AI wants your electricity

Episode 1 of Power Play, Quartz's six-part investigation into AI and the US power grid.

In Ohio, one utility held 30 gigawatts of data center requests. When it asked developers for money upfront, requests dropped to 13 gigawatts. Jackie Snow follows the phantom demand clogging utility queues, the decisions already being made on numbers nobody trusts, and who pays if the guesses run high.

Reported by Jackie Snow, Catherine Baab, and Joseph Zeballos-Roig.

Hosted by Marco Pimentel.

Produced by Simon Litt.

Edited by Georgina Kemp Walker.

Andy Cvengros: We're getting called by an old lady who owns a strip mall in western Illinois that wants to know if it's a data center property because there's power lines behind her.

Marco Pimentel: She's not the only one. All over the country, people who own a patch of land near a power line are calling their local utility companies with the same question. Am I sitting on a gold mine? The biggest companies on Earth are racing to build artificial intelligence, and AI runs on staggering amounts of electricity. Whoever powers it first, they believe, wins the next era of technology. So America is about to rebuild its power grid faster than it has in 50 years. Hundreds of billions of dollars. The thing is, nobody actually knows how much electricity AI will need. The most credible forecasts wildly disagree, but the utilities have to start placing those bets now. I'm Marco Pimentel, and this is Power Play. Over six episodes, we will follow the race to rebuild the American grid for AI, the fight over who ends up paying for it, and who exactly stands to profit from the change. This is Episode 1. For the last few years, the world has been told a specific story about artificial intelligence. We've been told it's software. Code in the cloud, a prompt box on a screen, a quiet, weightless revolution. But behind closed doors, the people building this technology have known the truth. AI isn't just software. It is the most energy-intensive, resource-hungry machine human beings have ever attempted to build. Recently, NVIDIA CEO Jensen Huang, the man supplying the silicon behind the entire AI boom, stated the raw reality out loud.

Jensen Huang: I've been over the last half decade helping people understand the amount of compute that's likely to be coming. The amount of energy that we need for computing is likely, probably a thousand times more than we currently have. And that's an enormous amount of energy.

Jackie Snow: That single question marks the end of the software era and the beginning of something much bigger and much more volatile. Overnight, the race for AGI has collided headlong with the physical limits of our planet. Trillion-dollar tech giants are crashing into 100-year-old utility monopolies, rural communities, and an aging power grid that was never engineered for this scale.

Elon Musk: So over a year ago, the shortage was chips, neural net chips. Then the next shortage will be electricity. So I think next year you'll see the electricity that just can't find enough electricity to run all the chips.

Jackie Snow: We are currently witnessing the largest, fastest infrastructure shift in a generation. Multi-billion dollar bets are being placed right now behind closed doors on pure guesswork. And if those bets go wrong, it won't be Silicon Valley investors who pay for the mistakes. It'll be regular ratepayers on their monthly utility bills.

Marco Pimentel: But this isn't just a story about gridlock and inevitable crisis. We also look at the solutions, from smart grid software and digital twins to virtual power plants that could unlock the massive pool of stranded energy we already have without thrashing the planet or inflating your electricity bill. This series is an investigation into that collision. Over six episodes, we're pulling back the curtain on the energy side of the AI boom to understand where those gigawatts are actually coming from and who gets left holding the bill. I teamed up with Quartz reporter Jackie Snow to break down the numbers. Maybe we can just start with demand. My understanding with demand and from research and interviews is that there was an expectation that energy demand prior to what we would call the AI wave was that demand for energy was increasing, electric cars from batteries at your home to just the electrification of different things. What was the reporting kind of looking at this? Because it seemed like it was very difficult to get a forecast.

Jackie Snow: Yeah, we had pre-electrification, pre-data centers from the 1990s up through 2010, 2015 even. We had this era of flat growth. The population grew, our economy grew, but there's all these ways that we are getting more out of what we had. So even LED light bulbs apparently were a big thing that helped reduce the demand on the grid. And I think the electrification question, people have seen this coming, EVs electrifying different industries. We saw it coming, but it wasn't particularly a sexy thing. People aren't like, ooh, EVs electrification, oh, my heavy industries will now be electrified. So I've been listening to energy podcasts for years and they've been talking about this before the generative AI boom that we're in. They were like, we're going to have to upgrade the grid and consider all these things. You could go read industry presses and they were talking about that, but it didn't break through because this was something that was going to happen behind closed doors, slowly in ways that were invisible to us. Maybe we'd see our electric bill go up or down, depending on how they did it.

