A data center does three jobs. It takes power in, turns it into computation and heat, and gets the heat out. And it must never stop doing any of them.
That is the whole model. Everything else on this site, the generators, the chillers, the procedures, the people on watch at 3 a.m., exists to keep those three jobs running without interruption. Fiber, the network that connects the work to the outside world, is a supporting system in this model, not a fourth job. If you can hold that sentence in your head, you can follow every article that follows. If you lose it, the details will bury you.
Start with the building itself, because the outside tells you almost nothing. A data center has a warehouse-like appearance: a plain, windowless building in a field or an industrial park, with very few people coming and going. It is not a warehouse. Nothing is stored there in the way a warehouse stores goods. What passes through it is not physical at all. It is computation, delivered over fiber optics to people who are somewhere else entirely. A fully operational site can look sleepy from the road while serving far more people in a day than live in the town around it.
That invisibility is part of why the building is so hard to think about, and part of why it matters.
Inside, the vocabulary runs in a strict order, and it helps to learn it once and use it the same way every time. A campus is a site with one or more data center buildings plus the shared support infrastructure: substations, generators, the cooling plant, and the operations buildings. A building is a single building on that campus. A data hall is a room inside the building that houses the IT equipment. Inside the hall are rows of racks, and inside each rack are servers, the computers themselves. Campus, building, hall, row, rack, server.
That is the address of everything that happens here.
Now trace the three jobs through that address.
Power in. Electricity arrives from the grid, usually at medium or high voltage, and is stepped down through substations and transformers until it reaches the servers. Along the way it passes through switchgear, generators that stand by for a utility failure, and uninterruptible power supplies, UPS systems, that bridge the seconds between losing utility power and the generators taking the load. The path has a strict order, and the order matters: generators and the automatic transfer switch sit upstream of the UPS, and the UPS never feeds the generator. The UPS is a bridge, not a long-term source.
Its batteries are commonly cited as lasting five to fifteen minutes, long enough to bridge to generator start, and no longer. That is the first thing to understand about data center power: every piece of equipment in the chain exists to make sure the next piece never sees an interruption.
Computation and heat. A server turns electricity into computation, and in doing so it turns essentially all of that electricity into heat. There is no separate waste stream to carry away; the heat is the byproduct, and it never stops as long as the servers are working. This is why the second and third jobs cannot be separated. You cannot talk about power without talking about heat, because the power bill and the cooling bill are the same energy, counted twice, once coming in, once going out.
Heat out. Cooling captures the heat at the equipment, moves it through one or more transfer steps, and rejects it outdoors. In an air-cooled hall, servers draw cool air in at the front and exhaust hot air at the rear, and that hot air is carried away by air handlers and chillers. In a liquid-cooled hall, cold plates sit directly on the processors, and a coolant loop carries the heat to a distribution unit and then to the facility water system and outdoors. The medium changes; the job does not.
Water carries far more heat than the same volume of air, roughly three thousand times more per degree (ESTIMATE, volumetric heat-capacity arithmetic: water about 4,186 kJ per cubic meter per degree against air at about 1.2, a ratio of about 3,400–3,500 to one, so roughly three thousand is conservative), which is why liquid cooling can move the same heat through much smaller pipes than air ducts. Either way, if the heat does not leave the building, the servers throttle and then shut down to protect themselves. The computation stops because the heat did not.
Nothing down. The third phrase in the model is the condition on the other two. A data center must never stop taking power in and getting heat out. That is what makes it different from almost any other building. An office can lose power for an hour and be an inconvenience.
A data center that loses power or cooling for minutes can interrupt services that thousands or millions of people depend on, airlines, banks, hospitals, the phone in your pocket. The concentration is the point: a small crew in a plain building holds an outsized share of other people's daily life in their hands, and nobody driving past can see it.
That is also why redundancy is the first concept you will meet again and again on this site. Redundancy means having a spare: N means no spare, N+1 means one spare, 2N means a full duplicate. But redundancy on paper is not capability in fact. A spare UPS does not help if both power paths share one downstream panel. An N+1 plant in maintenance is just N, because the spare is the one being worked on. A standby pump that has never been exercised is only a hope.
Where something is my experience and judgment, it will be labeled as such. Nothing numeric here is published from memory.
That discipline is the point of the site. Journalists explain what technology companies intend to build. Almost nobody has explained, in plain language, what it is like to operate one of these places, to take rounds through a hall at 3 a.m., to know what normal looks like, and to know what to watch for when it stops looking normal. That is what the next fifty-one weeks are for.
Further reading
- Data Center Operations guide, the operations pillar: what running a data center actually covers, from the three jobs to the people on watch.
- How to Read a Data Center, the vocabulary in full: campus, building, hall, row, rack, server, plus white space and gray space.
- Scott Stillings, Author, meet the author.
- Contact, start a conversation; a conversation, not a pitch.
Sources
This article is the opening explainer for the site, drawing on Where AI Lives, Prologue and Chapter 1.
- UPS battery runtime of five to fifteen minutes: a commonly cited range; the figure varies by project.
- Water carries roughly three thousand times more heat than the same volume of air per degree: arithmetic from volumetric heat capacity (water about 4,186 kJ per cubic meter per degree; air about 1.2), a ratio of about 3,400–3,500 to one.
Published Saturday, August 29, 2026. By Scott Stillings. Scott spent more than twenty years in critical infrastructure operations, as a submarine reactor operator in the U.S. Navy, as a commercial nuclear operator and senior-reactor-operator-certified instructor, and as a critical operations manager for a large hyperscale data center campus. He is the author of Where AI Lives: Inside the Hyperscale Data Centers Powering the AI Boom (Inside Hyperscale).