Episode Transcript
If you’re anything like the average American, you check your phone more than two hundred times every day. When you tap on the screen, you’re asking your phone to retrieve and display some data — maybe a message from your friend, or a YouTube video of a friendly Canadian showing you how to fix your dishwasher. You’re usually told that this data exists “in the Cloud,” which might make you think it’s floating around in some ethereal network in the sky. But the Cloud is much more material than that. When you hit play on that video, your phone forms a request. It directs that request to a cell tower, and fiber-optic cables funnel it to a computer server. All of these servers are inside thousands of physical warehouses, called data centers.
Raul MARTYNEK: If you think about everything on your phone, everything on a computer it really is carried by these physical building blocks and the data center is at the core of it, right? All this information originates and terminates in a data center.
That’s Raul Martynek, the C.E.O. of a company called DataBank.
MARTYNEK: I love being in a data center because it just kind of — it charges me up.
These days, much of corporate America feels the same. The data center market has never been hotter. Your local bank, hospital, and retailer likely all have servers in a data center somewhere.
MARTYNEK: We have customers like Carnival Cruises, and AMC Theaters, and we have customers like JPMorgan and PNC Bank. We have major healthcare institutions. There’s clients across the public sector. We’ll have state and local governments as clients.
That’s the whole infrastructure of our digital lives, housed in nondescript buildings scattered across the country. And with the rise of A.I., we’re only going to get more of them. For the Freakonomics Radio Network, this is The Economics of Everyday Things. I’m Zachary Crockett. Today: data centers.
* * *
From the outside, a data center doesn’t look like much. Most are giant, windowless masses of concrete and steel. They are often long and flat, less than three stories high. You might drive past one every day on your way to work, and not know what’s inside.
MARTYNEK: You have these things called data halls where all this computer equipment is in these racks, so there’s rows and rows and rows of computer equipment.
Those computers belong to companies that provide internet services. The data center’s main job is to keep them secure, powered on, and connected to the internet.
MARTYNEK: It’s kind of a quiet place overall. There’s a lot of humming of this computer equipment. You go into a 20,000-foot data hall where there’s thousands of servers and storage devices and network gear racked up and it makes a hum.
That equipment is doing two things — the first of which is storing all of this data.
MARTYNEK: A lot of what’s in a data center is, you know, industrial-size hard drives.
Every photo you copy into a cloud service — like Google Drive or Apple’s iCloud storage — lives in a server in a distant data center. Every movie and TV show on a streaming service, same thing. Let’s say you fire up Peacock and hit play on Wicked from your couch in Brooklyn.
MARTYNEK: When you click on that button, while the gold copy of that movie is sitting in that location in Washington, D.C., the actual movie was copied to the New York metropolitan area. And then that information is going from that server to let’s say that user in Brooklyn. They actually distribute copies of that movie to a whole bunch of different locations all over the country, all over the world.
But data centers don’t just store data. They also process it. According to some estimates, streaming a one-hour show on Netflix requires a little less than one kilowatt-hour of power on the server side. Asking ChatGPT to render an image of a porcupine moonwalking? That might call for around 3 kilowatt-hours — the equivalent of running a dishwasher for two full cycles. The industry has anticipated all of this new demand. The U.S. is now home to more than 5,000 data centers, according to the asset manager Apollo. These are operated by dozens of different companies. Some data centers are built from the ground up by Big Tech companies like Microsoft, Amazon, and Google.
MARTYNEK: They’re so big, and their platforms are so big, that it makes economic sense for them to spend money on data centers.
Most of the rest of the data centers, including those built by DataBank, are the digital equivalent of apartment buildings full of rental units. They’re called colocation facilities, and they host servers for dozens or hundreds of different companies at once — even the big tech companies, like Microsoft, who use them for extra capacity.
MARTYNEK: We go out, we acquire land in locations that we think are attractive from a data center perspective. We zone that land. We work with the local utility to bring power to that location. And then we hire a construction manager to build that data center on our behalf.
It’s an expensive process. The cost to build a data center is often somewhere around $10 million dollars per megawatt. That’s the amount of energy it takes to power a few hundred homes.
MARTYNEK: We just recently completed a 40-megawatt data center in the Atlanta market, so that data center cost us $440 million dollars.
While a data center is being built, DataBank works with realtors to find tenants who need servers.
MARTYNEK: It’s very similar to how they look at office space. One of these real-estate brokers will come to us and say, “Hey, listen, I have a RFP here for two megawatts for customer XYZ, and we’d like you to bid on it because we think you have a good location in a market that the client wants.”
