The Future of Electricity Demand in the AI Era

Electricity demand growth has returned with a vengeance in the United States due to an increase in manufacturing and, most dramatically, the growing use of AI. Across the country, technology giants are racing to build AI data centers, the largest of which will consume as much electricity as an entire mid-sized city.

Yet our electrical grid was not built with such large and immediate new sources of power demand in mind, and it has become clear that solutions are urgently needed if our grid is to successfully accommodate this new load. Adding to the challenge is the fact that forecasts of future demand have been frequently and dramatically revised upwards. The future of electricity demand looks big, but just how big remains uncertain.

Rob Gramlich, president of power sector consultancy Grid Strategies and a frequent expert witness on grid issues before Congress and regulatory agencies, explores the future of electricity demand. Gramlich discusses data from a new Grid Strategies report on the pace of demand growth, and a variety of strategies by which our electric grid might meet that demand. He also considers implications for the cost of electricity and the pace of grid decarbonization.

Full Transcript

Andy Stone: Welcome to the Energy Policy Now podcast from the Kleinman Center for Energy Policy at the University of Pennsylvania. I’m Andy Stone.

Electricity demand growth has returned with a vengeance in the United States, with the reshoring of manufacturing and, most dramatically, the growth of AI. Across the country, technology giants like Google and Amazon are racing to build new AI data centers, the largest of which will consume as much electricity as a mid-sized city. Our electric grid was not built with such large and immediate new sources of power in mind. It is increasingly recognized that solutions are urgently needed if our grid is to successfully accommodate this new load. Added to the challenge is the fact that recent electricity demand forecasts have been repeatedly and often dramatically revised upwards. The future of electricity demand looks big, but just how big remains uncertain.

On today’s podcast, we’re going to explore power demand and efforts to address it with Rob Gramlich. Rob is President of electricity sector consultancy Grid Strategies and a frequent expert witness of grid issues before Congress and regulatory agencies. He’ll discuss data from a new grid strategies report on the pace of demand growth. He’ll look at policies that might help our electric grid meet that demand, and he’ll discuss implications for the cost of electricity and the pace of grid decarbonization. Rob, welcome back to the podcast.

Rob Gramlich: Thanks, Andy. It’s great to be here.

Stone: So your recent report highlights rapid demand growth on the US electric grid and most notably, the rapid upward revision of demand forecasts. Could you put into context the magnitude of demand growth that we’re seeing, and why are demand forecasts being so frequently and substantially revised upwards?

Gramlich: Yes, a little more than a year ago, we put out the report called “The Era of Flat Power Demand is Over.” At that time, most people didn’t realize that was happening. So the forecast doubled then, and then just a couple of months ago, we updated that and found, in fact, the forecast had doubled again. So we’re looking at something like 130 gigawatts of new demand in five years, so that’s 15% in five years, or 3% a year. And so 3% a year is back to 1980s-level load growth, and we hadn’t had that since then. We’ve been going along at a half a percent, so a six-fold increase in annual power demand. It is a bit of a shock. It’s coming very suddenly. I think most of the world is now well aware of this dynamic. I do want to emphasize that there is still a lot of uncertainty, so that’s part of the upward revision. Another part is just as the utilities factor all the different factors into their own forecasts, and then those get rolled up to national numbers, more and more growth gets incorporated in. And so here we are. That’s what it looks like, and I think the next year or two are going to be all about really trying to reduce the error balance, both with specific utilities, as well as regionally and nationally on these numbers.

Stone: And again, we’re looking at about 3% national growth now. Is that right? That doesn’t sound like a lot, but again, from where we’re coming, that is quite a substantial uptick, right?

Gramlich: Yes, I know. When you talk to people outside the electric industry, 3% annual growth doesn’t sound that great, but when you’re talking about 50 to 80-year assets you have to build to serve it, that are very expensive, that’s a big deal.

Stone: So this new demand is coming at a time that is already challenging for the grid, where there is concern that power supply can’t keep up with demand, and there are also related concerns around the future reliability of the grid based upon this. Could you describe a little bit more where that backdrop into which all this new demand from industry and data centers is arriving?

