Are You Subsidizing AI Data Centers?, w/ Ari Peskoe

Speaker A: Foreign. Hello, and welcome to AI Curious. My name is Jeff Wilser. I'm a journalist, I'm a human, and I'm curious about AI and let's just say it plainly. Of all the things people are anxious about with AI the lost jobs, the deepfakes, the sci fi doom, the fight that's actually boiling over in real towns right now is, of course, about data centers. It may be the single biggest political lightning rod in all of AI One of the rare issues that unites the left and the right in every corner of the country in shared fury, really. And here's the part that makes many people furious once they see it. When you move into a new house, you pay for your own power. The meter runs, the bill comes to you, and nobody else on your street sees their bill go up because you had the nerve to move in, but now swap yourself for a data center, a hyperscaler. They drop a facility into some rural county, and somehow the person whose bill climbs isn't just a trillion dollar tech company. It's others. It's a retiree three towns over. Which begs the question, are, uh, private citizens like you effectively subsidizing part of the AI buildouts? My guest today would say, effectively, yes, and that's Ari Pesko, who directs the Electricity Law Initiative at Harvard Law School, which makes him one of the few people who can explain all of this clearly without either cheerleading or bursting into flames. We get into all of it. The size, the size, scale and scope of the problem. The interstate power market, where the tab hit $63 billion and data centers account for about 30 billion of it. The cruel little twist of marginal cost, where the last most expensive electron sets the price for everyone else. And the fact that back in the 1950s, a new factory actually lowered your bill. And why that math has flipped. It's a fun episode on, um, a topic that everyone has a lot of opinions on, and we try and cover it and hopefully a measured, balanced way with that. Please enjoy my conversation with Ari Pesko, director of the Electricity Law Initiative at Harvard Law School. Welcome to AI Curious. I'm sure every listener of his podcast knows, like, oh, wow, data centers are just popping up everywhere. People are upset about it. I saw some study that 71% of Americans are like, do not want data centers or are upset about them. It's maybe the one issue uniting the left and the right in all corners of the country, which is really rare. Um, so there's a sense that it's a big problem, um, and taking lots of energy. Can you Give us a little more of a sense of, of scope and parameters here. So how much of the grid are these data centers actually gobbling up and how fast is it changing?

Speaker B: Yeah, so data centers have been with us for decades. They're the backbone of the Internet. There are thousands of facilities around the country that call themselves data centers. You can qualify them as data centers. Uh, what's changed in just the past few years is the size of these facilities in terms of how much energy, uh, they use. And just to give you an example, uh, the most extreme example, meta, uh, is building a facility down in Louisiana that will be capable of using 5 gigawatts of electricity. A nuclear reactor produces 1 gigawatt of electricity. So you're talking about a single facility that's, you know, would, would need five nuclear reactors, uh, to power it. And that's, that's perhaps the largest energy consuming facility in the country, uh, that I, that I'm aware of. Once, uh, it's, once it's actually complete,

Speaker A: just to give a sense of scale. So that would be more than like several city blocks or more than like

Speaker B: much more than that. So the city of New Orleans, biggest city in Louisiana on a hot summer day when everybody's got their air conditioning going full blast. Electricity demand is at its peak. New Orleans uses about 1.2 gigawatts of electricity. So you're talking four to five cities of new Orleans popping up in a very rural part of the state that just hasn't had anything like that, uh, ever. That's the most extreme example. But we're seeing large scale facilities that are hundreds of megawatts, even, you know, a gigawatt, uh, popping up in several different parts, uh, of the country. And so when you have a basically new city in terms of energy consumption suddenly, uh, come to fruition, you're going to need a lot of infrastructure to make sure you have enough power, uh, being generated as well as to deliver all that power, uh, to that new site.