Once the contract is secured, the company will install its own servers in a data center. That tenant is on the hook for their electricity costs, and they pay DataBank a monthly rent.
MARTYNEK: Today, pricing is around $150 per kilowatt.
For DataBank, that translates to $150,000 per megawatt per month. Meaning, a fully occupied data center with a 40 megawatt capacity is getting roughly $6 million per month in rent. But those numbers can vary based on what market you’re in.
MARTYNEK: Markets like New York, given land costs, labor costs, can be more expensive.
Coming up: Location, location, location.
* * *
Data centers can now be found in all kinds of places. They’re on abandoned farms, in old telecommunications buildings, and even a former church. A company called HavenCo once placed data centers on a sandbar in the middle of the ocean. But when you stream a TV show or hop onto a video call, there’s a good chance you are being routed back to a data center in Northern Virginia — also known as “Data Center Alley.”
Clayton ROSATI: If you ask any booster in Northern Virginia, what makes Northern Virginia the internet capital of the world, they’ll tell you it’s that 70 percent of the world’s internet traffic flows through that part of the country.
That’s Clayton Rosati, an associate professor at Bowling Green State University in Ohio. He studies data centers, and he says there’s a simple reason for Virginia’s dominance. The first modern computer network, ARPANET, was largely developed by the Department of Defense, and most of the project’s infrastructure was centered near the Pentagon. Soon after, AOL set up a facility there, and other companies followed.
ROSATI: These companies essentially started creating a kind of snowball effect. the more fiber that got laid, the more high-speed internet capacity that was available. Now they have the highest concentration of data center capacity in the world.
DataBank has maintained a presence in Northern Virginia for years. So has one of its top rivals, Equinix, which operates 270 data centers around the world. Like DataBank, Equinix has a roster of recognizable tenants, like Nokia, Walmart, Kohl’s, and Citizens Bank. But these days, Northern Virginia’s data center hotspots, like Ashburn, are running out of space — which means prices are going up
Chris KIMM: Ashburn is sort of beachfront property.
Chris Kimm is the Senior Vice President of Global Customer Care and Customer Experience at Equinix.
KIMM: We’re looking much further from the metropolitan areas than we did in the past because our core business was all about that interconnection challenge.
If you’re trying to stream content to millions of customers as fast as possible, it’s valuable to have a data hub on the east coast, close to major population centers. But A.I. companies don’t need that beachfront property. They use servers in data centers to train their algorithms. And in those cases, they aren’t delivering information to an end user. They just need a lot of processing power.
KIMM: These workloads can be far away. They don’t have to have snappy performance.
So, they’re looking far afield for places to store their servers.
KIMM: Places like Atlanta, Georgia, have become more popular and then increasingly into the counties that surround Atlanta, Georgia, further and further from the city center.
Kimm says there are a few features he prioritizes when choosing a location. First, he needs a strong power grid, ideally in a market with cheap electricity.
KIMM: We look for proximity to major electrical infrastructure. Is there a transmission line that runs through the area?
He also needs fiber-optic cables nearby.
KIMM: They don’t have to be built right to that piece of property, but are they sufficiently close by in enough capacity that you can cost-effectively build fiber to that piece of ground.
And then, of course, there’s the cost of land. Sometimes, Equinix has to buy hundreds of acres when it’s building a new facility.
KIMM: We need a place where we can buy a piece of property at a reasonable price.
The other big consideration is tax incentives.
KIMM: When a data center moves into a county in Virginia, the data center, like all real estate, is taxed. But the contents of the data center, all those computer servers and routers and network switches, some of which are very expensive indeed, are also taxed as personal property.
As data centers multiply across the American landscape, rural municipalities are jockeying to bring companies like Equinix and DataBank to their region. Kimm says elected leaders in underdeveloped areas will often try to sweeten the deal to entice them.
KIMM: “Gee, we maybe are a rural county. We don’t have a substantial tax base. We might be a county that’s got a central school set up with a high school, a middle school, and an elementary school. But we don’t have the tax base to raise the money to actually make the kinds of improvements to those schools that we wish we could. If we could attract a data center or a couple of data center campuses, what economic impact would that have on our county?”
Many critics question that logic. One of them is Clayton Rosati, the professor. In Ohio, where Rosati teaches, data centers are increasingly popping up on farms.
ROSATI: Here in, in Ohio, in Wood County, they’re located on old — or on previously farmed land for soybeans or for corn.
