With AI becoming increasingly present in everyday life, the race to build AI infrastructure is only speeding up. At the center of that race is the rapid creation of data centers, with new ones opening on a nearly weekly basis in America. But as more data centers begin to integrate AI infrastructure, the amount of electricity required to operate them is growing at an alarming rate. Data centers are now expected to account for roughly 40 percent of US power demand growth in 2026, and the gap between what we need and what we can build is widening fast.

On today’s episode of Explain to Shane, I am joined by Lynne Kiesling, a nonresident senior fellow at the American Enterprise Institute, where she leads the Electricity Technology, Regulation, and Market Design Working Group. Kiesling also directs the Institute for Regulatory Law and Economics at the Northwestern University Center on Law, Business, and Economics, and is a member of the US Department of Energy’s Electricity Advisory Committee. I am also joined by Steve DelBianco, president and CEO of NetChoice and a seasoned expert on internet governance. Their combined expertise on this issue can help us understand how we can power the AI revolution.

Below is a lightly edited and abridged transcript of our discussion. You can listen to this and other episodes of Explain to Shane on AEI.org and subscribe via your preferred listening platform. If you enjoyed this episode, leave us a review, and tell your friends and colleagues to tune in.

Shane Tews: Steve, we’re both in Northern Virginia, and that area — my neighbors are convinced that our utility bills have gone up. So how’s it going out there?

Steve DelBianco: The utilities could be forgiven for not spotting the price signals or the demand signals over the last 30 years. The total electricity demand in the United States has grown by about one-half of 1 % compounded for 25 or 30 years. At that rate of growth, we saw a retirement of fossil fuel and nuclear facilities. We saw states really encouraging through renewable portfolio standards, standing up solar and wind, which is not truly a one-to-one replacement for a baseload coal, gas, or nuclear plant. So all of that was proceeding in a relatively acceptable manner until about two years ago.

When you think about it, my industry has to build a data center, a billion-dollar data center, about once a week on average, just to hold your movies and music and videos, all of the streaming that you do, the messages that you send with copies of all of that. And cloud migration from internal IT infrastructure to the cloud, things like AWS, Google Cloud. All of these factors were contributing to an ever-growing demand for data centers, particularly because none of us ever deletes our old videos, messages, and photographs. So there’s an ongoing need, and our industry was able, on average, to add a billion-dollar data center once a week. We would simply pick the parts of the country where there was adequate power to provide a couple of hundred megawatts to a billion-dollar data center. And we can put those data centers anywhere and still serve users everywhere because of the 4 nation’s superior connectivity.

But it became clear in the last two years that something new was afoot, and that is the addition of AI. So the very same building, about the size of an aircraft carrier, that comprises a data center also holds a bunch of server blades with Nvidia chips strapped to the side. Those are $35,000 to $40,000 chips, and they burn five to 10 times more electricity than just the storage chips that we had been using for cloud-based computing and photo and video storage. So suddenly, the same buildings are consuming 400 megawatts of power. And that insatiable demand for power continues to be shoehorned into parts of the country that have generative capacity. And unfortunately, we cannot rely upon our regional grid of electrical transmission to deliver power from areas that have access to areas that need it. That grid is also serving us poorly when it comes to the part of the day where, say, solar and wind in one part of the country are just cranking out the electrons, but another part of the country is in darkness or has no wind. We don’t even have an adequate grid to move that kind of power around right now.

I want to talk about the geography of it. So let’s say you’re in Virginia: data center alley. Why there? Was it that they were permissive? They had the regulatory structure? Is it because they’re close to a lot of usage? What makes that area important?

Steve DelBianco: In the 90s, Shane, there was an internet interconnect point called MAE East. There’s a MAE West, of course, and there’s a MAE East. And it was in Tysons Corner. It was a block away from the offices of the first company I started. That meant that connectivity was readily available, especially for the transatlantic cables. And add to that the notion that we had the federal government located here in Washington. We had a lot of very inexpensive land west of Fairfax County, in an area called Loudoun County. Dominion Power, which was Dominion Energy at the time, or Virginia Power, had adequate capacity. They have a broad, diverse mix of generation, including nuclear. And add that together, and Loudoun County became very attractive to build a data center. Once you constructed one data center, then the architects, engineers, contractors, and subcontractors who built that one are ready to start on a new building as soon as it’s done in two years. So 1,800 workers will labor for two years to construct a billion-dollar data center, but before they’re even finished, we’re building another one adjacent to it on the same hundred acre campus because you’ve already done the zoning. You’ve already built a substation for interconnect. So it makes sense to continue to build them out.