Speaker A: And I think it's useful for additional context to kind of talk about how quickly this has been growing and how likely it is to grow in the future too. Right. Because just for even a wider lens here, we all know that in the last several months, um, the amount of tokens that companies have been burning has just exponentially increased as we've entered the quote unquote agentic work era. Right. So just to back up even further where a year, a year plus ago, most AI usage, um, in terms of how many tokens were used, were for like chatbots someone saying, hey, how do I make this recipe for jambalaya? Right? And that is shifting to companies saying oh, um, this agent, uh, can code our entire, um, this entire app can be built with AI and so it hasn't increased tokens by 10% or 20%. It's like 5x10x100x. Right. And that directly requires more compute, a lot more compute. And that more compute requires more data centers and long winded way of saying there's still uh, most people, the average person using AI is still not really using AI agents yet. They probably will, they probably will get more mainstream in the future. More and more companies using more agents. So this is not going to slow down. We're going to continue to need more compute, more data centers. And some experts I've spoken to have said it's we're uh, in the area, we're getting around 5% of the grid being used by data centers and it could potentially grow to 50% by 2040. I've heard that estimate, um, by focus study this, that it could be as much as half of the grid. Is that kind of how you're seeing things as well, Ari, that like this is, it's likely there will be a, um, aggressive continued uh, growth of how much power is being consumed.

Speaker B: I think the uncertainty here is a huge challenge for the power sector. Um, you look at the history of the power sector, it's a story of growth for 100 years. Uh, and then that growth basically stopped around 2008. We had some turmoil in our economy. Um, I think the, you know, end users were capturing efficiencies. The replacement of your old uh, incandescent light bulb with LEDs, uh, as an example, you know, a lot of manufacturing was going overseas and from around 2008 to around you know, 2023 or so on a national level, our electricity consumption was flat. And that was kind of the world that the industry was largely expecting. It wasn't planning for growth. It really didn't see this boom in energy use coming. And so the past couple years have really been a scramble, uh, to try to build faster, get projects permitted. And at the same time they're being hit with supply chain issues. And some of those go back to Covid era, some of them just because they hadn't been planning on this growth. And you can't just ramp up to build up a transformer, for instance. It takes time to put those kind of factories uh, in place. And so there's a number of factors that make it hard for the industry to meet the power industry to meet this demand growth, uh, for uh, data centers. And what's the future of growth going to be? I mean there's certainly a solid amount in the pipeline right now. I think it's hard that there aren't a lot of great national, uh, national estimates are kind of, they vary. We uh, know there's significant growth. We know that. The best estimate I've seen is, is before this growth started, data centers were using about 4% of U.S. electricity. 50% seems high, uh, to me by 2040, just because that would require so much construction from the power industry to meet all that. Uh, but certainly it's, it's increasing and uh, you know, there's a lot of factors that are going to go into play here as to whether how high it's going to go and is it going to plateau in 10 years, is it going to keep going up? I mean there's a lot of uncertainty and that, that actually creates a lot of trouble for utilities and for consumers.

Speaker A: Before we go further, um, that's interesting that you mentioned that the demand has been roughly flat. Can you explain that a little bit? Because just as a layperson here, I would imagine given how much population growth we've had, technological improvements, there's, we're, we're doing more things, we own more things. A lot more happening in 2026 than in 1926 is it, is the reason why we've gotten so much more because we've gotten more efficient overall. The power demand has stayed about the same because even though we're doing more things, are more efficient. How would you characterize this?

Speaker B: Yeah, so just to clarify, I mean the, the flat period is roughly from 2008 to 2023.

Speaker A: Okay, sorry, got it, got, got it.

Speaker B: Uh, so the last eight.

Speaker A: Okay.

Speaker B: You know, roughly 15 year period and flat on a national level. So certainly there are regions where there is great population growth, where there can be more industrial growth in those regions and there may be a slight uptick, but that's you know, roughly balanced out by other regions where, where you know, there's been declines. Um, and you know, I said it's efficiency gains, um, and it's just lack of industrial growth. Uh, generally, um, I think those, those are two of the main, uh, factors. Um, and you know, I think that, you know, historically growth in electricity tracked economic growth and we've seen sort of a split there where our economy does continue to grow, but it no longer correlates or at least didn't uh, correlate with growth in electricity consumption.

Speaker A: M. Um, one, I ah, want to, I want to get to your, your work and your studies, um, in a moment. And one way to ease into that just some, I think, let me try and describe how maybe a curious listener or even myself might imagine this in a vacuum. Not knowing how it actually works. I'm thinking, okay, cool. If these new data centers are entering the grid, they should pay for additional consumption. Just as if I were to like, give me a very simple, probably flawed thought experiment here. If I were to move to Cleveland and I bought a house in Cleveland that would consume power, and so I would happily or unhappily pay my power bill in Cleveland. And I'm take, I'm doing my part to pay for my use of consumption, but I'm guessing that's not how it's working here in the data center lands. What's, what's actually happening. And also let me know if my analogy is flawed in any way.

