Transcript
INTRO
Jamie Nolan: Welcome to Energy Empire. My name is Jamie Nolan, and I'm a clean energy communications executive.
Tim Hade: And I am not Jigar Shah. I am Tim Hade, but I'm happy to be here.
Jamie Nolan: Hello, my friend. My first choice for a Jigar stand-in. He is on vacation, and he's actually attempting to be on vacation instead of just working the entire time.
Tim Hade: Well, one, I don't believe him. And secondly, I know Arnav is on a plane right now, so I'm happy to be your second choice. We all know that was your first choice.
Jamie Nolan: I do love Arnav as well, but you do bring mini Jigar Shah energy, as I explained when you joined us last time.
Tim Hade: Amen. And before we talk about the incredible podcast we're about to do: the Energy Empire merch store is killing it, Jamie. I've been told that Simon is responsible for this, but it's going really well.
Jamie Nolan: Yes, Simon, our executive producer, is responsible for that. I love that we hopped on this recording and I happened to be wearing an Energy Empire shirt, and you happen to have the sweatshirt behind you. This is my first time actually rocking the merch on the pod. So thank you for representing, and we'll forgive you for not wearing the sweatshirt, since you're in California.
Tim Hade: Yeah, I'm a genius. I bought a sweatshirt in the middle of the summer, and it's 85 degrees in my office right now, and Simon made me turn the air conditioning off. So we're not going to wear the sweater, but we will represent.
Jamie Nolan: We'll get you a t-shirt. And we hired a graphic designer, and we're going to get some awesome new merch designs coming out soon, so keep your eyes peeled, folks. I'm really excited about today's guest, who I think is one of the most effective communicators in our space.
Tim Hade: Amen. We're talking about Michael Thomas, who runs Cleanview. One of the things Jamie and I have been talking about for a long time is that it's just so hard to get a clear picture of what's going on in the energy sector right now, specifically around data center development. And the place I go to get the best information, the place I've found that's the most accurate and up to date on the cutting edge, is Cleanview, which is run by Michael Thomas. I'm really excited to talk to him today.
Jamie Nolan: Me too. I'm really curious to know how he gets all of this incredible insider information. He recently broke the story about this massive Amazon data center that's being constructed in the middle of the West Texas desert, and he broke that story nationally. It's been reported across mainstream media, tier one outlets, in the last week, citing his analysis. He really is doing incredibly important work for us to even be aware of what's happening in the rapidly evolving data center sector, because there's really just no single good source to get that information.
Tim Hade: Yeah, it's crazy. For most of us that have been in the power sector for a long time, the one thing you can count on is at least having access to information about what's going on. Forever, if you wanted to build a power plant or a new transmission line or whatever you're trying to build, that's a public record somewhere. It gets consolidated. There's a bunch of different outlets you can go to, and you can figure out what's happening in a particular geography, a particular county, a particular zip code. That just isn't the case today with the data center stuff. Trying to figure out what exactly is happening in this market right now is much harder than I expected it would be. And one of the people making it a lot easier to understand and wrap our heads around is Michael.
Jamie Nolan: Absolutely. And I'm also happy to have you sitting on the other side of the screen from him, because I want to talk about this choice that we're seeing, for these big projects to choose off-grid gas instead of the more flexible options that we know are available with virtual power plants, batteries, et cetera, which is what you're working on at Voltus.
Tim Hade: Yeah, for sure. Ultimately, I think this is a generational opportunity to do a lot of good things. There's a lot of questions and a lot of negativity around the AI boom and data center development right now, and there are bad outcomes. There's a path we can take that leads to really, really bad outcomes. But I think there's also a path we can take that leads to really, really good outcomes. We've talked about this before, but there's a world in which, because we develop a bunch of data centers and connect them to the grid, people get cleaner, more reliable, more resilient energy, and cheaper energy, which is, I think, the most important thing in the world today. How we navigate this point in time is really, really important. And whenever I think about trying to solve a challenging problem, having good information is always the starting point. All of this starts with having a clear view on what's going on in the market today, and there's probably not a better person in the world to talk to than Michael Thomas about that.
Jamie Nolan: Awesome. Well, with that, let's get right into it.
CATCHING AMAZON IN THE DESERT
Jamie Nolan: Michael Thomas, thank you for joining us. I'm so excited. I'm personally very excited to meet you, because I just think that you are one of the most effective communicators on clean energy of our time. I follow you on LinkedIn and your content is incredible, and I was so excited to have you on the pod. So thank you for being here.
Michael Thomas: Oh, thank you so much for saying that. And thanks for having me. I'm excited to be here.
Jamie Nolan: So you have just broken a pretty big story that's getting a lot of pickup in national news. I actually just saw my friend Mark Chediak at Bloomberg covered it today, and he credited you. You're getting all the credit for this one. Two weeks ago, you broke the news that Amazon is behind what could be the largest gas power plant ever built in America, a 7.65 gigawatt project in the West Texas desert. So tell us the story. How did you find it, and what exactly is Amazon building out there?
Michael Thomas: In January of this year, there was a behind-the-meter gas plant that came onto our radar called Gigawatt Ranch, and it came onto our radar because we track all the Texas air permits. At first, I was sort of skeptical of this project. This is a 7.65 gigawatt gas plant in the middle of nowhere. The initial developer had never built a natural gas plant before; their previous experience was limited to utility-scale solar development in Japan. So it had all of the signs of a speculative project that would never get built, and a lot of people dismissed it, even though there was some news and hype when they announced that they had secured the largest air permit ever given in America.
