Michael Thomas: Inside America's Off-Grid Data Center Boom
Michael Thomas is the founder and CEO of Cleanview, a data and research platform tracking more than 10,000 clean energy projects and thousands of data center projects across the country. Back for his third appearance on Inevitable, Thomas has spent the past year investigating the explosive growth of AI infrastructure and the increasingly extreme measures companies are taking to get power online.
In this episode of Inevitable, Thomas explains how the AI boom has transformed the data center market and why companies are increasingly building their own power generation rather than waiting years for grid interconnection. He breaks down the rise of off-grid and “behind-the-meter” data centers, including xAI’s Colossus, Meta’s rapid-deployment data centers, and Amazon’s massive natural gas project in Texas.
The conversation also explores the economics driving the race to power, the environmental and political backlash against data centers, and whether building gas generation today will accelerate or undermine the transition to cleaner energy. Thomas shares what his research reveals about electricity prices, water use, local pollution, and the growing public distrust of Big Tech.
Finally, Thomas discusses the tension between AI growth, U.S. competitiveness, and climate goals—and why he sees providing objective information about these projects as one of the most important roles he can play.
Episode recorded on September 2, 2026 (Published on September 16, 2026)
In this episode, we cover:
(0:00) An overview of Cleanview
(2:44) How policy uncertainty changed Cleanview’s business
(4:15) Why Cleanview started tracking data centers
(6:16) How Cleanview investigates infrastructure projects
(7:49) Using satellite imagery to track data center construction
(8:57) How Cleanview uncovered Amazon’s gigawatt-scale gas project
(11:26) Investigative reporting on secretive data center projects
(13:16) Why AI changed the economics of data center power
(14:44) How xAI’s Colossus pioneered off-grid data centers
(17:25) The explosion of behind-the-meter power
(19:13) Meta’s rapid-deployment data centers
(22:58) The economics behind the race to power
(25:29) Why available power commands a massive premium
(26:35) When building your own power plant doesn’t make sense
(27:06) The regulatory challenges of off-grid gas generation
(29:02) Why Amazon’s Indiana data center got online so quickly
(30:32) How off-grid power can become a bridge to the grid
(32:16) xAI’s plans for 10–20 GW of off-grid capacity
(33:14) Why data centers have become a bipartisan political issue
(34:11) The real drivers of public opposition to data centers
(36:17) Are data centers raising electricity bills?
(40:00) Why data centers should pay for their infrastructure
(41:08) Data centers and the water-use debate
(43:48) How power generation contributes to data center water use
(46:19) The carbon emissions of AI infrastructure
(48:16) The noise and local impacts of off-grid power
(50:54) Can AI accelerate the development of clean energy?
(52:00) Why the data center boom could slow decarbonization
(54:26) Where the public can have a voice in AI infrastructure
(56:40) Why better information matters for the future of AI and energy
-
[Cody Simms] (0:00 - 1:54)
Hey, it's Cody. Before we get into today's episode, I want to tell you about something happening during Climate Week NYC. Jeff Lawson founded Twilio and turned it into a $4 billion public company.Now he's running Inertia, building laser fusion, and he just closed a $450 million series A. We're taping a live episode with him on September 22 in New York at the JP Morgan Chase Tower. The link to join us will be in the show notes or just go to inevitable.fm/NYC2026. Hope to see you there. From MCJ, I'm Cody Simms, and this is Inevitable. Each week, we talk to the people powering the world and keeping it running to learn what comes next.
Today, on Inevitable, our guest is Michael Thomas, founder and CEO of Cleanview, back for his third appearance on the show. Michael has been diving into some of the largest data center projects in the country to understand how they're getting built, and in some cases, the extreme measures they are taking to source power. He publishes excerpts of his findings at his newsletter and has even deeper source materials via his subscription service, Cleanview, which tracks more than 10,000 clean energy and data center projects across the country.
Michael brings a unique perspective as a data provider for the energy space, as well as a pseudo-investigative reporter. He always brings deep original insight, and today's conversation is no different. Michael, welcome back to the show.