So when you think about it, Shane, 25 years ago, we were both here, in Northern Virginia. When you went out to Loudoun County, out near Dulles Airport, you’d see data centers, and there’d be nothing anywhere near them. Three decades later, there are homes right up against those data centers. And in most cases, the data centers were there first. And Loudoun County grew without adequate setbacks and ended up having neighborhoods cheek by jowl with the data centers themselves. But the ecosystem of contractors, architects, engineers, service personnel, and subs — that 6 ecosystem means you can build in Northern Virginia faster, higher quality, lower cost than you can anywhere else in the world. It’s the largest concentration of data centers on the planet. Beijing is number two, and they’re not even close.

Lynne Kiesling: There’s a related phenomenon going on now, that what’s relatively scarce is the power, right? And we’ve seen so many people talk about this, from Jensen Huang, who very charismatically says that the real limitation on our AI growth and on data center growth is power. And I think increasingly what we’re seeing, to Steve’s earlier point, is not just the generation. It’s really more the grid and how we both build the grid, but also the fact that we don’t do a very good job of pricing and managing our use of the grid. As grid capacity becomes scarce in some places, but it’s abundant in others, we don’t really do a good job of that coordination because we don’t use prices really well. So power has really become the limiter because it’s the thing that’s relatively scarce.

But you see things like in Texas, where big data center projects are being sited out in West Texas, right in the Permian Basin. And so the gas is right there to fuel the turbines, to generate the electricity, to run the data centers. And I also have to give a pitch in here for one of my favorite companies, which is Crusoe, because not only are they, you know, siting data centers right out where the gas wells are, but they’re also buying gas that would otherwise be flared because there’s not a whole lot of storage and there’s not enough pipeline capacity. And so, you know, gas producers end up having to flare stuff, especially if it’s gas co-located with oil — the gas comes out, and you can move the oil, but you don’t have anywhere to put the gas. So you just flare it, which is both economically and environmentally costly. And so if Crusoe goes and buys your flare gas, that’s a win-win economically and environmentally. So we see a lot of what Steve described in terms of the construction and labor shifting to the power being the main constraint, but also the same underlying driver for why you get this agglomeration in Texas.

If you are to spend your time trying to do corrections in the market, are you spending a lot of time in front of state PUCs, local city councils, the legislature? Where does the energy need to be done, from the human capacity, to fix the energy problem for the data centers?

Lynne Kiesling: I think it’s probably all of the above, but my work tends to focus most on public utility regulation — so at the state level, but also touching on the Federal Energy Regulatory Commission, because data centers are… and it’s complicated for several reasons. Number one, data centers tend to connect to the grid at the transmission level, and technically speaking, the transmission grid crosses state boundaries, right? So there are organized wholesale power markets. So, for example, in Virginia, Virginia’s in PJM, as are Maryland, Pennsylvania, West Virginia, et cetera. And so that transmission, that market, and the grid that allows for that market to transmit and deliver on the transactions, crosses state lines. And so it’s under federal jurisdiction.

But at the same time, that investment the utility is making, because the utilities own transmission and distribution, transmission and delivery wires networks. And so it’s both a state regulatory jurisdiction and a federal regulatory jurisdiction. And right now, there’s a bit of a turf fight going on, because in November, the DOE issued an ANPR, or advance notice of proposed rulemaking, to direct FERC to do some rulemaking around large load interconnection to try to standardize it and streamline it. And this is going to have a big jurisdictional fight between the federal and the state regulators because the state regulators very jealously guard their federalism. And so that’s one place where it’s going to show up. And it’s really complicated, because the incentives are just so murky in ways that we probably don’t want to get into — it’s deep in the weeds.

But it’s really, really difficult to get new transmission capacity built. Building new transmission lines is very difficult, whether it’s siting, permitting, or just the utility regulation parts of it. And so, in my work, what I look at is the more demand side, right? How can we get better pricing? How can we use markets to enable better utilization of the existing grid that we have? Because the average grid utilization rate across the country is only like 60 % or something. There are huge amounts of excess capacity on average.