We still continued to track it. And then, a couple of weeks ago, we saw a building permit pop up in Texas, and it was from Amazon, and it was located just down the road from the location of this gas plant. So we looked at satellite images, we looked at the construction permit, and we felt pretty confident that it was the same project, and that the locations were just slightly off on those two permits.
So we reached out to Amazon. Usually companies don't like to talk about their projects. Often you won't get responses, or you'll get a standard statement about not commenting on rumors. But Amazon, I talked to them for about an hour on the phone, and they issued a statement. Then we published the story, and everything that we had found ended up being accurate. So the project now is being developed by Amazon. The gas plant is being developed by an unnamed IPP that will be building it. And it's still that same size, 7.65 gigawatts of nameplate generation capacity, and we can expect dozens of data center buildings on that campus. So it's a very large project.
Jamie Nolan: Wow. I would love to have been a fly on the wall for that phone call you had with Amazon, as someone who has done communications work for a long time and had a lot of very tense conversations with reporters over the years. Were they trying to talk you out of reporting on it, or trying to suss out what you knew? What's your sense of their goal going into that conversation?
Michael Thomas: To be honest, I don't want to comment too much on the conversations, and some of it was on background, so there's only so much I can say. But I would say they were fairly transparent. They gave me all the information. They verified all the information. They gave me additional information, and so I was able to publish that. Of course, they're going to have their kind of standard corporate speak, as any company will in their prepared statement, and we obviously don't publish a lot of that. But they did share how many clean energy projects they have, their plans for water, which we don't cover as much as a company focused on the power sector, but we include them in the story. To their credit, they gave us everything we needed to confirm the story, and it meant that we had a very accurate story coming out of that process.
Jamie Nolan: Well, good for you. I'm very impressed. And honestly, I'm surprised to hear they didn't try and kill it so they could fully control it. So you must have come with unimpeachable information. Great job. And of course, the national news media is picking this up. It's a huge story.
THE $650,000 MEGAWATT-YEAR
Jamie Nolan: And this project, there are several others like it in store. Microsoft just signed a 20-year deal with Chevron for almost three gigawatts of dedicated off-grid gas. It's something like $650,000 per megawatt-year just to cover the capital. So Tim, this is your whole thing. You're an expert in this topic. Can you explain that number in plain terms, the $650,000 per megawatt-year?
Tim Hade: Yeah. To put this in perspective, the PJM capacity cap, which I think is a good proxy for essentially how much capacity used to cost, is $122,000 a megawatt-year today. So $650,000 a megawatt-year kind of backs up what people are seeing in the market, which is that the hyperscalers are willing to pay a premium for capacity that can come online as quickly as humanly possible.
And I guess that's the thing I would really love to unpack with you, Michael. I think what you guys have done an incredible job of at Cleanview is just giving people information about how many of these data centers are actually being planned, constructed, and operational. And my first question on that topic would be: why are those numbers so hard to suss out? You would think that it would be pretty easy to figure out how many data centers are being planned or built in the United States, but the numbers are all over the place, and I think people really come to Cleanview as a source of truth on this issue. Why is it so hard to get that information? What's going on there?
WHY NOBODY KNOWS HOW MANY DATA CENTERS EXIST
Michael Thomas: I think a helpful way of understanding it is actually to compare it to the power sector, where we have a lot more information. Energy nerds listening will be familiar with EIA, the Energy Information Administration, a federal agency that collects all kinds of data on energy infrastructure across the country. They have this form, the 860M form, and I promise we can talk about more than EIA forms on this podcast, but I think it's illustrative. What they do is they go to utilities and IPPs and anyone building a power plant, and there's a federal rule passed in the seventies that said: if you're going to build something, you need to tell us as the federal government, so that all the grid operators can be prepared, because it's going to impact the rest of the grid. And so there's this queue, basically, where you can figure out how many are under construction, how many are operating, how much gas do they burn, how much coal do they burn. It's this incredibly rich repository. It can sometimes be difficult to join some of it, but the data is there.
Compare that to the data center landscape right now. There's no one large load queue for the country. There's not even a large load queue for most states. The way it used to work, as recently as a couple of years ago, is a data center developer would call up the utility. They'd put them down on a list. It might have been in a spreadsheet somewhere, and there's maybe five or ten large loads. Why would you need a queue or a process when the only large load is Meta that's coming to town, and everybody at the utility knows that?
Now we've got a situation where there are hundreds of data centers under development across the country, with a combined hundreds of gigawatts of proposed capacity, spread across 50 different states. There's no one place where everybody is tracking this, and everybody has their different rules. If you just try to understand, for a single project, does it have all of its permits? You're talking about city permits, county permits, maybe a clean air permit from a different agency, and you can go down the list. There might be 30 different permits. So even to ask the question of, is this project permitted, there are kind of 30 questions within that question. It ends up just being really difficult, largely because we don't have that equivalent of the 1970s rule that says, hey, put your name on this list and let us know that you're going to connect to the grid.
Tim Hade: One of the things I've been reading a lot about is the hyperscalers acknowledging that they might not have been as transparent as they wanted to be in the past, and sort of implementing those kinds of changes. So is that information becoming easier to access? Are you having more experiences like you had with Amazon, where you just call them up and they're happy to talk about it? Or is there still this veil of secrecy happening in the industry that makes this stuff really hard to figure out?
