What a data center is, and why they're being built.
A plain-language, neutral guide to what data centers are, how they actually work, what they cost, and what they bring — the general picture, independent of any single project.
UpdatedMay 28, 2026
The basics
What a data center is.
A building full of computers that run continuously — the physical place where the cloud, online services, streaming, and AI actually run. At scale it is less an office than an industrial facility built around two systems: getting power in, and getting heat out.
What they do
The everyday internet
Cloud apps, file storage and backup, video streaming, websites, banking and hospital systems, and AI training and inference all run on servers inside data centers somewhere.
Efficiency
PUE, in plain terms
PUE compares total facility power to the power actually reaching the computers. 1.0 is perfect; the recent global average is roughly 1.54–1.56 — though older U.S. figures ran nearer 2.0 and the best new designs approach 1.1, so treat any single number as a rough benchmark.
The core challenge
Two hard problems
Almost everything in the design comes down to getting enough electricity in, reliably, and getting the resulting heat back out. Most of the cost and footprint serves those two jobs.
Inside a data hall: rows of racks running continuously, with the power, cooling, and networking that keep them online.
What is a data center?
A building where the internet actually lives.
Inside are long rows of racks holding servers — the machines that store data and do the computing. Around them sits everything that keeps them alive: power distribution, cooling, networking and fiber, and fire suppression. The work behind streaming a show, backing up a phone, a banking or hospital system, or an AI assistant happens on servers in a facility like this.
A single rack: stacked servers, dense cabling, and the networking gear that ties them together.
What's inside
Racks, servers, and the systems that keep them alive.
The hardware is more standardized than it looks. Servers mount into 19-inch racks; switches and fiber move data in and out; and several support systems run alongside them:
Power distribution with battery and generator backup so a grid blip doesn't drop the load.
Clean-agent fire suppression designed to protect electronics rather than soak them.
Redundancy (spare components, duplicate paths) so a single failure doesn't take the site offline.
Cooling equipment removes the heat servers produce — and the method chosen sets the water-versus-electricity balance.
How they actually work
Two hard problems: power in, heat out.
Servers turn nearly all the electricity they draw into heat, so cooling is half the engineering. The method is a trade-off: air cooling is simple but power-hungry; evaporative cooling cuts electricity but consumes water; closed-loop and liquid cooling use little water but cost more. Water versus electricity is the recurring tension of the whole industry.
Scale is the headline: a single 1-gigawatt campus running flat out would use on the order of ~20% of all the electricity Utah consumed in 2024.
That is why power and cooling — not the computers — dominate a facility's cost and footprint.
A typical exterior: large, low, and windowless. New builds increasingly land in rural regions with spare power and land.
Why they're everywhere now
AI and cloud demand are driving a build-out.
Three forces are pushing demand at once — AI training and inference, the shift of business software to the cloud, and rising video and streaming. The numbers are real but the forecasts are uncertain, so they are best read as ranges:
U.S. data centers used roughly 176 TWh of electricity in 2023 — about 4.4% of national use — climbing at close to an 18% annual rate.
Globally, installed capacity is on the order of 122 GW, with roughly another 31.7 GW under construction.
Forecasts for the U.S. share by 2030 span widely — from under 10% to as much as ~17%, a roughly fivefold spread — so treat any single 2030 figure as one point in a contested band, not a hard fact.
Because grid power and land are the binding constraints, builds are spreading from crowded hubs into rural, high-desert, energy-rich regions that can offer both.
The power connection is the hard part: large, steady demand that has to be planned against the local grid.
The costs
Real impacts, stated plainly.
Any data center, anywhere, creates the same categories of impact:
Power & grid strain — a big, steady load that can raise costs or reliability pressure for existing ratepayers.
Water — direct on-site cooling use, plus a larger indirect amount used generating its electricity. One widely cited Berkeley Lab figure puts indirect use on the order of ten-plus times the direct figure, but that multiple comes from secondary summaries and should be verified before being treated as settled.
Land — hundreds to a thousand-plus acres; converting rural land is often the most contested step.
Noise — continuous low-frequency hum; distance and setbacks are the main buffer.
Air — backup diesel is minor as operated, but continuous on-site gas generation is a much larger question.
Environmental & construction — e-waste, stormwater, largely unregulated decommissioning, and a heavy build-phase traffic peak.
A campus from above. The fiscal upside can be real where equipment is taxed; permanent jobs are usually modest.
What they bring to communities
Tax base, jobs, and lasting infrastructure.
The upside is real but varies by deal. Where the equipment inside is taxed, revenue can be large — in Loudoun County, Virginia, data centers supply on the order of a third or more of the county's general-fund revenue. Power lines, substations, and fiber can outlast the project, and a facility can diversify a single-industry economy.
Permanent jobs are usually modest — often ~100 to a few hundred even at multi-billion-dollar campuses; independent work (Brookings, 2025) finds industry job claims tend to overstate by roughly threefold.
How many roles go to local versus imported specialized crews has no clean, generalizable figure — it depends on the local labor pool.
Examples span the range: Quincy, Washington funds schools and a hospital from its cluster; Eagle Mountain, Utah drew a Meta campus (~$1.5B, ~210 permanent jobs) and a QTS campus (~$6B, ~100 jobs).
The balance
The trade-off ledger.
Set plainly side by side, this is the ledger any community weighs — neither column is spin.
Costs
Large, sustained electricity demand
Water use (direct + indirect)
Land conversion & visual footprint
Continuous noise; construction-period traffic
Air emissions (especially on-site gas generation)
E-waste & unfunded decommissioning
Contributions
Property-tax base (largest where equipment is taxed)
Construction jobs (temporary, often large)
Permanent jobs (real but usually modest)
Power & fiber infrastructure that can outlast the build
Economic diversification for the region
Foundational digital / AI infrastructure
The rules
How data centers are regulated.
There is no single "data center permit." A large project clears a stack of approvals — most consequentially a local land-use decision (rezoning and/or a conditional-use permit, decided in a public hearing where binding conditions can attach), then state air and water permits, utility interconnection, and tall-structure notice. Regulation has been tightening as the boom accelerates: Texas (SB6) lets the grid curtail very large loads in emergencies, and Georgia and Ohio have adopted tariffs making big data centers bear more of their own connection costs.
Utah's HB 76 (effective July 1, 2026) requires large data centers to report their water use — diverted, discharged, and reused — adding transparency rather than a hard cap. As recently enacted law, its details and enforcement are still settling into practice, so treat specifics as current-but-evolving.