Gernot Wagner: There's been this long-term, decades-long trend in the U.S. and elsewhere, Europe, where we've seen a downward trend in electricity demand growth.

Jackie Snow: I also took this to Gernot Wagner, an energy economist at Columbia Business School, who made the same point from the economic side.

Wagner: For the most part, it was efficiency gains. It was the good kind of using less electricity because we found ways to do more with less. Productivity is good. Electricity, power, energy is an input. If you can do more with less, all the power to you.

Marco Pimentel: At the same time, we hear that we want more electric cars and we want more batteries and we want more things that are electrified because it's just the better kind of option. So that's the big kind of scary question that I'm really grappling with is how do we do that? There seems like a major, major disconnect between all those numbers.

Jackie Snow: Yeah, it's tough, but it does seem like there's a sense of playing musical chairs and just having to be up and going around in the circle and getting a seat and having the energy at your disposal, whether or not you're for sure that the need will be there, because even before we get to the grid part of the equation, these AI companies, they don't want to pay a bunch of money to build these data centers. Utilities have guessed wrong on demand before. I talked to Brendan Pierpont at Energy Innovation, where he directs electricity modeling.

Brendan Pierpont: In the mid-90s or so, utilities were still projecting this rapid demand growth that they'd experienced for the last few years, few decades even at that point. In some parts of the country, they got way out over their skis. They were building expensive coal and some nuclear projects that were really not supported by the fundamentals. Utilities sort of missed the mark on load forecasting and overbuilt and had a bunch of stranded costs. And that was a driver of some of that decision making.

Marco Pimentel: Stranded costs, meaning even if demand never shows up, someone still has to pay for it to get built.

Jackie Snow: What matters most for us right now is how those flat decades changed the utilities. Growing the system stopped being how they made their money.

Pierpont: Where utilities really sought their growth is in replacing aging infrastructure, in smart grid, grid modernization investments. And so now we're faced with actually having to grow the system and grow capacity to meet this load growth. And it's a muscle they haven't used in decades.

Marco Pimentel: So we're handing a once in a generation build out to an industry that spent 20 years learning how not to build.

Jackie Snow: Some of the uncertainty about how big it is doesn't help. If we knew, okay, it's going to be this percent of the grid, it looks pretty certain, this is the growth. I think that's something that people could understand. And having some uncertainty there about the size too isn't helping some of the fears that people have around this technology.

Marco Pimentel: What was kind of your just gut reaction to that question after all this research?

Jackie Snow: The projections are all over the place. And you have to take anything that the AI companies are saying with a giant grain of salt, because right now they have to project that we're growing, we are important, we see such use and utility of our product that we're going to need as much energy as we're using today just for us. I put the same question back to Wagner.

Wagner: Estimates are all over the place, which is, I would say, classic or typical of where we are, right? So everyone is going to talk of their own book. So if you ask the Sam Altmans of the world, you get a very booster-ish answer for obvious reasons. Well, let's take that with a grain of salt.

Marco Pimentel: So the biggest numbers tend to come from the people who profit most from them being big.

Jackie Snow: The 2030 projection is like a 5x spread, right? I've seen numbers that start at 200 terawatt hours up to 1,000. And that is a crazy spread. And the fact that we don't know more about that really does just showcase how much we're dealing with people saying what they need to, to sound like the AI future is big and bright and also just the unknowns about what this future could look like with the technology and user base that really maybe will turn up in the future.

Marco Pimentel: To confirm that spread, you mean is what's potentially projected between how much energy is needed by 2030 is like a 5x difference between different reports of what AI companies are potentially needing?

Jackie Snow: That's right. Yeah.

Marco Pimentel: And part of why the low end is even plausible is that the technology keeps changing under everyone's feet.

Jackie Snow: And right now what we're doing with these models is you're kind of building them to be as smart as possible because everyone's in a race for artificial general intelligence. But there is a future where maybe you want the smartest, but maybe you will to some level be like, oh, but we also need these to be fast or more energy efficient. So we're going to start to see models that are more energy efficient and maybe do not need the data center load that if we are to continue to build it exactly like we are today that we'd need in the future. Also, there is this idea of doing this, your AI needs on your device. Apple is really big on that. Maybe you don't need to actually ping a data center. So there's this interesting tension right now between what they're saying, what the technology could potentially do in the future, and then what the grid is capable of doing right now.