Michael Thomas: It's a great question. I think there are two parts to it. One part is this larger story you hear about, with companies that are signing NDAs with a local county or a utility, and so there's just no way you can get that information. And then there's a separate challenge, which is just the dysfunction of how all of these files are managed. If you go into the Texas environmental agency and try to get a list of all of the power plants or data centers that are trying to get an air permit, it turns out it's a really hard database to search. There's no equivalent of a Ctrl+F search or an advanced search, and they're not indexed on Google. If you're googling that Microsoft project you mentioned, Project Kilby, the only way you'd know it's for a data center is by going into an 800-page document, the permit, and on page 83 there's one mention that says a data center customer will be using this power.
So we've built a ton of data pipelines to collect all that information, to go through it and parse it out. But often the bigger issue is just finding a county and trying to figure out: where would they be publishing the site plan for this data center that says how many buildings there will be? And then at the state, where is there going to be a permit that shows how many backup generators there will be, which is part of the method we use to calculate capacity? It's just a really difficult task. It's not in a clean, nice database that you can search on Google, which would definitely be the dream for us. We've spent many hours as a result of that lack of organization.
SPEED TO POWER
Jamie Nolan: Well, thankfully you are, because otherwise we wouldn't have the transparency that we need around things like the most polluting power plant to ever be constructed in this country. I want to go back to this Texas project, though, because they're doing off-grid gas, and there's another option here, which is why our dear friend Tim is here, because this is what he's very, very passionate about. A flexibility capacity contract would cost a third of the cost of powering such a project with off-grid gas, would be deployed in months, and could lower everyone else's power bills. So why do these companies insist upon choosing gas? Are they so selfish that they're really willing to pay $650,000 per megawatt-year for themselves, but not something like $200,000 to lower everyone else's bill? And this question is for both of you, because I think you're coming at this from different perspectives.
Michael Thomas: I think I can sort of explain the logic that I see from these companies. The context in Texas is that they have, at this point, they say more than 400 gigawatts of speculative large load requests. That's like 40 New York Cities' worth of power demand. It's multiple times Texas's entire generation capacity today. Of course, not all of that's going to get built, but that's the kind of waiting line. And so ERCOT, the grid manager, has really had a hard time with the rules and the process to get through this. While we can all say, well, that Amazon project's likely, but maybe Gigawatt Ranch, before Amazon came along, looked speculative, there's a process and laws that they have to follow, and I think that's something all grids across the country are struggling with. That's what's resulted in these long waits you hear about. In Dominion's territory, in PJM, in Virginia, they're saying it can take as long as seven years for a data center of more than 100 megawatts, which at this point is basically all of them.
I think it's in that context that you're seeing a lot of these companies try to find a workaround. xAI was the first to do this with Colossus, where they wheeled in these mobile gas generators, and they were able to get a data center online in four months, when it usually takes 18 to 24 months. That project was seen by many in the tech industry as innovative, in the same way that it made national news and got a huge amount of criticism, rightfully, for the environmental impacts. If you're running a tech company and you're trying to build this capacity because Anthropic is telling you they'll pay you 30 to 50 million dollars per megawatt of capacity, you look at that and you say, hey, why aren't we doing this?
It was after that project that you saw this huge explosion of proposed off-grid or behind-the-meter projects. All of it really was a result of this desire to get the data center online as quick as possible, and that's all downstream from this enormous amount of demand that we see for products like Anthropic's Claude or OpenAI's ChatGPT and all the compute they need. That's the primary driver of this. It's all about what they call speed to power.
THE TEXAS MORATORIUM AND THE ILLUSION OF CONTROL
Tim Hade: I agree with 100% of that. And one of the things we have to look in the mirror on as an industry is there's a moratorium in Texas right now on grid-connected data centers. Part of the challenge the tech companies have, and it's just a real challenge, is if speed to power is your mission, it can be really, really difficult, maybe impossible, to build a 7.65 gigawatt off-grid data center. But at least you can kind of control that to some extent. That's within your control. What you can't control is if you're planning to build a data center that's grid-connected over here, and all of a sudden the governor of the state comes out and says, hey, there's a moratorium on grid-connected data centers. Well, now you're in a pretty tough spot.
Ultimately, what we're watching happen in real time is kind of a mess, and everyone's trying to do what's in their own best interest when incentives aren't perfectly aligned in any way. And look, I know a lot of folks that work at Amazon. I've worked with Amazon. I really like those people. And to a person, I truly believe that all of the folks I've met on the Amazon energy team are very committed to reducing carbon emissions. That's a core principle inside Amazon, and they really mean it. And yet the outcome of that is you're building a 7.65 gigawatt gas plant that, Michael, your work indicates might be the most polluting gas power plant in the country at the time it's commissioned. That juxtaposition is kind of why we're all trying to work so hard to figure this thing out, because it's not easy. We're clearly doing things that aren't consistent with what we said we were going to do as a nation, and there are definitely better ways to do it. But the better ways require collaboration in a way that's not exactly happening right now.
When you look at it from a tech company perspective, it's like: I need this compute to come online. If I go out and I build this myself, maybe it'll work or maybe it won't work, but at least it's within my control. I kind of control my own destiny. And if I try to get this connected to the grid and build this big load flexibility program around it, there are forces outside of my control that can come in and shut it down. That's something that's happening in the market right now, and it's maybe steering folks in a direction that we as a society probably don't want things to go. Coming to terms with that is the next big test our industry faces in trying to figure out how to get these things connected.
