Navigating site selection in the AI boom

Transcript:

Allie Barefoot: I'm Allie Barefoot with Cotality. Welcome back to Data in Context. The rapid explosion of AI has kicked off a modern-day gold rush for land, power, and infrastructure. But as developers push beyond established tech hubs, that commercial expansion is running directly into local power grids, housing markets, and municipal zoning boards. When a multi-megawatt facility lands in a secondary market, the ripple effects on local home prices and utility loads can be immediate. Traditional site selection models that really only look at acreage and fiber access simply aren't enough anymore. They could lead straight into regulatory delays and community pushback. To help us understand how site selection is evolving to meet this challenge, I'm joined today by Cotality's Head of Research and Development, Anand Srinivasan. Let's put the data in context. Hi, Anand. Welcome to Data in Context again.

Anand Srinivasan: Great to be here.

Allie Barefoot: Yeah! We'll go ahead and jump right into the first question. We've seen AI infrastructure remap the US housing market. Cotality found home prices, for example, in a small town like Abilene, Texas, jumped nearly 20% while the rest of the state saw prices drop. As data center expansion pushes into new secondary markets, where are some of the biggest friction points emerging across land, power, and housing? And what does the data tell us really about that site selection?

Anand Srinivasan: Let's start with each of those components. From a power perspective, the grid is old and wasn't built for this rapid of a demand expansion, and this magnitude of a demand expansion. Data centers consume as much power as a small city, and that is not to be trifled with. That level of capacity needs to be added very quickly, multiple times in a quarter, across multiple parts of the country. So just from a grid standpoint, the magnitude and the speed at which the demand is coming online is substantial. So it takes a little while to think about that. So that's number one. Number two is, where do we get generation of that kind of power? Particularly as coal plants get sort of shut down and alternative power sources such as even nuclear or solar and wind don't have quite the capacity ramp to offset demand. So you've got a fundamental demand-supply problem on the power generation side alone.

The second part of it is the distribution of power, right? If you look at the average home's power bill or electric utility bill, you will see that the power costs are anywhere between 8 to 20 cents a kilowatt-hour. But if you look at the distribution piece, that in many cases doubles it. So you have that second problem of you can't be too far away from a grid and production being local for that data center to be allocated or to be placed. So you have to be near power, you have to be near a lot of power. And this creates an alternate problem, which is that if you take in a lot of the power from a demand perspective, that leaves very little for the existing residents or for potential new development of residential community. So you can see that balance or that friction coming into play already just from power, right? Just from power. The second piece of it is land. So that creates another challenge: you need a lot of land, potentially pay a lot of property taxes, and you want to bring them in, but it creates land shortages. And after the initial influx of construction, you're left with a shortage from a residential standpoint. Which creates problem number three: you need a spike for short-term housing, but may not need that same level of supply for long-term housing.

So you can see how the three on their own create a little bit of an imbalance, and then when you couple the three together, you want to be close to a load center, so that puts demand—that's where the demand for AI is: large cities where offices and homes are located, that's where the people are, that's where the demand is. So you want the data center to be close, but you can't be too close because you'll be taking away housing, you can't be too far away, you'll create a lag. So site selection becomes inordinately important, a complex bunch of factors, and far more complex relative to, say, the land needs of an average strip mall or even a housing complex.

Allie Barefoot: Right. And Cotality's predictive modeling is pointing toward high-potential emerging regions—for example, New York's Hudson Valley, where energy corridors, vacant land, and proximity to major metros intersect. What specific signals is the data telling your research team that a market like the Hudson Valley is prime for the next wave?

Anand Srinivasan: Yeah, so look, it's a delicate balance for sure from a community perspective. As a resident of the tri-state area and a long fan of the Hudson Valley, on one hand, you can see the potential boom that it will bring to the economy of that location. It is a quick getaway for many New York City residents. So, but there is a lot of vacant land there that you can potentially house a large commercial development. But it will strain the power grid, right? So you have to plan for that demand consumption and you have to plan for that level of power occupancy, if you will, power load. The flip side of it is that it brings AI capabilities to the greater New York City area, and so that's a positive thing. But it then will create issues for potentially locating residential communities there. Power grids for the greater New York area will be strained. Power and cooling are roughly 40%, 20% to 40% of the average data center's energy consumption. And so these are important categories to be taken into consideration. And the Hudson Valley is anywhere between 20 to 150 miles, so it's not a big stretch of land, and it's closely located to New York. So potentially you could accommodate some parts of the driving demand need for housing of New York in the Hudson Valley. But this is a factor that you'd have to take into account because it'll compete directly with that.