Wagner: Data centers need power. There are projections all over the place, but the one fact they share is that we would expect to have significantly more power by 2030, by 2035, by 2050 than we currently have in this country, in Europe, in China, everywhere because of data center. Now we can ask the question whether some of the AI boosterism is justified based on the productivity gains that would come out on the other end. Just getting addicted to your new AI girlfriend or something, if that's the thing we are doing suddenly, that's maybe not a productive use of the technology. If there are productivity gains to be had, fantastic. Yeah, let's figure out how to get the energy.

Marco Pimentel: So while the direction is obviously real, the size of the ask is anyone's guess. And why?

Pierpont: There are two elements of the uncertainty story. One is how much do we need. The second is on what time scale. My sense from sort of looking at this from the electricity sector perspective is that there is a wave of speculative development and it is driven by the fact that there's this new technology unlocking all kinds of things, but there's just enormous uncertainty about what the future business models of AI actually look like. There's just this broad enthusiasm that if we build as much compute as we possibly can that there will be uses that monetize that. I don't really know where we land on that. I do tend to think that things like power, availability of chips, availability of capital. I think some of these tech companies are probably starting to run into challenges of financing this stuff on their own balance sheets and looking to other ways to finance this build out. Those are going to be some of the limiting factors, but really hard to tell what ultimately drives the amount that we build out.

Wagner: The demand projections may actually not turn out to be fully true. The bubble might burst by the time this podcast comes out and suddenly we're looking at demand projections that look very different.

Jackie Snow: To actually get power, a developer has to file a formal request with the utility, asking to reserve a piece of the grid. With demand this uncertain, you might expect those requests to be slowing down. They've done the opposite. Andy Cvengros co-leads U.S. data centers at one of the biggest firms in commercial real estate, JLL. They track the actual deals, and he'll tell you most of the requests flooding the utilities aren't even real.

Cvengros: There's real projects there. They're being developed by the hyperscalers, by very qualified data center operators. I think what you'll see is you have the hyperscalers, the data operators, and maybe a few others that are really the most qualified groups to do this. And they're not really going outside of that core group of people. Outside of that, you see thousands of people trying to get in the space. Anybody who owns land that's next to a power line is now saying, do I have power capacity available? Can I go get a power study? Can I sell it for seven times what I got it for? All those people are going into their utility provider and everybody's bombarding these utilities. And so I'm in ComEd territory, I think there's 35 or 40 gigawatts of power requests in the system. And that ranges from the institutional operators, hyperscalers, all the way to the little lady down the street who's doing that. So they have to hear everybody. They have to entertain everybody's request as the same, and they are trying to figure out how to curtail that.

Jackie Snow: When you want to thin the crowd, the first lever is to up the ante.

Cvengros: It used to be that you could get a power study for like $25,000, $30,000, it'd be back in 90 days, and you're off to the races. Now there's so many people requesting it, they're requesting a lot of times a million dollars plus to hold that power off the market. And so what they're trying to do is push people who don't have money and aren't real and just have land under contract and are trying to flip it out of the system. A lot of people have a lot of money so they were putting those deposits down and they're going to the next step, which was committing to the power. And what the utilities are now doing is putting down, requesting a letter of credit against the power that you're drawing. So a lot of, we call take or pay or must take, type of models where you have, let's say you request 500 megawatts of capacity, you have to pay for 85 percent of that power as though you're using it day one up front through a letter of credit, and you may not get that power for five years. Even the best operators, the most well-funded operators, the ones who may or may not have a tenant in tow, they're looking at the same. I can't agree to that. That just financially, I can't hold capital off the market for five years. I may not ever see a return. There's a huge risk on our part. Even hyperscalers are saying, that doesn't really make sense to us. So that is happening all over the place.

Marco Pimentel: The deposit doesn't really seem to be about the money. It looks more like an attempt to filter out the prospectors.

Cvengros: Columbus, Ohio, AEP, the power provider, they're publicly announced they had like 30 gigawatts of power requests. As soon as they enacted the 85 percent, it dropped to 13 gigawatts. And then even that, I think, has dropped to even below that.

Marco Pimentel: 30 gigawatts down to 13 just by asking people to put real money behind it.

Jackie Snow: And Cvengros thinks that projections will keep falling as more places do that.