Michael Thomas: I think the control point you made is a really important one, and it kind of gets to the psychology of some of the leaders of these companies. I lived in Silicon Valley for a few years, and you just see that it's a very different politics than companies that are headquartered on the East Coast. A lot of them, as we're seeing, are more libertarian and kind of averse to regulation or bureaucracy or process. So given the choice between, do I want to try to control this entirely and build my own power plant, or do I want to go through this utility process or a grid interconnection process through ERCOT, they're going to choose what they see as more control. Now, of course, they don't control the supply chain. They don't control the local laws and ordinances that could come in. But I think there's this perception of control that is really important in driving a big part of this as well.
Jamie Nolan: So you're saying it's not just that they want to be cool like Elon and roll a bunch of semi-trucks of gas generators in and blow up the neighborhood?
Michael Thomas: No, I think that might be one part of it for some, I'm not sure. But I think that there are certainly these other dynamics.
PRESS RELEASES VS. BULLDOZERS
Tim Hade: How do you think about this ratio of proposed data centers to operational data centers? It's one of the things I really struggle with, where, I forget the exact number, but I think peak demand on ERCOT is something like 60 gigawatts, and you just mentioned that there are 400 gigawatts of data centers in the interconnection queue. I think it's pretty well established that there are a bunch of developers who are just throwing interconnection applications into the queue, because why not? And that's something different electric grids are working on trying to fix. But one of the hard parts of your job is trying to figure out how many of these proposals are actually going to get financed and built, and then how many of the ones that start getting built are actually going to get completed. Because I think you've tracked a few data center projects that were under construction, and then the satellite images showed that they're no longer under construction. So how do you think about how this is developing? We have all these proposals, but a small fraction of those are actually turning into operational data centers. How do we plan for that? I think that's something everyone's struggling with right now.
Michael Thomas: I think it's a great question. It is definitely a policy challenge. For utilities, for example, they have to decide how much generation are we going to build, how much transmission are we going to build, and if you don't really know, it becomes a really difficult task. So I think there are a few different ways to look at it. All infrastructure projects, whether it's a data center, a solar project, a storage project, have these generally agreed-upon stages. For data centers, there's the announcement. It doesn't cost much to issue a press release. We could all get together and write up a press release. Jamie, you've got the experience and background.
Jamie Nolan: I have to interject here and say I sure as shit hope they don't start there. They better have that community benefits plan in hand, and they better have laid the groundwork. As a public relations specialist, please don't start with the press release. But I hear you. That's when the public first becomes aware of it.
Michael Thomas: Yeah. And depending on the developer, there are differing amounts of work that go into it. Some developers will announce once they've got everything buttoned up and all the permits, and others, a dynamic that you'll see in the industry is there's a group of developers who have generally not built things, and Gigawatt Ranch's developer, Pacifico Energy, would have been in this category before Amazon acquired the site. If they announce it, especially if it has some kind of controversy to it, or some kind of large number that the media can grab onto, like the largest air permit ever, the most polluting power plant ever, then it gets a huge amount of attention. And that puts it on the radar of a buyer, a hyperscaler or a neocloud or an AI lab, in a way that it might otherwise not be. So that's the first stage, the announcement, and I think most people who are watching this closely tend to be skeptical of it and want to see receipts, if you will.
So the next question is: do you have a tenant? Because in order to get financing for a project, especially projects this large, which can be billions of dollars, maybe tens of billions of dollars, you need to go to a bank, who's going to say, what's your future income going to be that I can lend against? And that was the gap that we saw was missing from Fermi America, this company that had proposed one of these world's-largest power plants. It was going to be more than 10 gigawatts.
Jamie Nolan: They named it after Donald Trump. A fantasy that was furthered by our former secretary of energy, who owned that one.
Michael Thomas: Yeah. So Fermi announced this huge thing. We did a report on Fermi. We looked through all their SEC filings, what they were telling investors, and then we looked at satellite images, and we found that they had paused construction. We weren't seeing any progress between these satellite images. So we looked into it, and the issue was that they didn't have a tenant. They could get everything prepared, because clearing land isn't that expensive. But once you start to build things with steel and put chips inside that cost billions of dollars, you need money. So that project stalled. What's interesting is that project has since gotten new energy. They did sign a tenant within the last few weeks, and that project is very likely to move toward construction.
So then the next stage is construction, and there are a few different definitions of this, but what we look for is actual evidence that a building is going up. Again, it's cheap to just clear some land. You can hire someone with a bulldozer and clear some land, and it kind of looks like it. What we've seen is developers will take these photos where it sort of looks like a big construction site, because of the angle of the camera. It's a low shot against a big bulldozer: "we've started construction." And that's true, technically, but you have a couple of people and a bulldozer. What we look for is the actual shell going up. That's showing real progress. And then the next stage is, of course, it's operating. It's got chips inside, it's drawing a bunch of power from the grid. Those are the major stages. Once a project has that tenant or is under construction, you can generally assume it's pretty likely to proceed.
Jamie Nolan: I'm not convinced about this Fermi project, but we'll see what happens.
Michael Thomas: Skepticism is important when watching all of this. Don't believe it until it's actually operating. And even then, verify it.
