Will The Data Center Get Built? (What You Need To Know)

The Real Estate Espresso Podcast

Will The Data Center Get Built? (What You Need To Know)

Spotify Google Podcast Apple Podcast Subscribe via RSS Leave a Review

Will The Data Center Get Built? (What You Need To Know)

Welcome to The Real Estate Espresso Podcast, your morning shot of what’s new in the world of real estate investing. I’m your host, Victor Menasce.

Every single gold rush in history has been followed by a hangover. That’s been true since the tulip mania of the 1600s. There was the gold rush in railways in the 1830s in the UK, and then later in the 1870s in the US. There was an actual gold rush in 1896 in the Klondike region of the Yukon, close to the border with Alaska. It happened in automobiles in the early 1900s. It happened in radio and then in TV and then the internet.

But seriously folks, this time is different. This AI thing is here to stay and the projections for its impact on the economy are well understood. They’re very well quantified.

While there’s increasing numbers of tools that are showing maps of both existing and planned data centers, the limiting factor for each new data center is often described as being the infrastructure. That means optical fiber, large supply of baseload electrical capacity, and in many cases, ample water supply to provide the cooling.

The latest tool is a new open-source database. It’s on a map called the IM3 Open-Source Datacenter Atlas. It provides locations and facility footprints of both existing and projected data centers across the United States. I’ll put a link in the show notes.

Projected data centers are forecast through the year 2035. These facilities are assumed to be hyperscale, meaning more than a million square feet, and are modeled as of 2035 across four data center growth scenarios and five market gravity scenarios. The market gravity scenarios indicate the relative importance of the proximity to existing data center markets or high population areas compared with locational costs.

These future data center locations are constrained to be within two kilometers of electric substations, five kilometers of water supply service areas, and two kilometers of high-speed fiber. They also include eleven additional geospatial exclusions. Future research is going to explore the intersections between the new data centers, the electric grid, and water availability. All of this is contained in the data repository. So this is a living, active database.

What this means is that the smaller data centers of, say, only 100 megawatts, which are not the hyperscale ones, are not showing up on the map. But here’s an interesting fact according to utilities analyst Andy DeVries from CreditSights: there’s about 100 gigawatts worth of power capacity expansion in the next five years, and there’s 52 gigawatts worth of data center demand planned over that same time period. Now that sounds like a mismatch.

Initially this power-generating capacity is going to be coming from natural gas plants, potentially from restarting older nuclear power plants that were recently retired or mothballed. Three Mile Island would be a good example, as well as existing nuclear power plants that might have their lives extended. Some of that power could come from solar but in the current environment, solar is expected to be a tiny contributor to data center supply.

The long-term forecast is that the nuclear renaissance will provide a clean source of power and that electric vehicles and data centers will in fact soak up that demand.

The large-scale nuclear power plants are massive investments that have historically been custom designed. The very last one of these to be built in the US was the Vogtle Electric plant in Waynesboro, Georgia. As of 2026, these remain as the only nuclear power reactors built from scratch in the US in the 21st century. Unit 3 entered commercial operation in July of 2023 and Unit 4 in April of 2024.

This project is widely cited as the most expensive infrastructure project in history. The original estimate for both units was 14 billion dollars, and it finished at 37 billion dollars. A massive cost overrun.

The final price tag was more than double the original budget. When we look at what this really translates into in terms of cost per watt, it’s somewhere in the range of 17 dollars a watt, making it significantly more expensive than virtually any other kind of power generation. It was more expensive than wind, solar, or natural gas. These were originally slated to go online back in 2016. The final project finished seven years behind schedule.

The fact is, the US has not been building nuclear power plants for some time, and they’ve really lost the brain trust. This was a first-of-a-kind design and it led to frequent engineering changes in the middle of construction. Back in 2017 the contractor Westinghouse Electric filed for Chapter 11 bankruptcy, largely due to the massive losses associated with this project.

As a point of comparison, a solar installation today would cost somewhere between four to six dollars per watt at the industrial scale. The panels themselves are only about a dollar a watt. Building a mega nuclear power plant with delays and risks makes no sense at all.

We also know that a large percentage of data centers that are planned by OpenAI were scheduled to be built by Larry Ellison’s Oracle Corporation with a significant amount of debt associated with those data centers. This past week Oracle shocked the markets by announcing they would raise 45 to 50 billion dollars in a single year to fund these data centers.

The structure includes a 25 billion dollar issuance of unsecured bonds and up to 20 billion dollars through an at-the-market equity program. The problem with the at-the-market equity program is that it effectively drips new shares into the market, diluting current stockholders and putting downward pressure on the stock price. As you can imagine, the existing shareholders are not too happy about that.

On the one hand we’re hearing the US has less than 10 gigawatts of capacity, not enough to supply the demand. And then on the other hand, a lot of these supply increases, most of them natural-gas based, are nearly double the forecast demand increase.

The power infrastructure growth is seen as a competitive race against China in particular. China has increased its power-generating capacity by nearly 500 gigawatts, about half of that through solar and hydroelectric. A bunch of it is through coal-powered generating plants. The difference is that in China the build-out is being nationally mandated. In the US the build-out needs to be supported through a viable business case and an investment mandate. Without applying any judgment as to which model is better, I can tell you that one of these models appears to be most certain. And frankly, that’s the China model.

As real estate investors, the lesson is simple. If your real estate business case assumes a large-scale data center build-out that may or may not be financially viable, then your own real estate project could be built on a house of cards. You definitely need to take these investments with a very healthy grain of salt.

As you think about that, have an awesome rest of your day. Go make some great things. Go make some great things happen and we’ll talk again tomorrow.

Stay connected and discover more about my work in real estate and by visiting and following me on various platforms: