What Is a
Data Center?

A building full of computers

Strip away the mystique and a data center is exactly that: a warehouse-scale building packed with computers that store, process, and move information — running every hour of every day, forever.

When you stream a film, send an email, check a balance, join a video call, or type a question into an AI, you are not doing that work on your own device. Your phone is a window. The actual computing happens on a machine sitting in a windowless building somewhere else, then travels back to you across fiber-optic cable in a fraction of a second. "The cloud" is not a metaphor for something weightless. It is these buildings. Concrete, steel, copper, and an enormous appetite for two things: electricity and cooling.

None of this is new. The internet has run on data centers for decades. What changed — the reason your town may now be arguing about one — is a single technology arriving all at once.

What is actually inside

Open the doors and the interior is surprisingly legible. Everything serves one of three constants — compute, power, and cooling — and everything is doubled, because a facility that stops is a facility that has failed.

• The RacksTall metal cabinets standing in long rows. Each holds a stack of servers — the computers themselves — packed as densely as heat allows.
• The ServersProcessors, memory, and storage. Traditional servers lean on CPUs; the new AI machines are built around GPUs, the chips that do the heavy mathematics of training and running models.
• The NetworkSwitches, routers, and bundles of fiber that connect every server to every other — and to the wider internet. This is what makes a room of boxes behave as one machine.
• The PowerGrid feed, battery banks, and backup generators. If the utility fails, batteries carry the load for seconds until generators catch — the lights never flicker.
• The CoolingElectricity becomes computation and heat. Removing that heat — with air, with water, or by piping coolant directly onto the chips — is half the engineering of the entire building.

The size ranges from a single room to a hyperscale campus: several buildings on one site, drawing hundreds of megawatts — the electrical load of a small city — under one roof line.

Then artificial intelligence arrived

For decades, a server rack drew somewhere between 5 and 15 kilowatts — warm, but manageable. The chips that run modern AI changed the arithmetic entirely. A single dense AI rack today can pull 120 to 140 kilowatts. By the IEA's projection, one refrigerator-sized cabinet can soon draw the power of roughly 65 households.

Two shifts compound the pressure. Training a model is a burst of intense, temporary work. Inference — actually answering the billions of everyday requests — is constant, and it now makes up the overwhelming majority of the load. A training run ends. Inference never sleeps. The result is a wave of construction moving faster than the electrical grid was ever built to absorb.

The scale, made concrete

In 2024, the world's data centers used roughly 415 terawatt-hours of electricity — about 1.5% of all the power on Earth. By one 2026 estimate that figure is climbing toward 1,050 terawatt-hours. Put differently: if the world's data centers were a country, they would rank near fifth on the planet for electricity use — between Japan and Russia.

Approximate, illustrative comparison · Projected 2026 figures

Then there is water. Most centers shed heat by evaporating fresh water — U.S. facilities used an estimated 17 billion gallons for direct cooling in 2023, a figure expected to roughly double by 2028. Newer liquid-cooling systems can cut that dramatically, but they shift the cost elsewhere, into the electricity bill. And because most of that electricity still comes from fossil fuels, the footprint is not only local. It is atmospheric.

Why your town is voting on it

Infrastructure this large no longer arrives quietly. In July 2026, New York became the first state to impose a moratorium on new hyperscale centers. By then, more than 300 cities and counties had already passed their own bans or pauses. In a single quarter, projects worth over $130 billion were delayed or canceled, opposition a factor in many.

The sentiment is not fringe. A Gallup survey in early 2026 found 71% of Americans opposed having a data center built near them — nearly half of them strongly. A federal moratorium bill now sits in Congress. The objection is rarely to computing itself. It is to a specific bargain: a facility that consumes a community's power, water, and quiet, while — critics argue — leaving behind comparatively few permanent jobs.

And yet the other side of that ledger is just as real. Which is where the honest difficulty begins.