HOW MUCH OF THIS ACTUALLY GETS BUILT
Tim Hade: Stepping back, you're kind of at the intersection of this. I think one of the problems the industry is having right now is just trying to figure out how many gigawatts of data centers are actually going to get built, and on what timeline. A year ago, 18 months ago, we maybe overestimated the queue. We looked at it like, oh, they're putting 400 gigawatts into the queue, and that means 40 gigawatts will get built. And that's not happening. But I guess the thing that I find hardest to handicap about all this stuff is the community benefits plan that Jamie spoke about earlier. Community resistance to these data centers is unlike anything I've ever seen in my career. When you guys are trying to think about, all right, there's 400 gigawatts in the ERCOT queue, some portion of that is going to get built and some portion is not, how are you trying to think about handicapping that? Have you seen developers who have done a really, really good job at this, and therefore you're more confident those projects are going to get built?
Michael Thomas: I think the truth is that it's really difficult to know, because you could see a huge amount of resistance to a project and it could still get permitted. Maybe the result of that is that a county commissioner loses their job in six months or a year, but the project still might get built. And at the same time, you can see a project start construction, and you think, okay, that's a done deal, or get a tenant, and you think that's going to get built, and then there's a moratorium or something that stops that construction, or there's a lawsuit that just slows it down. So it's definitely challenging to measure, and we've seen some examples of people who have published numbers on this that ended up being misleading. There was a very viral story about how, I forget the exact number, something like five gigawatts of capacity is actually under construction for 2026, and media outlets all took it and ran with it. And then if you look at what's actually under construction, and look at all the satellite images, the number is way higher. It's multiple dozens of gigawatts of capacity under construction.
Everyone's trying to figure this out. The thing that I watch most closely is that construction progress, because I don't really think anyone's going to be able to forecast very accurately what happens in 2028, 2029, 2030. There are a few methods we can talk about. But in terms of what's going to be built next year, and you can get higher accuracy for maybe the first half of 2028, those have to be under construction. The only way those get built on that timeline is if they've started construction right now. And those numbers are all huge. All of the hyperscalers just have a huge amount of capacity under construction.
So I try to hold both of these ideas in my head. There's a huge amount of speculative demand, and most of those will never get built, ever. Just throwing a press release out there. And at the same time, there's a huge amount of capacity being built, unlike anything we've ever seen, at a scale we've never seen. Both of those can be true. It's a difficult one to forecast, and that's why you see the difference in forecasts of AI demand by 2030 from all these different groups. I've collected a bunch from the different investment banks and consulting firms. Everybody's got a forecast, and they're between 100 gigawatts and 200 gigawatts. That's a huge difference. That's like, well, we could be off by ten New York Cities. I don't know that that's helpful at that point, if it's that big of a range.
Tim Hade: Well, so 2027. Obviously this is an estimate, and no one's going to hold you to this. But if you had to guess how many gigawatts of data center capacity are going to come online in 2027, just based on what you're watching happen via the satellites, what would your number be, nationally?
Michael Thomas: So we're working on a report that's going to be a kind of construction status: how much will get built, how much is under construction. So I'm going to have a much better number for you. It's probably going to be a big range, like 15 to 25 gigawatts. Again, it's like a New York City's worth.
Jamie Nolan: Okay, that's going much lower than the other estimates I've seen.
Michael Thomas: Well, that's next year. Just what is actually built.
Jamie Nolan: Just 2027. What is added. Okay.
Michael Thomas: Yeah, and that's just the net additions. And this year it looks like it's going to be more like 10 gigawatts that get added. So it's a big increase. But let me come back on the show once we've got the new report, and don't hold me to any of those, since that's kind of just in the works.
Jamie Nolan: Totally. We should have you back. Absolutely. It's so hard to account for rising community opposition, and these strange bedfellows that we're seeing between far-left-wing and far-right-wing activists bringing their shared values together to combat these projects.
BATTERIES, VPPS, AND THE TEN-WEEK OFF-GRID EXPERIMENT
Jamie Nolan: I do want to pivot back to power, because, Tim, this is your whole thing. So instead of 35 gas turbines in the desert, what could these companies build with batteries and virtual power plants, and how fast could they move?
Tim Hade: Look, I think generally speaking, the way I look at it is: the tech industry is going to build these data centers. They're going to find a way. And what's important for us as a society is figuring out how we can support them in building these data centers in a way that benefits all of us the most, and I do think there's a lot of overlap there. Of those 10 to 15 gigawatts of data centers that are going to be built next year, if you say 15 is the number, probably 10 are going to be grid-connected and 5 are going to be off-grid, something like that. I think the big challenge is going to be: what does the performance of those off-grid data centers look like? Because we've built enough on-grid data centers to know how to do it. We know how to make it work. We know how to load-balance. There are challenges associated with the scale of these data centers, but we know a lot more about that than what it looks like to build a microgrid to support a data center. In fact, the project that Michael talked about earlier, Colossus, is the only example of an off-grid data center that's been commissioned, and that was very, very quickly then connected to the TVA system and the grid. So it was only islanded for, I think, six to ten weeks, something like that.
Jamie Nolan: What's that running off of now?
Tim Hade: It's primarily running off TVA.
Michael Thomas: Yeah. It's got 300 megawatts of grid connection, and it still has another 200 megawatts of behind-the-meter, so they supplement on top of that.
Jamie Nolan: With old jet engines that are polluting the air for the community around the plant?
Michael Thomas: Yeah, essentially. And I would also say there are a couple more than Colossus now that are operating. There's one Vantage built in Virginia that's a little more than 100 megawatts, that's been operating since last fall. Stargate Abilene has their turbines operating; those are supplemental to a grid connection, so not fully islanded. And then Meta and Williams, the natural gas company, just commissioned a project at the end of July as part of more than a gigawatt they're building in New Albany, Ohio, known before the data center boom as the home of Abercrombie & Fitch, the clothing brand headquarters. So now home to one of the largest off-grid data centers in the world.
