Wikipedia Deep Dive

Data center

13 min read

Somewhere right now, as you read this sentence, a warehouse the size of several football fields is humming with the sound of tens of thousands of computers. The air inside is carefully maintained at precisely the right temperature. Diesel generators stand ready to kick in if the power grid fails. Security guards patrol the perimeter. And inside those racks of servers, your email sits waiting, your streaming queue knows what you want to watch next, and an artificial intelligence model is answering someone's question about the meaning of life.

These are data centers. They are the cathedrals of our digital age.

The Surprising Scale of Digital Infrastructure

Here's a number that should stop you in your tracks: in 2024, the world's data centers consumed approximately 415 terawatt hours of electricity. To put that in perspective, that's roughly 1.5 percent of all the electricity generated on Earth. It's more than many entire countries use in a year.

And it's about to get much worse—or better, depending on your perspective. The International Energy Agency projects that data center electricity consumption could double by 2030. The culprit? Artificial intelligence. Training and running AI models requires staggering amounts of computational power, and that power has to live somewhere.

The United States leads this infrastructure race by a wide margin, hosting over 5,300 data centers as of early 2024. That's more than any other country on the planet. About 80 percent of American data center capacity is concentrated in just 15 states, with Virginia and Texas leading the pack. Virginia alone has become something of a data center capital—if you've ever wondered where the internet physically lives, a good portion of it lives in Loudoun County, Virginia, a place most Americans have never heard of.

The economic stakes are enormous. One analysis estimated that without investments in data centers for artificial intelligence, United States GDP growth in 2025 would have been 0.1 percentage points lower. These aren't just warehouses full of computers. They're infrastructure as critical to the modern economy as highways and power plants.

From Room-Sized Calculators to Digital Warehouses

The data center has surprisingly deep roots. Its ancestor was ENIAC—the Electronic Numerical Integrator and Computer—which came to life in the 1940s. ENIAC was a monster. It filled an entire room, weighed 30 tons, and used 18,000 vacuum tubes that failed with frustrating regularity. It required a special environment just to function: controlled temperature, organized cabling, restricted access.

Those early computer rooms established patterns we still follow today. The raised floor, for instance—a design where the floor sits above the actual ground level, creating space underneath for cables and cooling—was pioneered by IBM in 1956. Walk into any major data center today, and you'll likely find yourself standing on a raised floor, just as engineers did seventy years ago.

For decades, these computer rooms remained specialized spaces, tended by dedicated staff in white lab coats. Then came the 1980s and the microcomputer revolution. Suddenly, people were deploying computers everywhere—on desks, in closets, wherever they would fit. Organization went out the window. But as these systems grew more complex and more critical to business operations, companies realized they needed to bring order back to the chaos.

The modern data center concept emerged from this need. Inexpensive networking equipment and new cabling standards made it possible to centralize servers in dedicated rooms. The term "data center" started appearing in common usage.

Then came the dot-com bubble of the late 1990s. Companies needed to establish an internet presence fast, but most couldn't afford to build their own facilities. Entrepreneurs responded by constructing massive "internet data centers"—facilities that could house equipment for dozens or hundreds of different companies. These facilities offered something crucial: redundancy. If one telephone company's lines were cut, traffic could be rerouted through another provider. The internet's promise of resilience required physical infrastructure designed for resilience.

The Four Flavors of Data Centers

Not all data centers are created equal. The industry generally recognizes four distinct categories, each serving different needs.

Onsite data centers are the simplest to understand. A company builds a facility on its own property, staffs it with its own people, and runs its own equipment. Banks, governments, and large enterprises often prefer this approach because it gives them complete control. The downside? They bear all the costs and complexity themselves.

Colocation facilities offer a middle path. Think of them as apartment buildings for servers. A company pays rent for space, power, and cooling, then installs its own equipment. The building owner handles the infrastructure; the tenant handles the computers. These facilities often serve as crucial internet exchange points, places where different networks physically connect to exchange traffic. Many colocation centers also host the landing points for undersea fiber optic cables—the physical infrastructure that connects continents.

Hyperscale data centers are the giants. These are the facilities operated by Amazon, Google, Microsoft, Meta, and similar companies. A single hyperscale facility might contain hundreds of thousands of servers. These companies have such massive needs that they've developed their own custom hardware, their own cooling innovations, their own everything. They operate at a scale that makes traditional approaches impractical.