Speaker B: Um, so there's two aspects of the utility business model we have to understand in order to describe what's happening here with data centers. Um, and first, what is a utility? Utility is a company that owns the local wires that go up and down your street that connects to your home, and they have a monopoly on delivering that electricity to your home or business. So that's one aspect is that they are a monopoly. The second aspect to know is that when they spend money, they basically recover their costs from everybody that gets service from them. So they socialize their costs. Um, and if you're moving to Cleveland and you're going to move into a house that's already there, utility probably doesn't have to do anything at all to serve you personally. There was presumably somebody living in that house before switch the name on the bill. That's. Even if you build a new house, you know, a new house doesn't use that much electricity, there's probably enough infrastructure already there to deliver whatever electricity you want to that site because there's not a lot of spending they have to do. And even if they did have to upgrade some local component because your house is going to use a lot of electricity, let's say, um, they're going to share that cost with every other consumer that they're serving. It's infinitesimally small. Even if they did have to make a very small upgrade, you're going to talk about a few thousand, several thousand dollars divided by whoever, you know, all the people of Cleveland, right, with a data center. That's not the case because these, these, there's not Enough spare capacity on the system, often to generate the energy, but certainly to deliver the energy. Because you're talking about again, plopping down a, basically a new city in terms of how much electricity it's going to use. You're going to need a lot of, put it this way, sort of thick pipes. But you know, in terms of electricity, large wires, uh, to move all of that electricity to this new facility and these, you know, so you can spend hundreds of millions of dollars in the wires and potentially billions of dollars in building, let's say a new power plant to serve that data center.

Speaker A: And, and just to restate this up,

Speaker B: if, if you sp, if you. The traditional utility business model. Uh, sorry, go ahead. I was saying, you know, the traditional utility business model is again, spread those costs to everyone.

Speaker A: If I hear you right, Ari, it hasn't been as noticeable in the, my silly example of buying a house in Cleveland. Or let's say to make it a little less silly, maybe a whole new neighborhood pops up of like a thousand houses and that might require a bit of infrastructure build out for those thousand homes in the new neighborhood. That it would be, I guess like the logic is it would be punitive and unfair to ask the new home buyers and residents of this new neighborhood to shoulder the burden of the entire cost of infrastructure for those homes. That's, they would have like thousand dollar power bills a month as opposed to a hundred dollars or whatever it is. That's not fair. So it's divided evenly over everyone and no one really notices or cares that much. However, we're now in a world where it's not a thousand new homes in Cleveland. It's as you put out this 5, you know, 4x the size of Louisiana energy consumption. These massive, star esque, um, warehouse type, massive sprawling things plunked down that requires tons of capital to build out. So when they're adding to the grid, even if it's only 5% of the total grid consumption, the average citizen's um, bill isn't going up by just 5%. It could be going up by a lot more than that because of the massive infrastructure cost that the utilities are forced to build. Am I summarizing that properly?

Speaker B: I think that that's, that's basically fair. But let me, let me just play with your hypothetical a little bit more now that you've changed it to building a whole new neighborhood. So in that case, the utility probably would charge that developer that's building a thousand houses for some of the local infrastructure that's needed just for those Houses. Right. Maybe for the uh, wires that go around this new neighborhood. It's possible, I don't know exactly what the deal would be, but they would be charged something for that local infrastructure, but they would not be charged for if there are any upstream costs on the system. Right. So if, let's say there need to be upgrades to the higher voltage, the higher capacity delivery system, if the utility even needed to build a new power plant because it was constrained and it needed a little more energy, typically the developer of the homes wouldn't be charged that. And so in a lot of cases the utilities started this data center build out process with that same idea in mind. What they're going to charge the data center for the, you know, the local connection is going to be the data center's responsibility. There's going to be, and that may be 10, $20 million because it's a really big facility. It may need some uh, uh, equipment to connect to the network. But if we need to build larger transmission lines that are going to be upstream of that we have to build a new power plant, then we're going to spread those costs to everyone. But very quickly consumers caught on to what was happening and it started to snowball and be part of this larger pushback that you referenced earlier about data centers. And so Ned, what we've seen is some utilities changing their approaches, I think often inadequately, but certainly better than where they started from initially.