Jamie Nolan: What a claim to fame.
Tim Hade: So I think that, Jamie, is the big question. In order for those data centers to generate revenue, those power plants have to work. If you start seeing situations where these off-grid data centers are failing, and let's say a turbine blows up, now you have to go out and find a spare turbine. But the turbine supply chain is crazy, and you don't have spares, you don't have the right O&M components, and that data center's down for six months. That's a huge problem. So one of the things to look at, now that we have this first wave of off-grid data centers coming online, is: how do they perform?
But at the same time, I think there's also an acknowledgment inside the data center community that the best way to do this is to connect to the grid and try to distribute some of the benefits from the data center to the communities around it. If you can figure out that sweet spot, that's the way you really get social buy-in to build these massive projects. Ultimately, the way I think about it is: the net number of data centers we build over the next five years is probably not going to change that much based on how we build them. But what will change is how good it is overall for society. If we could figure out a way to get 90% of those data centers to be grid-connected, and to help build VPPs and help create load flexibility and help drive rates down for consumers, that's probably going to be a pretty good outcome for everyone. If all of those data centers are off-grid and they've got these big baseload gas plants attached to them, it's probably not a great outcome for normal people and society. Where we're going is at least directionally going to stay the same. How we do this is going to have a big impact on society overall, and what we're all trying to work on is how we do this in the best way for as many people as possible.
"THERE IS NO OFF-GRID DATA CENTER"
Jamie Nolan: Absolutely. And really indicative of this: after, Michael, you confronted Amazon about this, and they issued their statement, it says in that statement, quote, that the campus is powered by on-site generation that won't raise electricity costs for Texas families, and that Amazon believes in, quote, paying the full cost of powering our operations. So do we think this is a winning message for them? Do we think they're right, that going off-grid with gas is the neighborly way to do this? Tim, I think you're suggesting that they're completely mistaken and they've got this backwards. They obviously do not listen to this podcast.
Michael Thomas: I noticed that part of it as well, Jamie, and I think it's one of the things I'm actually concerned about with the policy response to data centers right now, which is the risk of unintended consequences. This project is an interesting example of what could come if the narrative is: all these data centers, when they connect to the grid, just increase our power bills. People point to examples like PJM, where at this point it's without debate that data centers are increasing the cost of capacity. And then the policy of somewhere like Texas is: don't connect to the grid, we're going to pause all interconnection requests. One of the risks in the short term is that they all just go off-grid in West Texas, because there aren't as many people there, it doesn't connect to the grid, and they get to tell this message of, hey, we're not connecting to the grid.
And even in that case, people say, well, it's going to connect to the gas grid, right? A developer told me: there is no off-grid data center, only an off-electric-grid data center. You still have to hook up to a gas grid.
Jamie Nolan: It's a great point.
Michael Thomas: But in West Texas, you have a market where there's actually a surplus of natural gas, because one of the byproducts of drilling for oil is you get natural gas, and there's not enough pipeline capacity to get it out to market. For 90 days this year, the prices at this hub were negative. So the tech companies, the developers, all these people can go there and say, hey, we're not connecting to the grid, we're using all of this surplus gas, so we're not impacting anybody. But they're going to be running on 100% fossil fuels, using the least efficient turbines, and the result is that you have the highest possible carbon intensity of a data center that you could imagine, compared to connecting to a grid, leveraging renewables, leveraging even just the grid we have today, which is much cleaner. Even a 100% combined-cycle turbine data center would be much more efficient than using these least efficient models that are being gobbled up right now. That's my fear: that the current political environment could push them toward these places, or to other countries where there's less regulation and more coal, and that on net it would be worse for the environment.
Tim Hade: The other thing about it, and this is not a comment on intent: I do think there's a general belief inside the hyperscaler community that if you build off-grid, that's a way to insulate your project from having negative impacts on ratepayers. But the point Jigar's made a lot of times is that it all comes from the same supply chain. Those gas turbines that are going to an off-grid data center are the same gas turbines that could be peaker plants to help support the grid. The conduit that's being used to build that microgrid is the same conduit that could be used for the grid. So the idea that there's no impact is just mathematically incorrect. There's definitely an impact on the grid when people take all the equipment we need for the grid and use it to build off-grid data centers.
But again, Michael said the words "unintended consequences." The thing to remember, and one of the reasons I try to separate what's happening from the intent of the people, is that everyone in this industry knows we're making this up as we go. Twenty-four months ago, this wasn't really a thing. We were starting to see, oh, demand for AI is going to be crazy and we're going to need to build a bunch more data centers, but almost all of this work has been accomplished in the last 24 months. People are moving really, really fast, and they're making mistakes as they go, and there are a lot of unintended consequences associated with building these projects. That's one of the things we have to think hard about as a community: maybe we're moving a little too fast on this stuff, and maybe we need to think a little bit more, because once these projects get built, they're going to operate for 10, 20, 30 years, and we're going to live with this for a long time. To some extent, you can make an argument that maybe now is the time to measure twice and cut once. And that's definitely not what's happening in the market today. It's just kind of: cut, and figure out what happens.
HAVE WE LOST OUR DAMN MINDS?