Cvengros: In any given market, there may be one or two gigawatt projects that land, but there's a ton of requests out there. So I see it falling out considerably as that happens more. I just tend to think that there's a better way to do it than requiring big letter credits and big deposits. A question should be, have you ever built a data center? Or do you ever intend to build a data center? Where's your tenant? The qualifying questions that don't require big capital. I feel like you should be qualified to even request a gigawatt of power.

Jackie Snow: The biggest version of this is happening in Texas.

Cvengros: Dallas, there's 250 gigawatts of power request. 90% of them are not real. They're purely people trying to put projects together to sell to the hyperscalers. Most of it is people speculatively trying to put projects together that are likely never going to go anywhere because they're just not going to be able to get power, that can go get zoning, they don't have the capital, they're not able to secure a tenant.

Jackie Snow: Cvengros walks me through what one of these dead-end projects looks like up close.

Cvengros: People go through this power study process. So we call it a detailed load study. You pay the million bucks, you're really getting real in your request, and they're going to come back and say, hey, it's going to cost you 60 million bucks to bring the power to your site. And you have 90 days plus or minus or whatever it ends up being to decide if you want to take this power on or not. If you don't take it, then it's at risk of being stolen from you. So it's kind of an exclusivity period. And so what you see sometimes is people who are unqualified developers running around with a load study saying, who wants to buy my site? I have 90 days to close. Sometimes that works, but generally it doesn't. So they're going to lose their power. They've spent a bunch of money for no reason. The public perception of data center growth is way overstated because of what they hear in the local market and what they see of people trying to do projects that aren't actually real will never get leased and it's just people trying to make money.

Marco Pimentel: At the end of the day, does the AI companies become, it's obviously a tooling layer, but a big part of this market is owning the energy layer and who owns that. We are building and AI is building at a wartime pace, but the grid is at an Eisenhower type of pace and so there's a massive difference between the two.

Jackie Snow: There are places Elon Musk is getting a little trouble right now. He has a data center that is powered by 100 gas turbines that he just sort of set up in the community is not happy about it. He has not made a climate pledge as far as I know, not like Google and everything, but getting a hold of some of these fossil fuels is hard right now. These AI companies don't want to be energy companies and they don't want to have to be a gas line engineer either. That takes a lot of engineering, a lot of safety stuff, a lot of labor that they're not interested in. Whereas standing up a green field of solar panels is currently faster. So it's faster, cheaper and keeps them out of a little hot water that people will feel if all of a sudden we're going to repower up a coal plant that was shut down.

Marco Pimentel: But a solar field only makes power when the sun's out and a data center runs flat out around the clock, so even the fast clean option doesn't get you firm power on day one. What are they actually doing to bridge that? Andy Cvengros has thoughts on this.

Cvengros: The grid's not working because the transformers and the circuit breakers that you build, they use to make a substation, right, anywhere from three to six year lead time for a lot of those items. So you just can't build grid level power. And so these big campuses, they're coming to the table with their own generation. They're utilizing natural gas. They're basically running the turbines, generating electricity, and bringing the grid in later. So the grid comes in in five years, they'll take that generation, put it onto the grid.

Jackie Snow: And the thing about the turn to gas is that people who watch the grid see it as a step backward. Here's Brendan Pierpont.

Pierpont: One of the responses of the data center industry to the slow pace of being able to connect to the grid and get power has been this huge turn toward on-site natural gas. That's both a signal that the utility industry is not meeting the moment, and it's a huge missed opportunity because a grid that has a diverse set of resources that balances across a large region is going to be cheaper and more cost-effective. So this is the kind of on-site natural gas boom is likely just a whole bunch of capital investment that doesn't need to happen.

Jackie Snow: The buyers themselves used to refuse gas because it isn't as reliable as the grid. That's changing too. Here's Andy Cvengros.

Cvengros: The turbine model is something that's really attractive for some. It doesn't have the same reliability profile as grid-level power does. And so a lot of the hyperscalers will look at that like, I cannot go natural gas for these types of applications. Anything that uses a public cloud, performance is everything. You have to be able to access your files, whereas AI is more a processing engine that potentially can go down. It has less levels of redundancy, but they still want it, what we call five nines uptime. They're starting to come off that a bit, and now adjusting their designs and their leniency towards that to be able to use it more.

Marco Pimentel: Which brings us back to the companies that have stopped waiting on the grid entirely. They're building their own power on their own land.

Jackie Snow: Further down the desperation ladder, the fixes get more far-fetched.

