A server room at a data center.(Getty Images)
The sun was beating down as I made my way through a quiet community in Ashburn, Virginia. The roads were wide and lined with cars; there was a nice central playground with no one playing in it; the houses seemed monstrously large; and there were ample trees dotting the clean-cut lawns of the neighborhood.
In so many ways, it felt like your typical US suburb, until you looked past the thick wall of trees that defined its boundaries. The white walls of a large facility peeked out from above the deep green foliage. It ran on in the distance, the large houses making it impossible to see its full extent. Towering above it a little farther away were the high-voltage transmission lines that fed it with electricity. The trees acted like a shield, allowing those who lived in the community not to think about what stood on their doorstep and the growing demands that came from the surrounding industrial expansion. I’m talking, of course, about hyperscale data centers.
No location on the entire planet has more data centers than Virginia. The state holds more than 13 percent of global hyperscale capacity—double that of Beijing, which comes in second. There were nearly 600 data centers within the state’s boundaries in 2025, about 150 of which were hyperscale—the monstrously large variety that have become much more common since the advent of generative AI—and most of them were clustered in an area in the north of the state around Ashburn, which has since been dubbed “Data Center Alley.”
Data centers are unavoidable in Northern Virginia. When I sat on a bench next to a neighborhood lake, a hyperscale data center owned by Microsoft stared back at me from the opposite bank. As I drove through industrial parks, they were dotted throughout. When I pulled into a shopping area to grab a bite to eat, facilities that powered Amazon Web Services loomed behind the strip mall. I could even feel their influence in the street names: Cloud Plaza, Performance Circle, and, worst of all, Growth Mindset Lane. We can thank Microsoft for that one.
In short, they were everywhere when I visited in 2025, and following them wherever they set down were the high-voltage power lines needed to keep them running. Data centers were remaking the fabric of a whole region; there seemed to be no end in sight for their expansion. They had become a major industry, and neither worries over local resource demands nor growing concern from residents seemed able to stop the ambitions of some of the most powerful companies on the planet.
It’s hard to generalize about the data centers themselves. They come in many shapes and sizes, operate at various scales—from rooms in a nondescript building to entire parks of warehouse-sized facilities crammed with thousands of servers apiece—and seem to be getting bigger with every passing year. Some of them exist to serve the companies operating them; others exist as places where many different groups can rent space for servers of their own. Certain facilities are cooled with air-conditioning, while others depend more on water-based cooling, and even the way those systems are implemented can differ. I guess what I’m trying to say is that one description of a data center will never cover the gamut: There is some standardization, but there is also a lot of room for operators to put their own stamp on how their facility or network of locations is going to work.
The interiors of some Google data centers have clusters of brightly colored tubes running throughout them—echoing the primary colors present in the company’s branding—while the average Microsoft or Amazon facility might appear more utilitarian. Digital Realty is one of the largest data-center operators in the world. The company has two major lines of business: It builds entire facilities that then get leased to other companies, including the major cloud providers, but it also offers colocation services. Those are facilities where Digital Realty runs the data center, renting space in racks to other companies that they can then fill with their own servers. I stopped to walk the length of one of its facilities in Northern Virginia. The simplest way to describe it is to say that it looks like a massive box. The company’s logo was visible near the upper portion of the building, and it had some different shades of gray and blue across its exterior.
The data center seemed quiet that weekend afternoon, though even on a weekday there weren’t likely to be many people there. Data centers don’t need many employees to keep them clean, secure, and running; the bulk of the jobs they provide are created in the construction phase and mostly disappear once that’s over. The parking lot at the front of the building was surrounded by a large black fence that went around the entire property. I wasn’t trying to get in, but I did wonder if there was a security officer behind a desk somewhere inside the building peering at me through the video feed of one of the facility’s many security cameras. Data centers are notoriously well-guarded, to keep their data safe and to assuage the worries of their clients.