Gramlich: First of all, with the backbone transmission network we inherited, most of us who are professionals in this space came to it after there had been a huge grid build-out in the ’70s and ’80s. When I got involved in the ’90s, you just assumed there was enough transmission capacity and didn’t have to worry about it a lot. Well, that has all been used up, right? So the grid constraints are now well known, and they’re significant, and they affect all sorts of things, like the generator interconnection queues, where every new generator sort of becomes the straw that broke the camel’s back, and a big network upgrade is needed to connect them. And then that creates delay and cost.

So that’s the grid infrastructure, and on the generation side, it has just been slow connecting all the new generation, partly because of interconnection queues, but also the market didn’t think we needed a lot, and there wasn’t much firm generation being built in these markets. So now there’s obviously a very strong market signal in capacity markets and outside of the markets in the vertically integrated world and the integrated resource plans. Those are hot right now. There are lots of generation plans going on there, which again, is sort of new. So both in the market area and non-market areas, there is a hustling up of ways to meet the supply growth needs.

Stone: You’re talking there about the areas of the country where there are competitive markets for wholesale electricity, and also kind of traditional, vertically integrated utilities?

Gramlich: That’s right, yes. The Northeast has the competitive generation areas, plus Texas and California. And most of the rest of the country is the vertically integrated model, so you have a different process to go through for the generation expansion needs.

Stone: As I think we’ll get to a little bit later, the strategies maybe for addressing all of this new load in those two types of areas might be different, as well. I think we will get into that.

Gramlich: Right.

Stone: So load, I want to point out, is not growing at a uniform pace around the country. There are certain regions, as you point out in your report, that are experiencing much more rapid growth in electricity demand than others. Can you just give us the map of where are these areas, and to what extent are the supply and reliability challenges that we’ve begun to talk about intensified in these areas?

Gramlich: Right, there are hot spots for new large loads which includes manufacturing, but certainly include data centers prominently. The data centers traditionally have really wanted to just be near the fiberoptic cable network, so Loudoun County in Northern Virginia has always been the biggest, but also Dallas/Ft. Worth is attracting a lot. Silicon Valley, California and other places have had that, but more recently, it is really driven by power, as much or more than the fiber network. So other areas open up. The Atlanta area, Georgia Power had nuclear power coming online. They had some power availability. Texas is a relatively easy place to build generation and transmission, so that was attractive for new large loads, as well as crypto, part of the new large loads. Parts of the Northwest, Arizona, and some other areas, Illinois, are other areas that come into that picture. So yes, some of these areas are really getting a lot, in some cases maybe more than they can handle. I should have said Columbus area, Ohio. That’s getting a lot. So it is not uniform. There are certain areas that are attracting more because they had some ability to take it in, but they all have limits. They are all trying to deal with more demand for large load interconnection than they can handle.

Stone: So the 3% figure that we talked about a few minutes ago for the national load growth, that’s when you average it around the country. But when you’re looking again at these places like Virginia or Texas or the Atlanta area, those numbers, I would imagine, are significantly higher.

Gramlich: That’s right.

Stone: An interesting point here is that AI has very recently been framed very much in crucial economic and in geopolitical contexts. It doesn’t really seem like it’s feasible to simply hit the brakes on data center development at this point, as well as manufacturing development. Can you explain that backdrop in which it looks like this stuff is going to happen, regardless of whether the grid is ready for it or not?

Gramlich: Right, so a lot of states and more local authorities are really trying to attract these data centers for jobs and tax-base reasons, economic development reasons. So that is driving the various levers at those agencies and states to try and make it work. Public utility regulators are trying to enable and accommodate what their state policy may be to attract these and encourage utilities to do what needs to be done to connect them. It is also, I think, relevant whether it’s sort of a national priority to win the AI race against China. We’re certainly seeing a lot of comments to that effect from the incoming administration, but even before then, there was a top activity of the closing weeks and months of the Biden administration to accommodate new data center load, to do AI here. Certainly the national security apparatus here in Washington, DC is very interested in getting as much of that work, those machines, processors, in US-based data centers, rather than abroad.

So that affects our ability to do that. When you get tough questions about permitting a local transmission line to serve that need. It’s  hard for policy-makers to approve those things, but when you have this overlay of very high level governor and White House national policy pushing to get things done, then you can see that some of those tough local questions can get resolved, and not everybody is going to be happy about it, but things can get permitted, and infrastructure can get built.