Marco Pimentel: Maybe a naive question, but could any of this AI growth be stalled due to just not having enough energy resources over the next kind of four years? I think there was research that you did where they're using jet engine turbines to power, with unsubstantiated type of technology to do it. That seems to be fairly desperate.

Jackie Snow: I love that jet turbine story because it's that guy, he just went to Y Combinator when Sam Altman was in charge and just DM Sam and was like, hey, I have a solution. In reality, the problem is the production of these things are behind. It's like, I guess if you had a warehouse full of jet engines, maybe that'd be helpful, but also jet engines are meant to be run really hot for like eight hours at a time, not 24 hours a day at data centers like they need. So it's not quite a solution, but I think there is such this craze around the AI energy crunch that you put AI in your name and all of a sudden you have more funding. Investors aren't dumb, though. People are reaching for far-fetched solutions like jet engines because the more standard means, like turbines, are spoken for.

Pierpont: The turbine supply situation is, it's a really tight market right now, particularly for sort of the most efficient gas turbines. So the big gas turbines that you put in the combined cycle power plant that can run with greater efficiency, that have adequate pollution controls on them. Those types of turbines have the longest lead time now, probably five, six plus years to get a turbine delivered if you order it today. When you have GE and Siemens and Mitsubishi have been honing and optimizing gas turbine design for power sector for decades to think that someone can just sort of patch together some off the shelf solution with something that was never built to generate power. Yeah, definitely feels like a big risk.

Jackie Snow: And even the real version of this, the serious gas plan, is bigger than the companies reaching for it realize.

Pierpont: I don't think any of these companies are really thinking through the fuel supply side of this. A single large data center connecting to the gas system to fuel a set of gas turbines to provide, say, a gigawatt of data center load in a particular site. That's like a small city's use of gas. And so they're now in the business of trading gas as a commodity on a scale that most local gas companies don't even have to trade gas as a commodity. There's this complexity that I don't think many of these companies have thought through.

Marco Pimentel: Someone has to pay for all that building. And it ties back to where we started. A utility makes its money on what it builds. In the flat years, that meant upgrades. Now it means a build-out whether the demand shows up or not. Guess high, guess wrong, and it lands on your bill for years. Here's energy economist Gernot Wagner again.

Wagner: Just because there's demand doesn't mean we need to supply the power. The question is, how much are we willing to pay for it? Those hyperscalers, the Googles, Microsofts, Amazons of the world are so rich that they're willing to pay a lot of money for that extra power. And that does interesting things to markets, right? So electricity prices can be expected to go up for your wholesale, for other industrial users, households as well, right? Making it an election issue, making affordability an issue that's front and center in people's minds. I teach climate, 500 MBAs a year or so. The most fun question I like to ask is, can you tell me how much you pay for your electricity? And by and large, nobody knows how much to pay for the electricity, which by the way is sort of a luxurious position to be in because it basically means, oh, you can afford it. It comes out of your checking account. You don't really negotiate with your utility. You just don't think about it. Now, lots of people do have to think about it because it's a big portion of their paycheck. And of course, if it increases 10, 50%, you notice.

Marco Pimentel: Think about the price of energy. What are we forecasted for that? You're talking about turbines that are not available for five or six years. My assumption is whenever there's more demand than there's supply, that means that the price of energy typically increases. What does that look like right now for the future of prices for energy?

Jackie Snow: I think the best case scenario, the experts I've talked to see a future that if we do it correctly, rates go down. So the way the energy market works in the US is these are all regulated markets. They all work a little different, but the way the business model calculus is utilities earn a regulated return on capital investment. That's their rate base. They get more infrastructure, more earnings, so this has been a baked incentive to build, not optimize or innovate. And more utilities are starting to go to the model where they're not getting this incentive to build. They're going to get incentives to drive costs down, and that will be more aligned to this future of growth. Before, when we didn't have a lot of growth, maybe that was fine, but we cannot keep building increasing amounts and have the rate go up for everyone. We're just seeing a build out where everyone's bill is getting higher. But there are ways of changing the market incentives and then also getting a little bit more out of the grid we already have and including more payers on it could lower the rates for everyone.

Pierpont: I think so much of how the data center demand growth impacts affordability in particular has to do with not, there's nothing necessarily fundamental about that. A lot of it has to do with how that demand is treated in regulatory processes or in market structures and market designs. I think the devil is very much in the details and kind of how data center demand actually ends up impacting costs for other customers.