[Michael Thomas] (1:54 - 1:55)
Thanks for having me. I'm happy to be back.[Cody Simms] (1:56 - 2:07)
You might be starting to rival Jigger, I think.This is what, your third time on? You're getting up there, like Christopher Walken, I guess, from Saturday Night Live, in terms of appearances, so this is awesome.
[Michael Thomas] (2:07 - 2:10)
Well, I appreciate you having me back. It's good to chat.[Cody Simms] (2:10 - 2:44)
Last time you were on here, we were in the thick of the One Big, Beautiful Bill discussions where a lot of the Inflation Reduction Act rollbacks were being threatened. And, of course, as we all know, many of those actually did get rolled back and removed from legislation. You have found lightning in a bottle yourself, even in spite of that, for your business, Cleanview.Maybe start by walking me through how you have evolved the business over the last year or so and what you've been spending your time on.
[Michael Thomas] (2:44 - 4:11)
Yeah, the last time we spoke was about a year and a half ago, I think, in May of 2025. And that was the beginning of a pretty challenging period for me and the company, because starting in April after Trump did his first round of tariffs and then in May, with the first house version of what would become OBBBA, there was just this huge uncertainty. And so the entire renewable industry basically froze for between three and six months, depending on your business.People weren't sure about if they were investing in projects. They were not sure if they were going to have to cut their budgets in half. And so most people in the space really just struggled because of that uncertainty.
It wasn't even necessarily the policy outcomes. If they had just signed a bill in May, everyone would have known and would have kind of moved forward, but there was so much executive action and things that were going to take 90 days to get through. And so the business was growing right before then.
Cleanview was at that time really focused on tracking renewable projects, trends in renewable energy. And then it just sort of went to zero for about six months. So during that period, I really had to think about one, was there a future of a renewable energy industry in America?
There were some moments when I worried it was that dark.
[Cody Simms] (4:11 - 4:14)
That is a bleak thing to even have to consider. Oh my gosh.[Michael Thomas] (4:15 - 5:39)
Yeah. And I think that was maybe the more extreme versions of my thinking. I think that even the modeling at the time was showing that without any tax credits at all, in the most draconian scenarios, there'd still be a market.But yeah, would it be a market that was declining by 50% every year? And so I really was just thinking about how best to position the business and be resilient. And at the same time, we were seeing this explosion in data center demand.
And all of our customers were asking us all these questions like, hey, you're tracking all of these infrastructure projects on the solar, wind, and battery side, are you tracking data centers too? Because that's the demand side of the grid. And so in July, we started to kind of build the first versions of a data center tracker.
And then by the fall of last year, we launched that first version. It was immediately very interesting to our customers and to, I think, just the broader public. Since there is so much uncertainty around this data center buildout that people have so many questions, they want to know how many projects are there, where are they being built, which ones are going to be built.
And so we launched that first version and then have just kind of improved it since then as a second product, really alongside our renewable project tracker. And fortunately, that product has continued to grow as well. And so we're still tracking solar, wind, battery projects, but then we're also tracking the demand side of the grid as well.
[Cody Simms] (5:40 - 6:15)
And you continue to follow just, in my opinion, an incredibly smart content marketing strategy in that you have this database that's behind a paywall that is an enterprise product that people can buy. But you also publish roughly every few weeks a pretty significant exposé on something you've discovered in the data that maybe you can share where people can find those articles. But we're going to spend some time today discussing some of the things you've published over the last six or nine months that I think are in many cases truly original reporting.[Michael Thomas] (6:16 - 7:03)
It's been really fun to be able to do those stories because one of the things that I've learned starting my career writing for general audience about a lot of topics. So I wrote a newsletter about clean energy and climate change where I would talk about everything from EVs to solar to industrial decarbonization. And then in starting Cleanview, I've been able to really focus and develop an expertise and learn a lot about some of these infrastructure projects.And there's just a lot of really interesting stories in the data that we're looking at. It sounds incredibly nerdy, but there's a lot of interesting stories in public documents, in permits, in satellite images. And so it's fun to be able to tell those stories and share them with a large audience.