There wasn’t much to see as I walked along the fence, the spikes at the top of each pole warning me not to try anything rash. It took nearly five minutes to walk from one end of the building to the other, and this wasn’t even one of the bigger ones I’d seen that day. At the far end, a little beyond the fence, I walked onto a wooden platform where I could make out a serene pond obscured by tall trees and plants. The chirps and trills of birds and insects were just barely audible over the buzz of the air-conditioning system on the facility’s roof. The pond appeared impossible to access as its lifeless neighbor encroached upon it.
When I pulled into an outdoor shopping mall, I parked across from the shell of a structure that was soon to become a data center. Some tech companies will take over an existing facility and convert it into a data center, but most end up being built from scratch on vast tracts of newly cleared land. Towering outer walls of concrete reached far above even the retail stores that surrounded me. Each one was held up with steel braces as they waited for the installation of roof trusses that would be used to bind the various pieces together. After that, it would all be sealed against the elements with some utilitarian exterior cladding, then fitted with the cabling, temperature-regulation systems, power backups, and everything else needed to get thousands of servers running and wired to the Internet.
It’s not always easy to know when a data center is slated for a community before it starts rising out of the ground—or, when residents do get that information, who is actually behind it. Consider Lowell Charter Township, which sits about 20 miles outside Grand Rapids, Michigan. In 2025, its residents learned that a new data center was planned for their area, and they saw their local leaders moving fast to rezone the land and get any barriers out of the way to make it happen. Yet no one would reveal which company would ultimately run the facility.
The public was not happy, and near the end of the year so many angry residents crowded into a planning-commission meeting that officials were forced to put the project on ice. It was only then that Microsoft finally came out from behind the curtain to reveal itself as the proprietor.
Situations like this are not uncommon, though they rarely end in the outcome that the residents of Lowell achieved. Major tech firms use shell companies to hide their involvement in data-center projects until it’s too late for locals to stop them, just as Google did in The Dalles, Oregon, when it used the name “Design LLC.” Amazon has gone by “Razor5 LLC” in Indiana, and Meta deployed “Sidecat LLC” in Ohio and “Degas LLC” in Wisconsin.
Tech companies need a lot of data centers to fuel their ambitions. They do not need nearly as many offices where engineers create the products that justify the need for data centers. When I saw that shell of a data center in Virginia, I was also seeing the phase in which it would provide the most employment. During construction, a data-center project tends to employ over 1,000 people—even more for the monstrously large hyperscale facilities—but once they’re done, the employment numbers tend to drop into the dozens. These are not high-paying engineering jobs: Many of the workers are doing security, cleaning, or conducting maintenance on the systems. Is that really worth hundreds of thousands, if not millions, of dollars per job in tax breaks and incentives? A growing number of Virginians didn’t think so.
The rush to build out data centers in response to the generative-AI boom has led to a renewed rush for land, this time led by the companies competing in the AI race or trying to profit from powering it. With time of the essence, access to electricity has become one of the defining metrics of a site’s viability. Companies can’t wait years for capacity to be built or grid connections to be made. In some cases, that competitive pressure has led companies to bypass some of the demands they typically make of local governments—as was the case when xAI, Elon Musk’s AI company, did not demand property-tax abatements from Memphis, Tennessee. Musk wanted the data center up and running as quickly as possible. But the construction boom has also lifted some surprising companies along with it. Construction firms clearly benefit from all the additional projects they get hired to complete, while electricity companies and computer-hardware manufacturers find there’s a much greater demand for what they’re selling.
Communities like the one I visited in Virginia were feeling the brunt of the AI boom. As the data centers got closer, some property owners even started begging government officials to rezone their land for industrial use so they could sell and move away from it all. In Northern Virginia, the signs of that opposition are increasingly common. I spotted large banners hanging from gazebos along a central stretch of road in Loudoun County, warning that towering high-voltage transmission lines would slash the values of people’s homes. “Who pays? Loudoun families,” the banners read. “Who profits? Data centers and Dominion”—the local electricity provider and utility.
Residents also have to deal with tremors through their homes during the construction period, followed by the incessant hum of air-conditioning units once the center is up and running. Some people have to contend with the bright lights that stay on all night to illuminate the exterior of the facilities. It’s easy to dismiss some of these concerns as typical NIMBY complaints, but most Virginians were not initially opposed to data centers. For a long time, there wasn’t much of a concerted campaign to stop data centers at all. That only began to change as they started encroaching on communities and people experienced the tangible impacts in their lives.