Jamie Nolan: And I'd love to know how Amazon as a company marries the concept that they're about to build the most polluting power plant ever constructed in this country with their climate goals. And also, you're saying, Michael, they're not even using best-in-class technology. I haven't heard anything about carbon capture and storage. Surely there are things they could be doing to make this project cleaner. For one of the richest companies in the world, it just doesn't make sense to me. And there's no way a project that is going to pollute this much is neighborly. You will never win me over on that point. It's not okay. And I also know people who work at the hyperscalers who are very deeply thoughtful and caring people, and I understand these are big bureaucratic corporations, but I challenge people to throw themselves in front of this bus a little bit. Surely this can be improved upon. It's like we've all lost our damn minds, and we need to pause and think about what we're doing here. We are facing a historic El Niño, and terrible severe weather across the country, and a historic heat wave. A, B, C, D, and E: have we lost our damn minds? What are they doing? I would expect better from a company with climate goals. So I do hope that somewhere along the way, my hopes are low, it's West Texas, but that somewhere along the way they're challenged to just be better neighbors and global actors.
Michael Thomas: I think it's one of the challenges with these voluntary climate targets: at the end of the day, there's not a lot of teeth. And so when there's a trade-off of earning this revenue, and in the case of Amazon, they're making a huge amount of money on building this data center capacity. A lot of people talk about the AI bubble and whether these companies are making money, but on their latest earnings call, the CEO said they're getting a three-year return on investment. When they build a data center and stick it full of chips, they get that return in three years. The kind of money they're investing, their expectation is you're going to see this growth in revenue like they saw when they built AWS.
So I think the challenge for all of the folks on the energy teams at these companies is that they're going to the board or management or whoever it is and saying, hey, we could go do this on-grid and support these renewables, and it's going to take six months longer, because the grid takes a long time, or because there are these complicated interconnection queue processes, or whatever the reason is in that market. And right now they're saying, well, we can't do that, because we're in a race. And they kind of have the blessing of the Trump administration, who is saying, well, we're in a race with China. Everybody's racing, and the money that can be made in the short term is just so astronomical that right now you're seeing companies choosing to get that revenue rather than pursue their climate targets.
WHAT SHOULD THE UTILITIES DO?
Jamie Nolan: It's depressing. I mean, I know you're right. One more thing. I want to make sure that we pull in the utilities here a little bit, because they sit in the middle of this whole story. It's their seven-year average interconnection queues that are pushing these tech companies toward building their own power in the first place. And yet Google has shown that utilities will sign flexibility deals when someone asks. I would actually go as far as to say I think Google is doing the best job of all the hyperscalers in how they're going about this AI build-out. So what do you think the utilities' role is in fixing this, and what should they be doing that they're not?
Michael Thomas: Tim, as the resident expert in virtual power plants, I'd be curious to hear: what do you think is the difference between a situation where Google can get a gigawatt of flexibility, and one where a company is choosing not to? I haven't followed the flexibility space as much, but I'd be curious to hear.
Tim Hade: I mean, look, I think the answer is culture, and Jigar's talked a lot about this. We're in this situation where everyone's interests are somewhat aligned. I don't believe that tech companies want to build off-grid data centers with baseload natural gas power plants, and utilities definitely don't want that to happen, and society as a whole doesn't want that to happen, and the clean tech industry doesn't want that to happen. But we have to provide a pathway for getting this stuff done fast.
The challenge we have right now is that we built an electricity system that was very, very good at maintaining the status quo, and that's what we lived in for pretty much our entire lives. From the late nineties until today, there was very, very little load growth. The grid operated a certain way, and it worked, and we didn't need to make big changes, so we just got good at maintaining that stasis. Now we're in an entirely different situation, and this machine that we've built isn't good at ramping fast. Now, it can be. The same people that built the first electric grid are us. We can do this. We've done innovative things before, and we can come together and figure this out, and there's a lot of reasons to be optimistic. We have all the technology to do it. We have the Liftoff reports and all this kind of stuff that blazed a path for us to follow. But we're in a culture that was built to maintain the status quo, and now we're in a period of dynamic growth, and so far we're not doing great at navigating that.
Ultimately, what has to happen is the utilities have a leadership role to play here. They have to come together and start saying: okay, look, we're going to figure out a way to get this done. We're going to reach out. We're going to work with the industry. We're going to work with state, local, and federal governments to figure out how to do this the right way. But I don't think the utilities can do it themselves. There's also a role for the state and federal government to play, and that's happening to various degrees in different states, working a little differently everywhere in the country. And then there's a huge role for us as an industry to play. We get to decide what kind of projects get built, and what the cost of those projects is, and things of that nature.
So I think it's really about collaboration, and understanding that this is a once-in-a-generation economic opportunity. That's true. And these national security issues are a real thing. I do not want to live in a world where we're at a material disadvantage to China in the AI race. But also, there's a way for us to do this that sets a good example for the rest of the world and shows the rest of the world how to do this, and we're not doing a great job of that right now. It's ultimately a situation where we have to be very, very clear about what we want to do, and then organizations have to be willing to change. I think that's starting to happen right now. Whether it's going to happen fast enough is an outstanding question. But it's not a passive thing. We as an industry can go in and drive some of these solutions and compromise and find better ways to do things, and I think that's starting to happen in a big way. Change is needed, but it's also happening, and the question is just: is it going to happen fast enough for us to end up with a good outcome here?
Michael Thomas: The only thing I'd add to that is that utilities are responsive to the regulation in their state. We saw this in the build-out of solar and wind, where renewable portfolio standards were what pushed utilities to make those early investments. When they went into their planning process, which they have to do every year or every couple of years, they'd run all their models, and the model might have otherwise said, let's build this coal plant or let's build this gas plant. But instead there was a constraint that said: you have to build, early on it was a gigawatt, two gigawatts, three gigawatts of solar or wind. That really jumpstarted the first phase of clean energy technology deployment, that and what other countries like Japan and Germany were doing.