[Cody Simms] (7:04 - 7:48)
Some of the data I pulled around the amount of things that you are tracking, and this may already be out of date, it's 10,000 plus clean energy projects, 500 plus clean energy developers. And then just since your move to data centers, somewhere around 3,500 data center projects nationally, of which 1,300 of them are operating and over 2,000 of them are planned with hundreds of megawatts of capacity, maybe even much bigger than that at this point. And hundreds of developers on the data center side, which has grown by an absolute multiple over the last year.Where am I undercounting there?
[Michael Thomas] (7:49 - 8:37)
Those are all the numbers. The only one I'd add that has been more recent this year is we're doing really detailed construction tracking on the data center side. So we're tracking using satellite imagery, whether projects are under construction, if they are at what stage of construction are they at.And so we can really see a campus ramp up over time, or see in some cases like we did a big story on Fermi America before they had found a tenant that this was a multi-billion dollar IPO. They were telling investors that they had begun construction. And then we found some satellite imagery showing that they had actually paused construction.
And they were only doing land clearing, which is sort of the cheap version. And so we published a big story and shared that with a lot of media outlets. And so we can really see what's happening in real time.
[Cody Simms] (8:38 - 8:57)
There are a lot of these like nameplate gigawatt scale projects that get announced, Fermi being one of them. Maybe walk through what your process is to go explore one of these. Feel free to be as specific as you like.And what the sort of document trail tends to look like for you guys.
[Michael Thomas] (8:57 - 9:11)
Everything starts with some automated tracking that we do where there's a number of building permits or air permits, SEC filings, different documents that will pop up on a regular basis at the same location.[Cody Simms] (9:12 - 9:15)
This tends to be before or after a press release has already been issued?[Michael Thomas] (9:15 - 9:49)
Sometimes before, you know, we'll know about a project like in the case of a really big project and story for us, Amazon Gigawatt Ranch. So I can kind of talk about the process here. So we had identified a air permit in January of 2026 for a very large natural gas power plant that was going to be off-grid in Texas.But it was proposed by this developer that we had never heard of. And most people would have never heard of. They had no experience building natural gas power plants.
And so it looked like a speculative project, like a lot of them that we're seeing around the country.
[Cody Simms] (9:50 - 9:52)
Something something something LLC, right?[Michael Thomas] (9:53 - 11:17)
Yeah. And it's going to be 7.65 gigawatts. Even though they've never built anything before.We were skeptical, but we continued to kind of follow that one. And then in August of 2026, so just a few weeks ago, we saw a building permit pop up in the state of Texas. And in Texas, you have to get a building permit if you want to build a data center.
And that goes to a state database. And we looked at the location of that building permit, and it was the same as the gas power plant. And then we looked at the construction start date on that building permit.
And compared it to satellite imagery. It had just begun construction on that same schedule that Amazon was saying at the site of this gas plant. And so we thought we had enough information to be pretty confident that Amazon was actually behind what would be the largest natural gas power plant ever built in America, and would not connect to the power grid. And we reached out to Amazon.
Most of the time companies don't respond. They don't want to talk about projects. They'll give us a generic PR response.
But in this case, to Amazon's credit, they got on the phone, had an hour long conversation, they verified everything that we had found, and said, yep, we're building that. Or we've acquired the project and we'll be building it. And then we published the story a couple hours later.
And then the next day, it was in the New York Times, and the paper, and the weekend edition, and then some international headlines after that.
[Cody Simms] (11:17 - 11:25)
Have you had any where it's been more Woodward and Bernstein-esque, where people won't return your phone call and will pretend like something's not happening?[Michael Thomas] (11:26 - 12:25)
Yeah, I won't name the company. I'll say it's one of the top five largest companies in the world. And we had found a natural gas power plant that we suspected the company was behind.We just wanted to verify the information. And actually, in this case, someone had reached out through a back channel, had heard that we were kind of working on the story, and said, this is a complicated project. You need to know these details.
I can't tell you these details. So it really complicated the process because we got some information. But then we were able to actually take some other information we had found that they didn't know that we had to really verify the story.
And so we were able to publish what I think was a really fair story and describe the nuance of the situation. But let's just say that companies would prefer to own the narrative, put out their statement on their own timeline. And so having a small market research platform, publishing a bunch of documents and looking at satellite images of their sites is maybe sometimes a nuisance.