The switch from farms to data centers can have a significant impact on a rural county’s utility consumption. Data centers sometimes keep servers cool with huge air conditioning systems that require as much as five million gallons of water a day. And there’s some early evidence that the presence of data centers drives up the electric bills of nearby residents.
ROSATI: They draw so much electricity and use so much water that a lot of people speculate that that will increase a burden on the communities that kind of gets externalized in some ways by the digital landlords.
One thing data centers aren’t bringing is jobs. A large data center is typically staffed by only a few electricians, security guards, and cabling technicians.
KIMM: They’re the technicians that work in the building. They’re the critical facilities engineers who care for the equipment and the cooling systems, and the generators, and the electrical systems.
Tenant companies often send in their own technical specialists to monitor their servers.
KIMM: In the highest traffic data centers, there are more customers in the data center than we have staff.
These employees sit in conference rooms in Equinix facilities, waiting for something to go wrong.
KIMM: We have a large retailer — they send a staff every day into our data center to be there to make sure that they can support any problem that might occur in their e-commerce platform, because so much of their business is dependent on that e-commerce platform to deliver.
If a disk goes bad, or a software update fails, they rush into the server room to fix it. Any time you discover that an internet service has gone down, there’s probably an employee hustling down the aisles of a data center looking for the guilty server.
KIMM: You’ll sometimes hear about software upgrades that didn’t go right. We need to take the new software out and put the old software back while we figure out why that didn’t go well — but keep running the business. Those are the kinds of interventions that people do in I.T. every day, and often they’re doing that in the data center.
Most of the other staff members focus on security.
MARTYNEK: I would say it’s even more secure than a bank, right? Because anyone can kind of walk into a bank and go to the counter. with a data center, no one really who’s not allowed to be there can get there.
Again, that’s Raul Martynek, the founder of DataBank. At the company’s data centers, you’ll find a security fence, a security guard, and hallways protected by biometric passcodes.
MARTYNEK: Security is really, really important. Our customers are massively paranoid about it.
Every server is covered by a cage for protection.
MARTYNEK: We kind of subdivide the data center, the data halls, into different kinds of cages that are physically secured. And then the customers put their equipment in there.
But even a locked cage isn’t enough for some customers.
MARTYNEK: You have companies that will, despite all those security measures, encase all their equipment in a cage where there’s a roof, and even underneath the floor, so that no one can get in there. And some customers even go to the extent of hiring a personal security guard and putting them right next to their equipment, so that they can be 100% assured that no one is getting access to their data. That’s how important it is to these businesses.
It is strange to consider that the whole infrastructure of the internet is squeezed into a series of giant buildings like this. Even weirder, to appreciate that the internet inhabits the physical world along with the rest of us. Chris Kimm remembers a situation a few years ago when the power to one of Equinix’s data centers suddenly dropped out. The problem wasn’t a hacker, or a piece of bad code.
KIMM: A snake crawled up one of the pieces of equipment. And sadly the snake is no longer with us, but they shorted out the feed to the site.
For The Economics of Everyday Things, I’m Zachary Crockett.
* * *
This episode was produced by Michael Waters and Sarah Lilley, and mixed by Jeremy Johnston. We had help from Daniel Moritz-Rabson.
KIMM: There’s lots of reasons why you wouldn’t wanna spend your whole day standing in a data center cage.
Sources
- Chris Kimm, senior vice president of global customer care and customer experience at Equinix.
- Raul Martynek, C.E.O. of DataBank.
- Clayton Rosati, associate professor of geography at Bowling Green State University.
Resources
- “How Data Center Alley Is Changing Northern Virginia,” by Mac Carey (Oxford American, 2025).
- “Noisy, Hungry Data Centers Are Catching Communities by Surprise,” by Sean Patrick Cooper (The New York Times, 2024).
- “As Data Centers for AI Strain the Power Grid, Bills Rise for Everyday Customers,” by Evan Halper and Caroline O’Donovan (The Washington Post, 2024).
- “Data Plantation: Northern Virginia and the Territorialization of Digital Civilization in ‘The Internet Capital of the World’,” by Clayton Rosati, Aju James, and Kathryne Metcalf (Online Media and Global Communication, 2023).
- “A New Front in the Water Wars: Your Internet Use,” by Shannon Osaka (The Washington Post, 2023).
Extras
- “Why Is It So Hard (and Expensive) to Build Anything in America?” by Freakonomics Radio (2023).
- Data Center Map.
Comments