One of the things I'm watching for is: what does a policy response look like to this? How do you encourage data center development that is good for communities, that is good for our clean energy goals and climate goals, and isn't so restrictive that it pushes developers out of the state, so they go to whatever state is willing to let them build whatever they want? But one that takes this opportunity that we do have, that a lot of people talk about: these companies have billions, trillions of dollars that they are going to invest into this, and power is a small part of it, relatively speaking. If you can get them to invest in advanced geothermal, or, depending on your views, advanced nuclear or an AP1000, or whatever your clean energy technology of choice is that you think can get us to 100% clean energy, is there a way to develop policy that encourages that kind of development? Some states are trying different things, and everyone's trying to figure it out. But that's something I would like to see: a policy response that encourages those types of investments.
A DATA CENTER PLAN FOR EVERY GOVERNOR
Jamie Nolan: That's such a great point. Especially going into this fall's midterm elections, I believe 36 states are seeing governors' races. When we had Secretary Jennifer Granholm on the pod, she spoke a lot about what she would advise those gubernatorial candidates to do, and one thing she said is: you've got to have a plan around data centers. So I'll definitely be looking to these candidates, and then the elected governors, to see what they're proposing in terms of meeting this matter head-on, because it's a huge issue. Like you said, there's so much economic development to be left on the table if you don't get it right. And hopefully they can stop some of this insanity, because 7.65 gigawatts of off-grid gas in the desert with no carbon capture and storage, it just boggles the mind.
So, Michael, thank you so much for being here. It was great having you, and you are as good of a communicator verbally as you are in writing. We've got to have you back. We definitely want to have you back when your report comes out.
Michael Thomas: Oh, thank you so much. I really appreciate you both having me on, and I enjoyed the conversation.
Tim Hade: No, man, for real, thank you so much. I learn a ton from your work, and I know a lot of people do. So keep it up, man. It's super important.
Michael Thomas: Oh, thank you. I really appreciate that. That's nice of both of you to say. And I've been listening to the podcast, and I especially enjoyed the Intersect Power interview recently. That was a good listen.
Tim Hade: Sheldon's the man.
Michael Thomas: Yeah, that was awesome.
DEBRIEF
Jamie Nolan: Well, that was fun. We did it. I was telling you off camera that I've kind of gotten sick of talking about data centers, but Michael actually did make the topic fresh for me, so I'm really glad he was our guest today.
Tim Hade: Yeah, for sure. He is awesome. Trying to find true north on these issues is so complicated in the world today, and hopefully one of the things people take away from this is they go out and check out Cleanview. Look at some of the reports they've produced. You'll get a lot of good information, and you'll learn a lot more about what's going on in the sector today. If we can agree on the facts and what's happening on the ground today, that forms a foundation for us to have conversations about how we make progress. One of the things that's really frustrating overall is that everyone's working off a separate set of facts. What Michael is doing, all the information he communicated to us today and all the work he's doing at Cleanview, provides that foundation so we can start to make progress, and I'm incredibly grateful to him for doing that.
Jamie Nolan: I mean, I think there's just such a lack of transparency right now into what's going on with data center development. There are wildly different estimates for how much capacity will even be developed in the next few years, and no one is doing a better job than Michael at actually digging into it.
Tim Hade: So, Jamie, how did I do? How was I? This is my first time guest hosting a podcast in the history of my life. How did I do?
Jamie Nolan: Oh, you're usually just a guest. You were like a blended guest-slash-host, because you provided a lot of commentary, which I appreciated, because you're an expert on this topic. You did a great job with that. Very funny. Your answers weren't too long. I did have to ask several of your questions from our run of show. I definitely had to jump in and ask your questions, but we got through it, and we only went a little bit over time. So I'm going to give you an 8.5 out of 10, so that I can get you to come back, so that I can get you to the 10 next time.
Tim Hade: I will take it. Well, full disclosure, I read the run of show and then completely went off script once we started, because I was just interested in asking Michael the questions I wanted to ask Michael. So that was a curveball for Jamie. I feel like you handled it great. I also feel like, as my media consultant, Jamie is a lot meaner to me when we're not recording this for a public audience. So I'll get the real feedback later, and we're going to continue this journey of self-improvement. But I very much appreciate you having me and teaching me how to be a good podcast host. And next time I'll be better.
Jamie Nolan: You're a natural. You absolutely didn't even need the run of show. Everyone who has worked with me knows I do like things just so. I am very organized and I'm incredibly thorough, and so when there is a run of show, I do like to hit the run of show. And we did a really good job. Some things were out of order, but I still think that it made sense. So I am very proud of us, and I can't wait to have you back. Next time Jigar goes on vacation, you're invited.
Tim Hade: Okay, so like 2028. He shouldn't go on vacation.
Jamie Nolan: We'll see. We'll see how this one goes. He's not working, which means my daily number of messages has dropped from 25 to 10.
Tim Hade: He's never not working.
Jamie Nolan: He's never not working. That wouldn't surprise anyone who knows him. But bless his wife's heart, she's trying very, very hard to preserve their family time, and I'm supporting her. I'm not sending him any requests. So, Jigar, if you're listening to this and you're still on your Alaskan cruise, go back to hanging out with your family, and we'll see you next week.
Tim Hade: Amen.