Stone: I’d like to take just a step back here for just a moment, just to put this into further context. So as your report points out, we have seen rapid electricity demand growth before, and actually much higher demand growth than we are experiencing or are expecting to experience over the next few years. Looking at the ’50s and ’60s, you point out how high that had gotten. What makes addressing load growth, demand growth so difficult by comparison today?

Gramlich: Yes, that’s right, and that was, we hoped, one of the unique contributions that our grid strategies report could provide here because I think it is relevant that we’ve done this before as a nation. Three percent, actually, isn’t that high. It’s only half as much as in the ’50s and ’60s, so we can do it. But what has changed is we just have different processes for permitting everything these days than we did in the old days, right? Everything is harder, from housing to you name it.

We also, I think, lost a lot of muscle memory in the electric industry to do this. I call it a “load growth mindset,” which we haven’t had for decades, and so the people who built their careers doing it in the latter half of the last century are just not around. And then electrical equipment and things like that are in human resource capability. Talent is an issue. We don’t have the electrical equipment. There are so many engineering schools that don’t even have power systems engineering anymore. These are problems, and these things are slow to fix, but on the other hand, they are fixable. We have shown as a country that we can meet this type of growth. We just have to accelerate those things, accelerate supply chain, accelerate trained people to do the work. I think that’s one of the reasons we got out there, to try to raise the alarms a little bit about this, of all the things that need to happen to meet this type of growth.

Stone: It’s interesting you bring up the issue of nuclear and of manpower and know-how. My understanding is that one of the issues that the nuclear industry has faced in recent decades, as projects have moved forward or not moved forward, is the workforce of the ’70s and ’80s, that was building so many nuclear power plants, no longer exists. And that expertise has to have been rebuilt, at least for the large, traditional plants.

Gramlich: Yes, there are a lot of similarities between transmission and nuclear right now. In the ’60s and ’70s, we did a lot of transmission, and we did a lot of nuclear. We had supplies. We had workforce, et cetera. Neither of them did very much for the 50 years after that, but now we need them both. And now they’re both experiencing a huge tailwind. The market is telling them, “We need you, if you can get in here and help.” Both of them, of course, have their challenges, costs included. But there is great interest in doing both transmission and nuclear, two things we’ve done historically but haven’t done for a while, but the market wants them.

Stone: So there are two parts to this challenge here, just reiterating what we’ve been talking about. We have all this new demand coming online, or that wants to come online. That’s growth on that side. And then you also have to grow the grid, to accommodate the generation as well as the physical power line infrastructure to do that. And that’s kind of the Catch-22 we’re looking at, at this point. So much new demand the grid can’t accommodate. The drivers of this demand, the technology companies, are looking for ways around this. One option, referencing nuclear, is to bypass the grid altogether. So a high profile, recent example of this is the Three Mile Island power plant here in Pennsylvania, which was recently closed. Its owner, Constellation Energy, has signed a deal with Microsoft for nuclear colocation, or colocation of the data center with the nuclear power plant. Can you tell us what is colocation, and why are these types of arrangements potentially attractive to the tech companies?

Gramlich: The data center companies in particular have a need for speed. They want to get connected very quickly. They are in a race with each other, and we are in a race with other countries to connect these data centers. So when they’re looking at all these constraints and utilities around the country saying, “Sure, we’d love to have you connect. How does 2031 sound to you?” And they say, “No, how does next year sound to you? And please, how do we get connected quicker?”

Colocation becomes one of these creative shortcuts. There are a whole lot of creative shortcuts that these very smart, well capitalized companies in the hyperscaler and other data center sectors are looking at. So it shouldn’t at all be a surprise that looking at on-site at a nuclear plant is appealing, if you can get the grid connection through that, because obviously they have a big grid connection already there, then that seems like a great opportunity. I expect we’ll see more of that. The thing is, there are just not that many opportunities for that. Closely related to that is a nuclear expansion or bringing a nuclear unit out of recent retirement back online.