In early 2026, a cold snap hit many parts of the United States. Once the grid was stressed, data centers in Virginia were given permission to run fully on their diesel generators, with little regard for the air-quality impacts of burning all that fuel. They did not have to “reduce their consumption, take some of the AI offline,” said Chris Miller, the president of the Piedmont Environmental Council. Instead, the decision was made “to expose the general public to an increased level of public-health hazard” from the burning of all that diesel fuel. It’s no wonder that as data-center construction ramped up in 2024 and ’25, opposition began growing around it.
Between 2024, when I’d first spoken to Miller, and early 2026, when we reconnected, there had been what he described as a night-and-day shift. People were feeling the effects of data centers near their communities, and they were also getting confirmation that what they were feeling was real. Over that time, a lot more data became available on the facilities’ water use and electricity demands, and media reports shed further light on how they were contributing to higher electricity prices in places like Virginia. People were finally waking up to what was going on around them.
When you’re stuck on the outside of a data center, it’s easy to imagine the interior as sinister: ominous, towering black boxes filled with servers whose small lights and indicators give off an eerie glow in the dimmed light of a chilly server room. But the truth is that they’re typically quite mundane and utilitarian spaces. In bigger facilities, the floors can be concrete, but many others have elevated floors covered with removable tiles, while fluorescent lights keep things well lit—at least when humans are inside. Miles of cables running overhead and down into the racks enable rapid data transfer and the servers’ ultimate connection to the wider Internet.
There’s no question that the sensitive data held inside must be protected, and the extent of the effort to do so must be visible to clients. But it’s not a form of theater that ends at the door. Stringent security practices are present throughout the facility, beginning with the people who are allowed through the door. I had a hard time getting inside a major data center: The most recognizable companies rejected my requests, if they responded to them at all. Several of the experts I spoke with told me that’s typical. Since the generative-AI boom has increased the scrutiny placed on data centers, getting inside their walls has become much more difficult.
When I finally visited a data center for a tour, the extent of the security measures struck me more than anything else. Security officials already had my name when I arrived at the gate. Once I got through the door, I needed to check in and hand over my passport before I could go any further. No photos were allowed. As we made our way through the facility, it became clear that you couldn’t simply walk through any of the doors. Each one of them had its own digital lock, meaning the executive escorting me had to tap us into room after room as he explained how the different parts of the facility worked.
When you enter the server room of a data center, the warm air produced by all of the computer hardware envelops you, but far more striking is the incessant metallic hum that feels like an auditory assault—the product of all the servers and whirring fans in the room. Rows of tall lockers (called “racks”) fill the facility, each containing dozens of long, thin servers. Depending on the size of the data center, there can be hundreds, thousands, or hundreds of thousands of servers throughout the building. The server racks stand seven feet high, with bundles of colored cables connecting to the hardware and running along the back of each rack. When the servers are cooled with air-conditioning, you can feel the difference in the heat they produce simply by standing on either side of a rack.
Standing inside a data center, I couldn’t help but think of the vast network of supply chains that make them possible. The Internet has extended to touch every corner of the world, but so too do the trade routes whose finished goods eventually sit within the facility’s racks. The many thousands of servers overwhelmingly arrive from factories in Asia, with supply chains that snake out to mines across the world where the minerals and metals essential for their production originate.
The Internet can seem like an immaterial thing when we log on to social-media platforms or do Google searches that send us to any number of websites. But structuring data centers as real-estate-investment vehicles required defining and treating the network in a wholly different way. It’s an important element of the business of the Internet that often gets ignored, but it has significant implications for the Web itself. Those corporate structures shape the incentives that drive the companies ultimately building out the foundations of the digital world.
Paris MarxParis Marx is the author of Hyperscale: The Ambition and Excess of Big Tech's Data Empires. He hosts the Tech Won’t Save Us podcast and writes the Disconnect newsletter.