Edge data centers represent the opposite philosophy. Instead of concentrating computing power in massive centralized facilities, edge data centers distribute smaller facilities closer to where data is actually being used. This matters for applications where milliseconds count—think autonomous vehicles, industrial automation, or real-time gaming. The speed of light is fast, but it's not instant. Data traveling from New York to a server in California and back introduces noticeable delay. An edge data center in New York eliminates that round trip.

The Engineering of Reliability

When a data center goes down, the consequences can be catastrophic. In 2017, a British Airways data center failure stranded 75,000 passengers. A 2021 Facebook outage—caused by a configuration error that propagated through their data centers—cost the company an estimated $100 million in revenue and temporarily made it impossible for millions of people to communicate.

Data center designers obsess over preventing such failures. The industry measures reliability in "nines"—as in, how many nines can you put after "99 percent uptime." Five nines (99.999 percent) means roughly five minutes of downtime per year. Six nines means about 30 seconds.

Achieving this reliability requires redundancy at every level. Power systems are a good example. A typical high-reliability data center might have utility power from the grid, backed up by uninterruptible power supplies (essentially giant battery banks), backed up by diesel generators, backed up by a second utility feed from a different substation. The philosophy is "N+1"—whatever the minimum required capacity, build one more than you need.

Critical servers connect to two completely separate power distribution systems, called the A-side and B-side. If everything on the A-side fails simultaneously—every utility feed, every battery, every generator—the B-side takes over without interruption. Static transfer switches can flip between power sources in milliseconds, faster than a computer can even notice something went wrong.

This redundancy extends to cooling. Data centers generate enormous amounts of heat. A single server rack might produce as much heat as a small apartment. Multiply that by thousands of racks, and you have a serious thermal management challenge. Traditional approaches use industrial air conditioning, but modern facilities are getting creative. Some use outside air when weather permits. Others employ evaporative cooling, which uses water's transition from liquid to gas to absorb heat. A few facilities have been built in cold climates specifically to reduce cooling costs—Facebook and Google both have facilities in Scandinavia where the cold Nordic air does much of the work for free.

The Hidden Environmental Costs

All that cooling requires either electricity or water, often both. This is where data centers intersect with broader environmental concerns.

The electricity demand is staggering and growing. A study by the Electric Power Research Institute estimated that by 2030, United States data centers could consume between 4.6 and 9.1 percent of the country's entire electricity generation. That's a range because the future is uncertain, but even the low end represents a massive increase from today's roughly 2 percent.

Where does this electricity come from? Increasingly, data center operators are signing deals with renewable energy providers—wind farms, solar installations, even nuclear power plants. Some hyperscale operators have committed to matching 100 percent of their electricity consumption with renewable energy purchases. But critics point out that these arrangements don't necessarily mean the data center is actually running on renewable electrons. A facility in Virginia might buy wind power credits from a farm in Texas, but the electrons actually flowing into the building come from whatever mix powers the local grid.

Water consumption gets less attention but matters enormously. Evaporative cooling systems—among the most efficient approaches for managing heat—consume vast quantities of water. In water-stressed regions like the American Southwest, data center water use has become politically contentious. A single large facility might consume millions of gallons per day.

This is the environmental tension at the heart of the AI boom. These systems require immense computational resources. Those resources require physical infrastructure. That infrastructure requires electricity and often water. Every time you ask an AI assistant a question, somewhere a data center's power meter ticks upward.

The Sound of the Future

If you've never been near a data center, you might not realize they have a characteristic sound. Thousands of cooling fans spinning continuously create a persistent drone. Air conditioning systems rumble. Generators stand ready to roar to life.

For neighbors, this can become a quality-of-life issue. Residents living near data centers have described the sound as "a high-pitched whirring noise, 24 hours a day, 7 days a week." One comparison that keeps coming up: it's like being on an airport tarmac with a jet engine idling, except the engine never takes off. It just keeps running.

The Occupational Safety and Health Administration requires noise monitoring inside data centers when levels exceed 85 decibels—about as loud as a lawn mower. Server areas can reach 92 to 96 decibels, loud enough to cause hearing damage with prolonged exposure. This is one reason the industry has been moving toward what are called "lights-out" data centers—facilities designed to operate with minimal human presence.