So those, I think, we’ll see, as well. Also not really a huge solution, not a nationwide solution to the demand, because there are so many, so limited opportunities. Also within the colocation category are on-site gas or a sort of hybrid solar storage/gas options that I think will be pursued. Those are probably the next quickest opportunities. I think we’ll see plenty of that. I don’t want to overstate the colocation story, though, because I really don’t think any data center developer or any other large load for that matter, wants to be off-grid. There’s a lot being made of the off-grid opportunity. It gets a lot of press, and there has always been this sort of mystique about going off-grid. I don’t think anybody wants that who is in the business of any large load. The network provides such great value to them, and when you look at the options, you realize that any particular supply source that you might have on your site or anywhere else, has a forced average rate, or it needs refueling, or it has other reasons why it might not be operating all the time. That’s why we have electric powered networks. That hasn’t changed. The network means every supply source balances every other supply source.

So every hyperscaler knows that there’s value there. They want to be connected to the grid. If they have to do some things in the very near-term to reduce their reliance on the grid, we can talk about that. They may do that in the near-term, but it doesn’t mean they’re off-grid for any significant period of time. It means they’re in the process of getting the grid connection they want.

Stone: I don’t want to over-emphasize this point, either, but it’s interesting that Mark Christie, who is now the FERC Chairman as of January, has named colocation as a priority issue that the FERC needs to get resolved this year. So there obviously is some pressure for these types of arrangements. PJM itself, the  electricity market operator here in this Eastern part of the country, engaged in the stakeholder process to try to come up with rules governing these colocation arrangements, and who would pay for the grid. There were a lot of concerns around that, and they weren’t able to get to the endpoint. So there’s a lot of controversy around this.

Gramlich: Yes, that’s right. I would divide it into a couple of categories. I think the FERC case– I think if you’re a typical FERC staff, just looking at the cases coming before you, if somebody wants to plug in a new large load or whatever they’re doing, and they might use the grid. They are grid-connected. Their power source could be drawing large amounts of power from the grid in certain situations. You kind of look at it as, “You know, you’re responsible for paying grid costs. You’ve got to pay your share of the infrastructure, just like everybody else.” So you can’t just say, “Well, if most of the time I’m not relying on the grid. I don’t therefore need to pay for it.” That’s just not how cost responsibility works for the transmission system.

So I think for the actual grid, the transmission component of cost, then these new large loads are going to pay for that. There’s a separate question about the capacity market, the generation part in these nuclear units that were selling into the wholesale market, but then would large be selling to the new large load. To me, I think ultimately this is about regions that have competitive generation. The generation owner can sell to whomever they want to. If they choose to sell to this new customer, as opposed to everybody else, that’s their prerogative. The tariffs and the protocols around capacity markets in certain regions make that not quite as simple as I just described, but I think that’s where it leads eventually. Anybody who has power to sell is going to sell to whomever they want to, and these new large loads are looking for those power purchase agreements, and so they’re going to find them. There is a question, I think, that then goes to the states, your state of Pennsylvania, as well as other states, of: “Oh, wait a minute. How come my normal residential rate-payers aren’t getting the benefit of this hedging or this procurement of power long-term? How come this valuable power is just now all going to these new, the wealthiest companies in the world? Why don’t regular people get to have that power?” And I wish they had asked that question three years ago. They could have procured it, right? There was power out on the market, and nobody bought it. So I think the states need to get better about hedging for regular people.

Stone: You just started to broach a new subject here, and I want to dive a little bit more deeply into it. This is a question of who is going to pay for all this grid infrastructure. You’ve got these large, very wealthy tech companies that are coming in. It’s interesting. I’ve been doing some reading, and they’re not even so concerned about the cost of energy. They just want access to it at this point because their businesses are so critically dependent on getting the electricity that they need. But there is, as you mentioned also earlier, there is some uncertainty around how much demand these data centers are going to drive. It is still to be seen. There is a risk, for example, that in anticipation of there being much new demand, the grid is built out to accommodate that over a number of years, and then that demand doesn’t materialize, but somebody gets paid for these semi or stranded assets, whatever you want to call them. Can you talk about the risks, who they might fall upon for these costs, and some of the solutions that have been proposed to make sure that those who are most benefiting from these new developments might actually pay for them?

Gramlich: This is a major area of work and activity, particularly at state commissions and with states now to avoid this scenario that you described well of somebody saying, “Oh, I need 100 megawatts to serve my new, very large load.” And then the utility goes and starts to build out the network to serve them. And what if they don’t materialize, or they end up using much less energy in the future for whatever reason? Then it’s like the existing rate base. The residential, commercial and industrial customers who are there are left holding the bag.