Speaker A: Can you give us a sense of the scale of these costs and maybe if you can, how much um, on a, per our energy costs in communities where data centers are being just kind of plopped in, are costs going up by 10%? Are they doubling? Any sense of, kind of what we're talking about here?

Speaker B: Yeah, it's hard to, I can't generalize because these are really local and often regional, uh, uh, issues. So a national average would be kind of meaningless. I can give you a couple of sort of anecdotal data points. Um, one is, you know, the hub of data center development is Virginia, um, and that it's been that way for uh, decades. The legislature there commissioned a study that was released in 2024 where they hired a, uh, you know, industry consulting group that did a study that said well basically nobody right now is really paying for data center energy costs. But if we continue the current rate structures and we see the growth that's going to happen with data centers, if we don't make any changes by 2040, the average person or the average household in Virginia could, could be paying as much as $400 a year in data center related costs. Um, so that's an extreme, I think example, uh, that is motivating some reforms uh, in Virginia.

Speaker A: I'm glad you called out that my thousand person home theoretical is substantively different than me moving to Cleveland. Personally I'm wondering now it's kind of on that continuum from on the far end of it, me moving to Cleveland, one house on the other end of the continuum, um, the data center, you know, 5 billion gigawatts drop down. Something occurs to me might be more kind of um, a little bit more fair. Fairer. Um, apples to apples is if a company um, builds a massive new office building. Right? So like when Exxon moves at headquarters to Houston and puts up this huge building or campus, um, how is that typically handled? And what's typically considered fair is Exxon uh, typically bear the brunts of that additional infrastructure builds. Is it taxpayer subsidies, is it our residents, they just shoulder that and kind of suck it up across the board. What's the usual method and how is data centers difference or the same as the usual approach when companies create a structure like that?

Speaker B: Yeah, I think actually again push back really on your hyper. I think a better comparison might be something like uh, aluminum smelter or you know, some giant energy using industrial uh, facility. Um, um and what often happens there is that the utility will negotiate a deal with this new factory. Um, and that contrasts with how rates are set for everybody else. For the rates that you and I pay, those are set in a state regulated process by a specialized agency called a public utility commission that basically sets uniform rates for a particular type of ratepayer. So all residential ratepayers pay the same rate. All, all commercial customers pay the same rate. But when you have one of these really large facilities, many states have specific laws that will allow special rates to be negotiated. Often, you know, based on the idea that these sorts of facilities can bring a lot of benefits to a community and we want to attract them uh, to our locality. Um, and states have different approaches for how they require the regulators to evaluate these contracts. In some states there has to be a showing that there is no kind of subsidies from um, other ratepayers to that facility. In other words, that facility is really paying for all the energy and infrastructure it needs. That's, that's problematic for reasons we can get into. And then in other, other states they actually explicitly allow subsidies based on, you know, the idea that economic development is good and so everybody should subsidize through their electricity rates. Economic development, uh, tax breaks that you mentioned. Those are totally separate. Right. That's a different pile of money. Uh, and so that may be the lot of data centers are getting those uh, just as many industrial facilities do, uh, as well. But that's a totally separate uh, issue from the utility rates.

Speaker A: All right, thank super helpful broader context. Thank you for that and walking down the kind of those basic questions. Now let's get into kind of the, the red meat of, of your study and your work. What recent paper that came out kind of describe for us. I love the title of it, the title of it and kind of the main thesis of your, your focus.

Speaker B: Yeah, so it's called Extracting Profits from the Public. Um, it's about how utility ratepayers are paying for big tax power. And the goal of the paper was really just to explain how it is that your, the rate that you're paying, your bill, uh, is effectively paying for infrastructure that is being built for these data centers or is being paid for energy that is being supplied, uh, to these data centers. And we've kind of already hit on the basic mechanics of it, which is that the traditional utility business model is to socialize costs. Um, and you know, we get to a deeper level in specific types of costs because there are ah, ah, complex regulatory mechanisms both at the federal and state level for different types of costs. And so we explain how different methods, uh, for sharing costs, uh, can lead to everyone paying for data center, uh, infrastructure. And in addition, in many parts of the country, your utility is actually buying energy from an interstate market, uh, that's regulated by the federal government. And there's one market in particular, uh, in a region called pjm, which stretches from New Jersey to Chicago down to North Carolina, uh, where prices have spiked over the last few years due to data center demand in the market. Because in any market when demand for a product goes up and supply can't keep pace, prices shoot up. That's sort of like econ 101. Uh, and we, we're seeing that in this m. In this particular market and consumers are paying a lot of money. Uh, uh, just to give you a number here, the past few years the, the, the total cost of this market has been $63 billion and data center demand is responsible for about 30 billion of that. But they're not paying 30 billion. So the cost is being spread to everyone in the region.