[Cody Simms] (12:26 - 12:30)
Yeah. I guess, thank goodness for free speech, right?[Michael Thomas] (12:30 - 12:34)
Yeah. And some good permitting paper trails as well.[Cody Simms] (12:34 - 12:34)
And public records.[Michael Thomas] (12:35 - 12:35)
Exactly.[Cody Simms] (12:36 - 13:16)
You know, it strikes me that this whole movement of off-grid gas data centers, and that phrase gets thrown around so much, maybe you can describe really what that actually means, was largely pioneered with the Colossus projects by xAI, which I think to date are still the largest ones that are actually running, like truly operationally running that would fit that description. Can you maybe describe what that project was and how unique it really is in terms of the amount of conversation we hear about off-grid gas?[Michael Thomas] (13:16 - 14:35)
The typical way that a data center was built for the last couple of decades really was that a developer, maybe it's Meta, maybe it's Google, Amazon, building their data centers to power all these products that we use, they would go and find a site that had land, fiber, water, and then they would call up the utility and say, hey, we want to connect this for a hundred megawatts of power over the schedule. And they would, in some cases, get in a little spreadsheet with like five other large loads or large industrial customers.It wasn't a very formal process, and it wasn't power itself wasn't that important of a consideration for these companies. And what changed after the launch of ChatGPT really, and this AI boom, was that there was this explosion in demand from data centers, and a huge increase in the size of these data centers, in part because of these scaling laws where the larger the data center is, the more concentrated the compute is, the better the models, at least up until now. And so they want to connect a gigawatt.
Well, connecting a gigawatt is no easy task, and it might take five years, or in the case of Dominion in Virginia, which is data center alley, about 20% of the country's compute today. They were telling developers over the last couple of years, it would take about seven years to connect.
[Cody Simms] (14:36 - 14:44)
In all of your articles, you tend to remind readers that a gigawatt, as much as we hear it in all the press releases these days, that's like a city scale's worth of power.[Michael Thomas] (14:44 - 16:05)
Yeah, it's a nuclear power plant. New York is 10 gigawatts peak load. So it's a lot of power.And so in August of 2024, xAI, Elon Musk's AI company, went to some utilities and tried to go through this process, and they were told that it was going to take years to connect data center that was as large as they wanted for what would become Colossus. And Elon Musk, being Elon Musk, sort of taking his own unique approach, in many cases, very unpopular approach, they just decided to wheel in some backup generators and turbines on the back of literally semi trucks. Like if you look at a graphic or an image of these, it's just a semi truck.
And then on the back, it's a trailer and inside is an industrial turbine. And they wheeled in a couple dozen of these. And within a few months, they were able to connect a very large data center and get it online.
And that's compared to a typical timeline, even if you have power of two years, for example, to average meta data center that we've tracked historically. And then they scaled it up to more than a gigawatt. So very soon, it was the largest data center in the world.
Today, it kind of goes back and forth between some of the other large data centers, but it's likely the largest data center in the world.
[Cody Simms] (16:06 - 16:10)
When you say industrial turbine, describe what that actually is.[Michael Thomas] (16:11 - 16:56)
So historically, we've had industry sort of co-locate power in America. And so what that means is someone who's running a factory may also have a steam turbine that can generate some of their own power. And so there's these turbines that are sized specifically for that.In the case of xAI, they were using turbines that were actually meant for oil and gas fields. So if you think about drilling for oil and gas in the Permian Basin, you're out in the middle of West Texas, hundreds of miles from any infrastructure, and you're drilling these wells, well, you need a lot of electricity or power to run this equipment. And so they'll use diesel generators or in some cases gas generators, they're using industrial turbines like that in that case.
[Cody Simms] (16:57 - 17:25)
xAI rolled these in, created this gigawatt scale facility, which is actually producing power at that scale. Where are we today in terms of the number of off-grid projects that have been announced or that you all are tracking and know are in the works? And from a power capacity perspective, how much reservation is there in that universe?And how much is actually producing today?
