Why AI Data Centers Are Heading to Rural America

[00:00:00] Speaker 1: This is arguably the most important infrastructure map in America today. It shows the country's major hyperscaler data centers, broken down by what's already operating, what's currently being built, and what is still to come. But the map also points to a more unsettling trend. This new era of data centers, driven largely by AI, is rapidly pushing outward, away from the traditional tech corridors and increasingly into rural America. But why is this happening? Why are some of the most advanced facilities on earth being built in rural counties? And what does a community actually get when one of these projects shows up? Before we get ahead of ourselves, let's build a data center. And yes, we're tracking how often I repeat that phrase. The term "data center" gets thrown around constantly now, almost like it's one type of building. But in reality, a data center can mean a lot of different things. At the smallest end, it could be a room inside a company's office building, filled with servers that store internal files or run basic business systems. Then you have co-location data centers, where different companies rent space for their own equipment. Then you have cloud data centers, built for companies like Amazon, Microsoft, Google, Meta, and others to run the digital services we all use every day. But what we're focused on here is something much larger, hyperscale data centers, and more specifically, the new generation of them. So let's start with the basic building. This part is the outer shell, the long, windowless structure that most people see from the road. It can look like a warehouse, but what's inside is obviously much different. First, let's add data halls. These are the rooms filled with rows and rows of server racks. In a normal cloud data center, those racks might handle storage, websites, apps, or cloud computing. But in an AI data center, those racks are increasingly packed with GPUs and specialized chips designed to train and run massive models. And that immediately changes the rest of the building. These machines use enormous amounts of electricity, which we'll touch on in a minute, and you need an entire electrical system around them. So next, we add transformers, switchgear, power distribution equipment, and batteries. In many of these newer campuses, the power infrastructure is a major part of the site itself. Then there's cooling. Every watt going into those servers eventually turns into heat. So we need chillers, cooling towers, air handling units, pumps, piping, and in some facilities, liquid cooling system that bring cooling much closer to the chips themselves. Then we had backup generators. These are usually diesel or gas systems designed to keep the facility online if the grid fails. Then there's security fencing, stormwater systems, water infrastructure, fiber connections, and in many cases, a dedicated substation. All of which I didn't feel like making a 3D print for. Basically, these new data centers are places where electricity is converted into computation at massive scale. Now that we built the thing, the map starts to make more sense. If a data center were just a building, you would put it near customers, workers, fiber routes, and the existing tech economy. And for a long time, that's basically what happened. The older data center map is heavily urban and suburban. Northern Virginia, Dallas, Chicago, Phoenix, Silicon Valley, Atlanta, New York. But the new map shows something different. To be clear, this is a recreation my team and I made using data from Brokavich data center tracking. Which focuses on these major AI projects. It's not perfect, I found a few sites that are missing, but it still gives a strong sense of where the largest hyperscaler in AI projects are heading. The broader research confirms the same direction. 87% of currently operating US data centers are in urban areas, while 67% of planned data centers are in rural areas. And 39% of those planned data centers are in counties that currently don't have a single one. So why rural? The first driver is that power availability is now more important than location. The biggest AI data centers are mainly being built around electricity. And by no means is power magically better in rural America. In most cases, rural grids are weaker than urban grids. But these sites can offer room to build around power: substations, switchyards, and transmission connections. That's why you'll often see these projects sitting adjacent to massive transmission lines. The second driver is that old data center hubs are running out of usable capacity. All those cities I mentioned earlier are still important. But after years of building, many of these markets are crowded and expensive. Data center vacancy fell to a record low 1.4% at the end of 2025. Even as supply increased 36% year over year. So among everything else, the next wave is forced to look elsewhere. Sites with power access within 18 to 36 months are now highly sought after. Developers are prioritizing sites with more than 200 megawatts of power. And grid power capacity for existing projects is largely booked through 2030 in most markets. The third driver is probably what you think about: land. But it's not just about cheap land. It's about buildable land, for all the components I've mentioned so far. The fourth driver is speed. Some people believe what's called speed to power is now the primary criteria on driving site selection. That means developers are prioritizing places where the path to approval, grid connection, construction, and energization is faster and more predictable. There's also a political reason these projects move rural. For a county that does not see many billion dollar investments, a data center can look like a major economic development prize. Years of construction work, upgraded utilities, fiber infrastructure, and a much larger tax base. And this is where things get complicated. Because for a lot of rural communities, the issue is the scale of what is being asked of them. The first concern is electricity. If a facility needs hundreds of megawatts, or if an entire campus eventually needs gigawatts, residents want to know who pays for the new power plants, substations, and transmission lines needed to serve it. In Georgia, regulators approved nearly 10,000 megawatts of new generation, with about 80% expected to power data centers. US electricity use in data centers rose from 58 terawatt hours in 2014 to 176 terawatt hours in 2023. By 2028, that could rise to as much as 580 terawatt hours. That's 580 billion kilowatt hours in one year. For context, that would be roughly 14% of all electricity used in the US today, or about the same electricity used by 54 million American homes in a year. The second concern is water, and this is what I personally seem to hear about the most. This is 2 gallons of water. Multiply this by about 3000, and you get the approximate amount of water used each day for every 1 megawatt of data center capacity. And when many of these new projects are measured in the hundreds, or even thousands of megawatts, you can see why water has become such a major concern. Only in drought-prone regions, where many of these projects are being built. It is important to point out, though, that not all data centers use water the same way. The cooling system makes a huge difference. For example, closed-loop systems recirculate most of their water, so only a small amount is actually lost. Air-cooled systems use virtually no water at all, but they require significantly more electricity than water-based cooling. Even with those differences, the scale is still enormous. Nationwide, US data centers directly consumed about 17 billion gallons of water in 2023, with hyperscale and co-location facilities accounting for 84% of that total. After power and water, the next issue is what these campuses physically do to a place. At this scale, it means a permanent industrial footprint in places that may have been farmland, desert, forest, or open space. A community may be asked to absorb a huge industrial campus, while the permanent job count is often tiny. That's why some communities are trying to stop these projects entirely. In Virginia, the Prince William Digital Gateway was planned as a massive data center district near the Manassas battlefield. The developer eventually terminated its part of the project after years of local opposition and litigation, even though the project had been expected to bring major investment in tax revenue. In Michigan, Saline Township's board voted against a major data center proposal tied to OpenAI and Oracle's Stargate effort. But the developer sued, and the township later settled, a case where local rejection didn't actually end the project. In Utah, Kevin O'Leary's proposed Stratus project quickly became another flashpoint. Weekly Descriptions framed it as a 40,000-acre development with gigawatt-scale power demand, with some reports claiming it can consume more than twice the electricity used by the rest of the state. After backlash, its size was drastically reduced. The project is still deeply contested, and will probably have its own future BuildCorp video. The point of this whole video is not to make people dislike data centers. Depending on how you feel about AI, this infrastructure has to be built somewhere. These facilities are becoming part of the physical backbone of the modern economy. And to be fair, they're also becoming one of the biggest forces propping up the US construction industry today. All this to say, the cloud used to feel invisible, but the AI cloud is becoming very visible. Rural America is where the country is being asked to decide what that trade-off is worth. I'm Josh. This is BuildCorp. Thanks for watching.