center is a giant, always-on building full of computers. Today it uses serious power and water, and that is a real tradeoff worth taking seriously. Tomorrow it may cool itself with recycled water and run on a small reactor next door. Worth understanding before you cheer or boo.

Where these numbers come from

Figures are as of 2025–2026 and move quickly. Numbers are given as ranges and rough comparisons on purpose, because the exact figures are disputed and companies do not always disclose them. The future-technology items are real and underway, but most are still being built or licensed.