[Michael Thomas] (17:25 - 17:58)
Since Elon Musk built xAI with his team, built Colossus, there has been this explosion in off-grid data center announcements. We're now tracking more than 100 gigawatts of these behind-the-meter data centers. Many of those have no plans to connect to the grid.In some cases, they would connect to the grid, as in the case of Stargate in Abilene, where they've built their own turbines with Crusoe, but they're still relying on grid power. So it's sort of like a super backup, where maybe the grid...
[Cody Simms] (17:58 - 18:01)
I think the phrase for that is bring your own capacity. Is that right?[Michael Thomas] (18:01 - 18:57)
Yeah, it's sort of a subset of that, where they're bringing their own capacity on site, and they're in charge of that generation. So that generation is under their control, and they can help get a site online faster that maybe could provide 90% of the power, but the utility tells them, hey, listen, in August, we really can't give you any power from 3pm to 6pm, as air-conditioning units are running, and we're kind of tapped out. And so since xAI, there's been this explosion, and there's a lot of announcements. The important thing to note, as you alluded to, is that a very small fraction is actually operating.So we're looking at a couple of gigawatts right now operating, and then we're in the middle of our latest update report on this, but probably 5 to 10 gigawatts that's under construction. So it's accelerating very quickly, but it's still most of the gigawatts are in the announced stage.
[Cody Simms] (18:58 - 19:13)
You've described this in your writing as the Mad Max phase of power development. Maybe give a few more examples of some of the extreme measures you've seen companies take. I like your Meta story that you wrote about, maybe as an example.[Michael Thomas] (19:13 - 20:34)
Meta is certainly the next best example. I wrote a story about their data center strategy a few months ago and described it as this Mad Max phase of the data center buildout, where Meta, in about 2022, realized that their data centers were taking about two years to build. And part of the reason for that is they had really perfected the most efficient building in the world in some ways. This was highly efficient water use, highly efficient electricity use by the metric in the industry, very low PUE, which meant almost all the electricity was going to economically useful activities.And they did that over more than a decade. Starting in 2022, you can actually see this in time lapse satellite images, where they were building the next version of these buildings to continue to buildout their data centers for Facebook and WhatsApp and all their services. And then you see the building just get demolished.
It's like halfway under construction, and then they tear it down. And then in its place, there's this tent, just like a hundred and something thousand square foot, tent. And you look at some images from helicopters that have shot these, or a kind of product footage from these companies.
And it's a rapid-deployment structure as the formal name, I think "tent" is a better description of it, because that's exactly what it looks like.
[Cody Simms] (20:35 - 20:44)
It reminds me of the early days of the Tesla Model 3 production lines, where they set up, I believe, tent production facilities as well, right?[Michael Thomas] (20:44 - 21:15)
Exactly. They've taken a play out of that playbook. And so what they were able to achieve was a 12-month timeline.So they cut the time in half. So they were able to buy up all these NVIDIA chips, put them in a tent. In some cases, they have no backup generators.
So these are probably going to training Llama and other Meta models, Spark, and not to serving Facebook, which needs a certain amount of uptime. But yeah, I mean, people are doing anything they can to build a data center a few months faster.
[Cody Simms] (21:15 - 21:17)
And how are they powering that facility?[Michael Thomas] (21:18 - 22:10)
In New Albany, which is their largest cluster of these tents, there's more than 10 of them in this business park in New Albany, Ohio. New Albany, Ohio has become a hub of data center development, formerly best known as the home of Abercrombie and Fitch headquarters. A different industry is sort of graphing up.And that cluster will be over a gigawatt. It's their Prometheus project. And they're powering it today with 200 megawatts of off-grid, totally islanded natural gas that they built with the Williams company, a natural gas pipeline company.
And then they're building another 200 megawatts that will come online in a few months here. And then they're expanding that even further to more than a gigawatt of off-grid gas at their campus there. And more than a dozen tents will eventually be there.
[Cody Simms] (22:11 - 22:58)
Let's talk about the economics of why companies are doing this. Because clearly, I can't imagine any big tech company wants to be a power company. Their core competency is not operating gas turbines in the middle of tent conditions.Their core competency is building consumer product and operating technology solutions. And there is a reason, though, that they aren't just willing to sit around and wait for a plug into the grid, that they are taking matters into their own hands. Can you describe the economic reason that they may be willing to invest all of this time, energy, capital, and risk, frankly, into doing these things?