So it’s a classic regulatory risk question, and the job of the public utility commission is to have fair allocation of costs and shield regular consumers and rate-payers from such risks. So there are solutions out there, and I think the hyperscalers themselves, and the new large loads generally recognize this as a concern and are trying to manage it and mitigate. From what I see, they are operating in this regulatory process to avoid that, and they are agreeing to things like long-term financial commitments so they pay up-front, and if they don’t materialize or use the energy in the future, they still are on the hook to pay, which means the other rate-payers do not have to pay.

Now there are a lot of details in those, and there are settlements going on in AEP Ohio. I think Duke has a new policy. Georgia Power has a policy on this. Every hot spot is already deep into this and figuring out some form of that, where there’s a financial commitment. The other aspect of that financial commitment is that then you can really kind of determine how much of this load is “real.” We started this discussion talking about the error balance. We don’t really have a solid estimate on whether that’s 3% nationally or whether it’s 10% at some specific utility, or 1% or 20%. It’s still uncertain. But the more these large loads lock in their deal, and then the utility says, “Okay, this much load is locked in. This is very certain because they’re paying me, no matter what. So they’ve got real skin in the game.” Then you can really sort out the locked-in, certain load, versus the uncertain, more speculative. And then we’ll get better load forecasting, and that helps the whole process, the regional transmission planners included. We’ll have much better knowledge in the wholesale power markets, and participants in that will have better knowledge about what the actual load is.

Stone: It’s interesting. Just on that, you mentioned AEP in Ohio and Indiana. I don’t know a whole lot about this, but I actually have seen that in the press, and I think basically the deal there would be that the Googles of this world — not the Google specifically involved in that — but those types of companies would be locked into, say, 12-year contracts where they would pay for the grid, whether they actually end up using it or not. It’s kind of an assurance that the utility doesn’t build this, pass the costs onto rate-payers, and then the data companies just kind of walk away.

Gramlich: Exactly right. And very similar actions in the other hot spot areas.

Stone: I want to also emphasize that this uncertain future load really is — There is a lot of uncertainty, right? This recent appearance of DeepSeek and all the discussion about, “Hmm, maybe AI data centers won’t use as much electricity as we thought.” This again throws another big old ball of uncertainty into what we’re looking at in the future, right?

Gramlich: Yes, that’s right. It has definitely been a question mark. I don’t think it really changes the estimate with any confidence. I think it’s a question mark for everybody, but I think the implications of DeepSeek have been overstated by many. But it’s one of these things where what kind of quality information are we getting out of China generally in this particular instance, to know what really happened there. So maybe we’ll learn more, and then maybe that does have implications. The implications on the stock market, of course, were significant and quick, but I don’t think that necessarily means there’s a change in the power demand. The power demand has the classic rebound effect of they still want to consume as much power as anybody can possibly provide.

Stone: I want to jump to the environmental implications of all of this. We talked about colocation, and we talked about these new data centers connecting directly to the grid. Regardless of what happens, there’s going to be a lot of new power that’s going to be needed, again to serve all this new demand. And there are some creative solutions that are out there, in addition to what we’ve spoken about. For example, in PJM there’s something called the Reliability Resource Initiative, which would allow what they call “shovel-ready generation projects” to leapfrog in the interconnection queue ahead of resources that are already waiting to interconnect, to provide electricity as soon as possible. That interconnection queue in PJM and in much of the country is actually largely renewable energy — wind and solar and some storage. There is concern that the leapfrogging resources might be fossil resources, particularly gas. Could you talk about these concerns? And broadly, again, all this is happening at a time when the grid is struggling to decarbonize, and now we have this additional new demand. Where does this leave us on an environmental and climate front?

Gramlich: Broadly, I think unfortunately, we’ve got a bump in the road here on the decarbonization trajectory for the electric industry. There are, I think, unavoidably some delayed shut downs of emitting plants. Unavoidably there are some new fossil plant additions that are coming and likely associated pipelines. I do think almost all of these new large loads are very committed to their decarbonization goals, but that doesn’t mean that if they have the opportunity to get a new data center online right away to compete with their competitors and other countries, they’re not going to do it. They have a business imperative to get connected as much as they can, right away, even if that means some period of some number of years’ delay in their decarbonization goals.