Speaker A: Can, my understanding is that the, the uh, marginal costs and marginal economics are part of what's going on here too. Right. And can, can you let Me know, I'm going to give a 15 an awkward 20 seconds my understanding of it. Let me know where I'm getting it right or wrong is that let's say if, um, uh, if some, the grid has a certain costs as it is a baseline cost, and then if this next. So the grid's full, someone else comes into the grid and requires more demand and energy, the price goes up. Let's say it goes up a lot. Like because we've already used the cheaper. We first use the cheapest energy. Right. We have. We use great. There's, there's wind we can use, there's solar we can use cheap stuff. And then we kind of get to the energy that's more expensive at the margin because we already cheap stuff. It costs more to service the last kind of folks in the grid to get energy. And the, the, the kicker here is that new marginal high cost applies to everyone else also in the grid, not just the latest person to come to the party. And if the latest person to come to party is data centers, that they're increasing the marginal costs for themselves, but also for everyone. So on top of the infrastructure buildout we've been discussing, there's this marginal cost phenomenon. Am I getting that right? And how much of a problem is that?

Speaker B: That's a great explanation. It's exactly right. Um, there are a number of reasons why and it's just, it's just more expensive to build stuff generally right now. So you have the supply chain crunch on certain power sector components that are driving out costs. General costs of building and borrowing are going up. And so you're absolutely right. The cost of that new energy and new infrastructure, new delivery infrastructure as well is going up. And if you use that traditional approach of just averaging the costs for everybody, socializing those costs, then everybody's costs are going to go up. And that's why I think the solution here is to isolate the data centers, to charge them the costs of the new infrastructure needed for them and the energy that they're buying and to try to separate them as much as you can from everybody else.

Speaker A: How? Um, I'm, um, guessing if we had a representative from Hyperscaler here, they might say, wait a second, this is no different than Exxon building a big office building. And in fact it's similar in many ways. We're adding jobs to community. You know, we're, we're the people need jobs and economy, we're adding jobs, we're adding opportunity. We're going to have a, uh, jolt into the economy. Um, this is bringing a lot of, of um, of action and economic activity into the community. So why shouldn't we act as we usually do when companies come to it? Why should AI be demonized and singled out? What's your, what's your response to that kind of um, uh, stance of the hyperscalers?

Speaker B: Yeah, I think a couple of things. One is the speed and the scale issue. Uh, this is happening. So it's a massive scale and it's happening. They want, you know, they want this to happen today. Um, and there's a huge cost to that. And you just explained it with the marginal cost that we were just uh, talking about. So I don't think there's really a good historical parallel. In fact, if you go back say to the 1950s, 60s when there was this baby boom, massive industrialization, huge growth in this country, the economics of the power industry were totally different where the marginal cost is actually lower than the average cost. So when a new factory came in and the utility had to expand to meet that demand, it was actually lowering everyone's costs. So it was a win, win. That's just not the case, uh, right now for reasons we just discussed. Um, and the second issue is that because of the speed and the scale, there's this huge uncertainty and therefore risk for everyone. So the risk is what happens if a data center goes to the utility and says we're coming, you know, build us to get, you know, enough. Uh, we want to have a gigawatt facility here, we're going to open it as soon as you can and something happens in the world and that data center never becomes operational. But the utility is already spent $2 billion to build all of this infrastructure. Well, the utility is going to want somebody to pay for those costs and it's going to go to ratepayers and ask for money. So one, one major improvement we have seen, uh, not quite across the board, but in many states, uh, is this requirement that if utility is going to build something for a data center, the data center first has to sign a long term contract right away that's going to guarantee some stream of payments for a 10 or so year period. Often I would say it's not enough, uh, but it's certainly better than zero.