[Michael Thomas] (22:58 - 25:03)
So I think if we start at the highest level, the largest driver of this is the demand for compute and for tokens or AI right now. So if you look at the revenue of Anthropic and of OpenAI, many listeners will have already seen this chart. But Anthropic started the year at about $10 billion of annualized revenue, and they're now at more than $65 billion.So they're the fastest growing company of all time. And every billion dollars of compute that they're selling is power somewhere, and a data center running somewhere. And so that's the growth that we're seeing.
And that's like software growth, or you can think of it as bite growth. It's much harder to match that in terms of the physical world, the world of atoms, where you're actually building tents or data centers. And then if you move down and where that $65 billion a year is going, in terms of where Anthropic is spending, it's going to AWS, which is seeing the fastest growth they've seen in, I think, five or 10 years.
And it's going to these data center developers or neo-clouds like Crusoe and other companies that are focused on building data centers. And the economics for a company like that, up until recently, it was about $10 billion of revenue per gigawatt of capacity. So if you build Colossus a year ago, you could expect $10 billion of revenue.
Well, if you look at the deals that SpaceX, which owns xAI, signed with Google and Anthropic a few months ago, the numbers are closer to like $30 to $50 billion per gigawatt. And so those are the economics that are driving this speed to power desire is every year that you can get online sooner is $10 to $50 billion of revenue. And in many cases, at good margins.
And so they're just trying to get online as fast as possible to capture that.
[Cody Simms] (25:04 - 25:28)
One could argue it's revenue and it's also mind share and market share. I would say xAI might be a good example of that where they weren't really even in the discussion a year or more ago, relative to ChatGPT and Claude and Gemini. And obviously they were first to market with this off-grid strategy.And now I would say are squarely in the midst of the discussion amongst the top models.
[Michael Thomas] (25:29 - 25:56)
And there's also like a decay function with all this where SpaceX can charge $50 billion per gigawatt. But if you look at power booked out a year from now or two years from now, these longer duration contracts, it ends up being closer to $10 billion per gigawatt. And so there's this incredible premium for available capacity.And the reason that Anthropic and Google signed with them was that it was the only available compute really probably in the world at that scale.
[Cody Simms] (25:57 - 26:34)
We've given a few examples now of behind-the-meter off-grid strategies that did work and that did accelerate time to market. But I think your reporting also shows it doesn't always work. And in some cases, building your own gas plant isn't necessarily faster than waiting for a grid interconnect.Have you come down yet on maybe describe a few examples where the grid connection actually ended up being a winning strategy. And have you wrapped your head around yet what those conditions are that matter in terms of making the decision of which direction to go?
[Michael Thomas] (26:35 - 27:03)
I've heard it described by some folks that I've talked to in the industry as there's no off-grid data center. There's an off electricity grid data center. But if you're off the electric grid and you're powering it with natural gas, you're just on the gas grid because there's gas pipelines that run across the country.And it's not like you're just like flying them in. You're connecting to a bunch of pipes that are all interconnected from wellheads and span the whole country.
[Cody Simms] (27:03 - 27:06)
And still under FERC jurisdiction, et cetera, et cetera, right?[Michael Thomas] (27:06 - 28:50)
Yeah, and really unpopular and lead to opposition and lawsuits. And so we're seeing this with one of the Stargate projects. So Stargate is a number of different data centers being built across the country with OpenAI and its partners.And their Stargate project in New Mexico, Project Jupiter, the plan initially was to build it with these natural gas turbines. Then they had challenges getting an air permit from the state of New Mexico because of the pollution concerns. That would have been the largest single source of pollution in New Mexico, larger than almost every gas power plant, and really, point of pollution in the country. And so the state of New Mexico was challenging that and they pivoted and said they're going to use Bloom fuel cells, which are still fossil fuel based, still result in carbon emissions, but have lower local air pollution and NOx emissions, for example.
But they need a huge amount of gas to power this regardless. And right now they're struggling to get a gas pipeline permitted. There was initially a lawsuit, the state land authority blocked the initial path.