In a lights-out facility, almost everything is automated. Servers are provisioned, configured, monitored, and managed remotely. Staff rarely need to enter the facility at all, eliminating both the need for lighting (hence the name) and reducing exposure to the harsh environment. These facilities can be located far from population centers, solving both the noise problem and often reducing real estate costs. They also improve security—a data center with no people inside is a data center with no insider threats.

The Architecture of Information

Data center design has become its own specialized field, with practitioners drawing on mechanical engineering, electrical engineering, computer science, and architecture. The challenges are unique.

Consider the raised floor. This seemingly simple innovation—elevating the floor to create a space beneath—solves multiple problems simultaneously. Cables can run underneath, out of sight and protected from damage. Cool air can be pumped through the void and emerge through perforated tiles exactly where it's needed. The floor itself can be easily reconfigured as equipment changes.

Airflow management has become an art form. The standard approach creates alternating hot and cold aisles. Server racks face each other across a cold aisle, drawing in cool air from the front. They exhaust hot air out the back into a hot aisle. Physical barriers—sometimes curtains, sometimes rigid enclosures—prevent the hot and cold air from mixing. Some facilities position cooling units directly between server racks, intercepting heat before it can spread.

The newest frontier is liquid cooling. Air is actually a terrible medium for moving heat around—water is about 25 times more effective. Some cutting-edge facilities now pump coolant directly to individual servers, or even immerse entire servers in specially formulated non-conductive liquids. These approaches are more complex and expensive, but for the most power-hungry AI workloads, they may be the only viable option.

The scale of these facilities can be difficult to grasp. A hyperscale data center might cover a million square feet or more—equivalent to roughly 20 football fields under one roof. Someone has proposed a 65-story data center, essentially a skyscraper for computers. The global count of data centers has exploded past three million in the United States alone, with more than triple that number worldwide.

The Aging Problem

Here's a fact that might surprise you: the average data center is only about nine years old, and industry analysts consider anything older than seven years to be obsolete. Technology moves that fast.

This creates a perpetual modernization treadmill. Facilities that were cutting-edge a decade ago struggle to support today's power densities and cooling requirements. The servers that ran early cloud workloads look quaint compared to the GPUs needed for AI training. Equipment that was efficient in 2015 wastes electricity by 2025 standards.

The industry talks about "data center transformation"—a continuous process of upgrading, consolidating, and optimizing. This includes virtualization, which allows one physical server to act as many virtual machines, dramatically improving utilization. It includes automation, which reduces the need for human intervention in routine tasks. And increasingly, it includes security hardening, as data centers become more attractive targets for attackers.

There's also a human dimension to this aging problem. The specialized knowledge needed to run data centers is concentrated among engineers who are themselves aging. One industry observer noted that "data center staff are aging faster than the equipment." Recruiting and training the next generation of data center professionals has become a strategic concern.

The Future Is Being Built Now

The next decade of data center construction will likely dwarf everything that came before. U.S. market demand is expected to double from 17 gigawatts in 2022 to 35 gigawatts by 2030. Global data creation is projected to reach 180 zettabytes by 2025—a number so large it's essentially meaningless without context. (A zettabyte is a trillion gigabytes. 180 of them is more data than humanity generated in all of history prior to about 2015.)

The artificial intelligence revolution is the primary driver. Training a single large language model might require thousands of specialized processors running for months. Operating these models at scale—serving millions of queries per day—requires even more hardware. And this is just the beginning. As AI capabilities expand, so will the computational requirements.

Some see opportunity in this growth. Data centers create construction jobs, operations jobs, and support the broader tech economy. They pay property taxes and bring high-paying employment to regions that might otherwise struggle economically.

Others see risk. The strain on electrical grids is real and growing. The water consumption is substantial. The noise affects communities. The environmental footprint contradicts many companies' sustainability commitments.

What's clear is that data centers have become essential infrastructure—as fundamental to modern life as roads, bridges, and power plants. Every email you send, every video you stream, every question you ask an AI travels through these facilities. They are the physical manifestation of the digital world, humming away in industrial parks and converted shopping malls and old salt mines, keeping the internet alive.

The next time your phone connects to the cloud, spare a thought for the building full of computers that made it possible. Someone designed that building. Someone is maintaining it right now. Someone is planning its replacement. The digital future is being built, one rack at a time, in data centers around the world.