Allie Barefoot: Yeah, right. And this brings up a critical challenge is community trust. Cotality's survey analysis found that consumer trust in AI dropped 14 percentage points over the last year. Technical feasibility and tax breaks, you know, sometimes they feel like they're not enough to secure that community trust. How can granular, municipal, parcel, and zoning data help predict and navigate community pushback before a developer even submits a plan for a site?

Anand Srinivasan: Look, I think surveys and an assessment of the population and the demographics of the population—where they live, how they live, what is the distribution or density of people in that area, and what are the industries they work in, etc.—all of that information can be found with Cotality, and we're happy to potentially mix that data with other data such as power needs and potential grid and structure-related information so that the company potentially locating a data center can have all the information with which to make a decision. I also wanted to share a few more data points, right? One is this notion of time, right? One of the things that's happening right now is the AI boom is driving an incredible, just a huge magnitude of demand over a very short period of time—very high velocity, right? All of the infrastructure build that we're talking about that is potentially needed to sustain this has to occur over a longer time period. So there's this huge mismatch between demand timing in terms of speed and supply timing potentially in terms of speed. So maybe there's a compromise, a way to compromise between the two and build out supply more slowly in a more informed and a community-friendly way so the community can see how things are progressing. "Hey, I put, I needed 300 megawatts, I put in 25 megawatts as my first phase, and it didn't destroy my community. It actually benefited my community. Let's do another 50," and you can do it in stages that way. So that could be potentially an option to go.

Second is the notion of hydro. Hydroelectric power is a vast, abundant power source in the nation, and there are a lot of dams that are yet untapped that could be used as power sources. I think, and we'll have to—more research will have to be extracted out of it to see if it's a favorable example, but Cheyenne, Wyoming, was pre-AI data center build, and there was a utility that was potentially set up to harvest hydro power to potentially drive the demands of that data center. This is pre-AI. So potentially there are some stories there that could be used as positive examples. The other part of it is from a velocity standpoint, right? This goes to the timing of it—goes to see if it's sustainable or not. Will the AI demand, is it a bubble? Will it burst? I think there is general recognition of the fact that AI is real, AI is here to stay, and AI can dramatically inflect productivity and be a benefit to society, right? But a lot of it is happening very, very quickly, and I think that's what is potentially kerfuffling a lot of communities. Maybe slowing it down and demonstrating—the tech community demonstrating that, in fact, this is here to stay—can be a positive force on turning the communities around actually to their advantage. And then you can use one community as the poster child for others, and use those examples repeatedly, not only in places like New York, but also potentially in Texas.

Allie Barefoot: And my final question here for you, Anand, is when developers, utility leaders, and city planners, when they gather at a negotiating table, what does a truly sustainable blueprint look like from a data architecture standpoint? You know, I know you just touched on it recently there with Washington, but what does a blueprint look like moving forward?

Anand Srinivasan: Look, I mean, historically when a factory has been located, the tax incentives are given to the company that is locating the factory, and in return what is expected is property tax, employment, and consumer discretionary income as a result of that employment, right? So the interesting part about AI-related data centers or data centers in general is that it is a power load, right? So one is, you're creating an incredible power draw to that data center from the community infrastructure, so you have to be able to handle it. In return, what you're getting is property taxes. So one of the other things that you can potentially do is have the company that is building the data center, or the company that benefits from that data center, invest in more than just creating an offset or adding to property taxes—create a lasting community. And that may not be a direct benefit to them at all, it may be a direct benefit to the community, but this is the company's way of showing "I'm here for you in the long run," right? So neighborhood parks, neighborhood schools, etc., etc. So lift up that community and demonstrate that that is a positive force for the community in addition to the property taxes. Because if you—you can't rest your benefits on property taxes alone, because the employment component is not there. The number of employees of said data center relative to a potential large auto factory, for example, is going to be substantially different. And as a result, you can get temporary spikes due to construction, but those are not going to be sustained. So you have to create additional potential benefits to the community.

Allie Barefoot: Yeah, right. Anand, it's always so eye-opening whenever I talk to you about this. You are very knowledgeable on this topic, and it's really great to break down that data and research that Cotality has. As AI density data centers are becoming a little bit more frequent in the US property market, so thank you again for taking the time to sit down with me.

Anand Srinivasan: Thank you.

Allie Barefoot: Thank you again to Anand for joining me here on Data in Context, and thank you so much for listening. If you haven't already, subscribe to Cotality's YouTube channel, and as always, if you want to find out more information, head on over to cotality.com.