That's now going to go to FERC. And so it's going to just go through all these different regulatory not loopholes, but hoops and challenges. And it could delay the project by a couple of years.
Now, if you compare that to Amazon's new Carlisle data center, which at the beginning of this year, it was the largest data center in the world before Colossus surpassed it, that data center was built with the electric grid. It took about two years to get a gigawatt of compute online. So incredibly fast timeline, and they didn't build any new generation or long distance transmission lines.
[Cody Simms] (28:51 - 29:02)
So that flies counter to the whole narrative of to go time to power, you got to be a cowboy and build it on your own, and that interconnects are seven years plus. How did they get that done?[Michael Thomas] (29:02 - 29:07)
That site is probably one of the last great sites in America for a data center developer.[Cody Simms] (29:07 - 29:12)
Location, location, location, right?[Michael Thomas] (29:13 - 30:05)
So the local government there in the state of Indiana had been preparing that site since 2017.And it was a almost pre-permitted set of land that was meant to be connected to AEP Ohio's high voltage transmission grid. So it could get a huge amount of power very easily. My read of it was that they were initially thinking that maybe an auto manufacturer would go there, maybe a battery manufacturer, someone who needed a lot of power and needed permanent land.
And then Amazon struck a deal with them, and it was already powered, permitted. They were able to accelerate the timeline a lot, so they were able to build it in two years. But there's not a lot of those sites.
And if you look at what Amazon built in Indiana next, it's going to trigger more than two gigawatts of new gas. So that sort of shows that there's probably not a next site available.
[Cody Simms] (30:06 - 30:32)
Let's talk about the bring your own capacity strategy, which is this notion that most of these tech companies don't ultimately want to be running power. They want to be grid connected. They're building power in the short term to get up and running.They might have to continue to operate some of that power at peak load, et cetera, but ultimately want to interconnect. Have we actually seen that happen yet?
[Michael Thomas] (30:32 - 31:57)
Interestingly, the first off-grid data center is also the first data center that has now connected to the grid. So Colossus, the first xAI data center that got so much attention and controversy when it was first built, was initially running on these mobile generators. And then they got a 150 megawatt substation built on site, and then another 150 megawatt substation built.And now they're moving those turbines. You can see month by month in the satellite images where those turbines are now going to Colossus 2, which is a site across state borders. They angered everyone in Tennessee.
And so they went to Mississippi where they were able to get a permit and have now been building up Colossus 2. But Colossus 1 is at this point primarily grid powered and only using those mobile generators for a boost on top of that grid power or backup. But they've bridged.
And I think it's an interesting example, because I think when a lot of people talk about this bridge solution, they assume that the bridge sort of goes away, that you build the backup turbines, and then you move them off site and kind of retire them. But in this case, they were a bridge to the grid, and then they just moved those turbines to a new site. And so it's still gas generation that is burning a lot of gas and having real impact.
[Cody Simms] (31:58 - 32:15)
I wonder how they think about that CapEx themselves, if they are truly reusing the invested CapEx across multiple projects. It feels like a cost structure strategy that maybe a lot of folks aren't paying attention to.[Michael Thomas] (32:16 - 32:54)
Right now, SpaceX has incredible economics on their data centers because they're able to sell at this premium. So these $50 billion per gigawatt equivalent deals were hugely profitable for xAI. And so we've just seen them accelerate their plans.They've now announced plans to build 10 to 20 gigawatts of off-grid data center capacity by the end of 2027. Now the asterisk there is the typical Elon Musk timeline asterisk, which is that's what Elon Musk is saying. And it may be like the Robotaxi or some of these other ones.
[Cody Simms] (32:55 - 33:13)
You're watching all this through a data and policy lens. There's also a political story running in parallel, which I think your Tennessee and Mississippi transfer story is a good indication of, which is the public opinion on data centers is, shall we say low?[Michael Thomas] (33:14 - 33:57)
It's hard to imagine something less popular in America right now. And it's also, I think, hard to imagine another political issue where you see Bernie Sanders having a similar position to Ron DeSantis, the governor of Florida, and Governor Greg Abbott in Texas to Republicans who are on the far right of the ideological spectrum compared to Bernie Sanders. And all three of those politicians have come out in some type of opposition or some kind of delay.So there's this unique bipartisan consensus that's developing. It's not like solar and fossil fuels, for example, where it just splits down political partisan lines.