Speaker A: Hm. Um, and to clarify my own stance, uh, as an aside, I do think there's real economic and positive benefits for these things potentially. And I think I just want to see that all costs are fairly captured and borne by them. Right. And so I'm not anti dentist center at all. Assuming that they are properly paying their fair share. Um, and I think so. I really appreciate how you've, um, scoped that out. I'm wondering. I think also part of fair share of costs has to be the environmental costs as well. And to ensure that any other external costs, um, are appropriately, uh, factored in. Um, do you have. This might be a little outside the scope of your official, um, studies and work. But how do you think about that? About like. Because everything we talked about so far has been the literal, like energy, the bill from the utility. But there's other social and environmental costs as well. How are you thinking about to what extent these hyperscalers are or are not, um, paying the full tab when you factor in all costs?

Speaker B: Yeah. And here, uh, um, some of these issues are hyperlocal. Right. So data centers can be noisy in part because they have so many backup generators. Those backup generators are often very dirty. They're powered by diesel fuel that spew off all sorts of noxious and uh, dangerous fumes that are bad for people, uh, who live very close by. Um, there can be water issues in some cases that are not going to, that are, that are very local.

Speaker A: Right.

Speaker B: I want to be clear. These are local issues. You're not, you know, they're not going to. One data center is not going to have a national effect on water consumption. But locally, you know, and certainly the construction process could as well. There are dust issues associated with construction, traffic issues, et cetera, et cetera, et cetera. And so, you know, there's this question is, do we want a data center in our neighborhood? What benefit do we as neighbors of this data center get? Or is it just all cost? Uh, and you can, then you can zoom out farther from there and talk about the greenhouse gas emissions from, uh, this, this build out. A lot of these companies had made clean energy pledges, uh, to power their facilities with wind, solar, nuclear, etc. Facility in a generation that wouldn't have greenhouse gas emissions. And as they suddenly see their energy needs accelerate, a lot of them are backtracking on that. And the power industry more generally is also going in the wrong direction largely because of this perceived need for urgent, uh, growth, uh, for these facilities. So these environmental costs are really, I think, are in the mix as to why there is this broader backlash. Certainly energy costs are an issue, but not by any means the only issue driving some of the opposition.

Speaker A: You mentioned water. And I think it's become a real lightning rod for a lot of folks, um, kind of right or wrong. Can you, how would you put that in perspective, um, the water I've a lot of the folks on kind of the um, more pro build out side would say hey, ah yes, they consume water. But A, a lot of the new ones are being built in a way that's more water efficient and B, for perspective, the total water used by data centers is less than the water used by golf courses and less than the water used to produce almonds in the world. Right. But then also locally it can be very different and it could have impacts on communities. How would you kind of.

Speaker B: Yeah.

Speaker A: What's the fair way to characterize the water issue?

Speaker B: Yeah, yeah, I think those national comparisons are pretty meaningless uh, because water is such a local issue and there certainly are parts of the country that are water stressed. Uh, and look, I don't know how to, you know, I don't know how to tell you, you know there are parts of Arizona or California, let's say where yeah, they're growing. You know, they're using way more water to grow alfalfa that's being used to feed horses or something like that. And so I'm not going to tell them how to use their water properly but any increase in water use can have uh, an effect but I'm not going to sort of make those kind of value, uh, judgments. And there's also, you know, there could be a lot more transparency from data centers on the energy and water use of their facilities. I think right now there's, there are a lot of these issues are shrouded in secrecy. In fact, a lot of data center development issues more generally are shrouded through NDAs and confidential agreements etc. And I think that just makes people suspicious.

Speaker A: Mhm. Let's say you mentioned um, earlier that you'd want to see data centers kind of pay more uh, for what they're doing here and the infrastructure build out. Let's say you are given a magic wand to create the policy that's the perfect policy here. What are the main things you'd want to see if you can write at a state or federal or hell global level. What are the key components of the ARI Energy platform?