So I think there are ways to manage that situation. I think first of all, renewables in storage are the most ready sources of new power. I think they should be accelerated. I think every interconnection queue around the country can quickly sign interconnection agreements with tens of new gigawatts each year. Compare that to PJM, that only connected a couple of gigawatts each year for the last few years. I think they could increase that by 20-fold, and they might, actually. Stay tuned on that.

So that’s an option. I also think there are some kinds of on-site gas turbine with solar PV-type options that can quickly connect, that whether the gas turbine isn’t operating all that much. Maybe it is in the near-term to get them connected, but long-term it doesn’t necessarily need to, which isn’t a terrible emissions profile. So there are ways to manage it. Of course longer-term, we could talk about small modular nuclear reactors. I don’t think we’re going to see those before 2032-ish. But longer term, there are some options there that would be carbon-free. So that’s an option.

You also mentioned the RRI, the interconnection queue policies that might connect so-called dispatchable units more quickly. I really fear that in our zeal to connect things quickly, we might break the law, and that doesn’t speed things up. That just means years of court battles. Let’s say a gas-combined cycle in PJM, in your region, Andy, has a 70% capacity value. That means its contribution to the capacity resource adequacy needs, based on how reliably it has operated. And let’s say solar and storage hybrid plants have a 69% capacity value. So they’re basically the same in terms of their contribution to capacity. And then let’s say you artificially define the gas units as dispatchable, and the hybrid unit as not, when in fact they are dispatchable. That’s illegal. That’s discriminatory. And it doesn’t serve the reliability need to say that gas plants get to go forward and jump the queue ahead of the hybrid plants. That type of thing can cause real damage to the long-term integrity of competitive markets, if you start offering discrimination opportunities to certain resources on the basis of an artificial label that has no engineering meaning.

So it’s obviously in a context of a weird political time right now. Certain FERC commissioners and policy-makers in Congress and Washington have different fuel preferences than many of the states have, but that’s definitely one to watch, in terms of whether the interconnection process remains non-discriminatory.

Stone: Yes, and this selection of types of resources is not just potentially happening in PJM. For example in Texas, Texas is offering 10 billion dollars in low-interest loans for the bid out of new gas-fired generation specifically. So we’re seeing these types of things happen in a lot of different places.

Rob, I want to ask you a final question here. Obviously we are under new political leadership in this country at the federal level, and we’ve also seen a change at the FERC, where Mark Christie, as I mentioned earlier, is now the new Chairman of the FERC, put in that position by President Trump. Recently the FERC has ordered some required changes in the way transmission is planned in this country. I want to get your general thoughts on movement to support these changes going forward under the new administration, whether the new administration seeks to build out the electric grid with the same priority that the Biden administration had placed on this. What are your thoughts?

Gramlich: Transmission is extremely important, and I’m certainly one of the folks watching very closely every utterance of the new people in charge, as to where things are going to go. I guess I’m cautiously optimistic at this point. The Trump executive orders didn’t call out transmission specifically one way or another, which is probably a good thing. It’s probably good to fly below the radar in this environment.

But Secretary Wright has said a number of very positive things about transmission in this confirmation hearing, and then in his statement is a written memo that he issued upon taking the reins over the Department of Energy. And one of the pillars was transmission expansion. He said something about looking for every way to help expand the grid. So I think that’s potentially very positive. And then over at FERC, which controls some of the key policies on planning cost allocations mainly, and then to some extent could even be permanent. Chairman Christie has been a bit of a skeptic, but in a low-growth area and where transmission benefits all new generation, and it’s open access and technology-neutral. I don’t think he’s opposed, and he did vote for Order 1920-A, and it was important to him to get some major changes to it, to involve states in that. But Order 1920-A is the regional transmission planning rule that now requires that a proactive, long-term planning to meet new growth and other needs.

So one would expect that FERC carries that through and rules on the compliance filings, and that regional transmission organizations, including PJM, get the memo. This is now the rule they must comply with, and they need to start doing these things. So I’m hopeful. Now, there are five commissioners, and as soon as any one of them leave for any reason, then there’s an opportunity to appoint somebody new. So over time, the make-up of agencies like FERC does change. We’ll see if things go a different direction in the future. But for now, things look pretty good at FERC and the Department of Energy.

Stone: Rob, thanks very much for talking.

Gramlich: Thank you, Andy.