Jackie Snow: That incentive runs straight into a market that locks prices in years ahead. Pierpont watches this in the Mid-Atlantic.

Pierpont: In the PJM region, that Mid-Atlantic grid operator, that capacity market is a three-year forward procurement. And so for the next three years, we've locked in some assumptions about how much data center capacity is going to come online. Already, the market operator is starting to revise downward the forecasts for the near term which essentially just means that we've locked in high prices that are based on an already outdated load forecast and customers are going to be paying those high prices for the next at least the next three years. And there are some vertically integrated utilities where they've made sort of emergency requests to the regulators to say, this load is coming, I need to build a bunch of new capacity to meet this load. And so if they do that, and then the load does not show up, those are costs that the utility is probably going to come back to the regulator and say, okay, we need to recover these from our other customers because we built this.

Marco Pimentel: Well, what happens to a normal person's power bill if the capacity is built for a data center, and then the data center it was built for leaves?

Jackie Snow: If they had to build out new wires, a new substation, a new peaker plant for that data center, and they didn't really see an increase in their bill because the data center offered to pay so much, and then it goes out of business or goes to space. Yeah, then they could see their bills go up considerably and stay there because a lot of these rate hikes, building a plant, you're going to be paying for that for like 10, 15, 20 years.

Wagner: There's a lot of uncertainty here, which also means there may in fact be some undue lock-in. That would be, of course, a problem for all sorts of reasons, economic on the one hand and climate, of course, as well.

Marco Pimentel: As with many negotiations, everyone at the table wants something different.

Wagner: Data centers want, right, flexible contracts and as cheap as possible. The power providers, the utilities, right, want guaranteed off-take agreements, long-term contracts, and of course want to make as much money as they can for building the infrastructure, right? Okay, so that's the basic tension here. And, well, what do households want? Well, households probably want data centers to be paying for the full investment costs, the full cost of the added power demand, and not be picking up the slack for the Googles, Amazons, Microsofts of the world.

Marco Pimentel: If a wrong guess sticks us with the bill, the real question is whose bill?

Wagner: I like to say there are three sources of money in the world. Households, ratepayers on the one hand, shareholders on the other, and taxpayers. That's the age-old political debate, of course, who should ultimately hold the short end of the stake, who should ultimately be responsible for the bill here. What we see so far is that it turns out that the political power as so often rests with those who have a lot of money.

Marco Pimentel: There's also one more real wrinkle. Whatever the real number turns out to be, the grid work still has to happen.

Jackie Snow: The devil is going to be in the details for this. So does the grid need to be modernized? Yes. Did everyone I speak to say even with all these technologies to get more out of the grid we have, do we still need to build? Yes. Our grid is old and needs to be updated for the coming electrification, regardless or not if data centers and AI happen in a big way. So it has to be done in the right way, because if this does turn out to either be a bubble or just let's say we find a lot more efficiencies in how we build these and train these AIs, we're like, oh, actually, we don't need this many data centers, then there could be a lot of people holding the bag.

Marco Pimentel: So who's actually exposed if the bet goes bad? Not the tech giants, really. They can cover whatever they signed for. It's the power that got built on spec without a signed customer behind it.

Jackie Snow: A lot of the data centers are signing more long-term agreements, so it'd be hard to get out. A lot of these companies, even let's say Meta decides that the juice is not worth the squeeze when it comes to some of these AI projects. They do have the money to pay these bills. It's not going to be like a lot of failed companies that there's no one to sue for or to get the bills to be paid for. So there's definitely some risk, but I think some of these utilities have looked into the future for these kind of things and are making sure that they've signed long-term agreements with companies that are going to be there in the future. The grid is something that's mostly invisible to us day to day, but we recognize how important it is. And I think when we are seeing cost of everything go up, including our electric bill, which feels like something that we can't really negotiate or bring down without introducing discomfort to our lives. And it is because of a technology that people are scared of, either because of the risk to jobs or to even our future livelihoods on this planet that we all need to inhabit together. I think it's kind of this perfect storm of something we are recognizing is really important, being met with something that we're not sure how important or how powerful it is.

Marco Pimentel: So that's where episode one leaves us. We're rebuilding the grid for a number nobody can pin down. And if we guess wrong, your rates help cover it. Next time, the size of the demand is only half the problem. Even the power we're sure we need can't get where it has to go fast enough. And the reason goes straight back to those flat, quiet 20 years.