Speaker B: Yeah, um, well I want to just highlight that this runs through utilities, not data centers. Right. We've been talking about data centers, but it's the utilities are the entities that are monopolies that are regulated by the state and they're the ones that have the ability to share these costs with everyone so that the regulatory focus is on them. Uh, and they need to just do a better. Some of them are Doing a good job already. But we just need to make sure that they are isolating the costs of the data centers and making sure the data centers pay for them. And that has to happen at every level. That's the local system, that's the higher uh, voltage transmission system. The costs are regulated there by the federal government. That's the need for new power plants, um, that we uh, want to make sure that the data centers and not everyone else are paying for that. And one way we do that is to have more competition and less monopolies in the power industry. Because you can have, if a data center needs a new power plant and it makes a deal with a company that builds power plants and there are companies that do this and it's just a private deal between the data center and the power plant developer, there's really no way to spread those costs to everyone else because it's a contract between two parties. It's only once the utility gets involved, which has this sort of captive rate base across the region, then it has this unique ability to throw some costs into everybody's bill. So in a lot of states, the utility has a sort of complete monopoly that it controls not just the local delivery, but also the higher voltage and the generation uh, of power, the sale of power. And so states can loosen these monopolies and even require the data center say look, if you need a new power plant, don't get it through the utility, get it through some other company. And that's going to have a huge impact for, in terms of protecting consumers. So I make that a centerpiece of my, uh, my policy here.

Speaker A: And it is fair to say that there's elements of kind of good old fashioned free markets principles that breed competition, which breeds lower prices. Isn't that part of this as well?

Speaker B: It feels like, I mean certainly, you know, here, I don't know that we're going to get, we'll certainly get lower prices for everyone. But it's because we're not going to socialize these, these data center uh, costs. So as much of the infrastructure that can be built through private arrangements with the data center. So they can still, they, they still will want to connect to the larger grid because there's reliability benefits for them to do that and that's fine. And so there's still, so you know, regulators still have to be involved in sort of what's the arrangement between the utility and the data center to make sure they're connecting in a fair way and paying the appropriate amount for those services. And the federal regulator right now is spending a lot of time on that issue. Um, so it's, it's, it's sort of, it gets really into the technical details. We won't go down, but that's an important issue. But really let's put a lot of the infrastructure costs as much as possible on the data center.

Speaker A: Uh, Ari, as we wind down, give um, us a kind of final question for you. Give us a prediction for or range of predictions of how you think this will play out or the next one year, five year time frame of your choice.

Speaker B: Well, I mean I said there's just a lot of uncertainty as to how much growth, how much demand there's going to be for from the data centers. And the biggest uncertainty is are we in the second inning here of data center growth? Are we in the eighth inning of data center growth? And that's really uh, the biggest uh, unknown. So um, my prediction is that we're going to keep kind of just chugging along. Um, we're not going to have any sort of sweeping reform from Congress here. A lot of this is going to play out state by state. We'll get different results in different states. And uh, at the federal level again, we're just going to sort of keep plugging away at this, chipping away at this problem slowly. And what we're going to see over time is that I think we'll get more of these costs on the data centers. But what I don't know is how much is already baked into the system already. In other words, how many projects already have shovels in the ground and the deals are in place such that data centers are not going to be paying for all of this infrastructure. That's pretty weak prediction. Uh, but that's what I got for you.

Speaker A: I think it gives a good sense of the kind of how wide the error bars are, if you will, of how um, up in play this is. But pretty soon data centers will all be in space, right? We'll have the starlink, we'll be sending it all to all the in space. We have nothing to worry about.

Speaker B: I mean, you know, the idea that people are even talking about that I think just illustrates the failures uh, here because this shouldn't be so hard. Um, and then it's unfortunate. And it's not just the power sector that's slowing things down. There's certainly a lot of other issues as well around data center development. Uh, but we shouldn't need to launch these things into space. We ought to be able to build them here. We have plenty of space for it, um, we ought to be able to build the infrastructure, uh, we need for it.

Speaker A: Well, Ari, thank you so much for coming on and making a very complex conversation more clear. Appreciate you and your work. Thank you again.

Speaker B: All right, thanks for having me.

Speaker A: Well, there you have it. Thanks again to my guest, Ari Pesko, director of the Electricity Law Initiative at Harvard Law School. Thank you to our show's producer and video editor, Jasper Chua, and thank you to Garrett Lang for theme music. And thank you to you, dear listener. If this is your first time at AI Curious, please subscribe. Rate it 5 stars. All that good stuff. Thanks again and see you next time.