[Cody Simms] (33:58 - 34:11)
How much of this backlash do you think is really about data centers specifically and how much is it a stand in for broader anxiety about AI and the economy and job loss, et cetera, et cetera?[Michael Thomas] (34:11 - 35:21)
It's definitely a mix of both. If you look at polling on why people oppose data centers, the number one issue that they'll mention is impact on the environment, which actually surprised me because as someone who has written about climate change for a decade, environmental issues don't poll that high. They're among some of the lower issues when you poll people on what's important to them or their priorities, at least.In the case of data centers, people mentioned water first and then the power use, and then they'll mention some other concerns like land use changes. So that, to me, makes me think that a big part of it is the physical infrastructure, but you really can't separate it from a few things. One is the broader anxiety around AI.
I think that's certainly real. But I think the even more important one is the lack of trust that people have in the companies building them. So if you look at what has happened in general with trust in institutions and large corporations that plummeted over the last few decades, but I think Big Tech in particular has a real trust issue with Americans after so many issues and trust breaches.
[Cody Simms] (35:22 - 35:31)
Are we seeing low approval even starker in areas that have actual data centers in their backyard?[Michael Thomas] (35:32 - 36:04)
That's one of the things that I find interesting is you see a lot of moratoriums passed in places where no one's proposing a data center. It's almost like a preemptive moratorium saying, don't ever come here. And in many cases, they're not necessarily sites that you would imagine to be suitable for a data center.So there's an element of that. But I do think that there's more opposition in the states that are seeing a lot more of this because they're important local issues and then politicians are talking about it a lot and it gets a lot of media coverage as well.
[Cody Simms] (36:05 - 36:16)
Let's unpack a few of the specific issues that you highlighted. You said water was the highest one. I actually want to start with affordability, which is are data centers actually driving up people's electricity bills?[Michael Thomas] (36:17 - 36:43)
It depends on the data center and the market. So this is one of these questions that it has a really nuanced answer. I was interviewed on radio recently, where you're given 10 seconds to respond on stuff. And the first question they asked me was, Trump has said that these don't raise electricity bills.What are your thoughts? And I had 10 seconds to respond. And it deserves more than 10 seconds.
So I think what I would say is....
[Cody Simms] (36:44 - 36:44)
Welcome to the podcast format. Go ahead, Michael.[Michael Thomas] (36:45 - 39:54)
Exactly. So I'd like to talk for an hour now about this issue if you don't mind. If you look at where electricity demand has grown the fastest in the last five years, and then you compare that to where electricity prices have risen or fallen, this data comes from an LBNL study. So one of the national labs, really trustworthy, they put out great research, have no agenda.And when they did the analysis, they found that there was actually a negative correlation between rising demand and bills. And so where demand for electricity grew for example, North Dakota, it exploded over the last five years driven by cryptocurrency, data centers, and oil and gas electrification, bills actually fell. And if you adjust for inflation, they fell quite a bit.
If you look at taxes, for example, they fell. And then if you look at states where electricity demand has fallen, states like California and Hawaii, bills have risen. And it actually sort of makes sense if you think about why that would be, because when you bring a lot of electricity from a single user onto the grid, if you don't have to build any new infrastructure, all you do is allocate more of the cost to that customer.
And so you and I get to pay a couple dollars less. AWS building New Carlisle, Indiana's site is going to pay a little bit more. It's sort of this economy is a scale.
And that's how the grid has worked for a hundred years. Where it breaks down is specific markets. So someone listening to this who's really plugged in will think about PJM, which is the largest grid operator in the country. It covers the mid-Atlantic.
I think states like Ohio and out into the East Coast, like New Jersey. And in these states, electricity prices have been rising. And the grid operator has been very clear in saying this is because of data center demand.
But it's also because there hasn't been growth of generation and storage in that market. So if you look at how much battery storage PJM has built in the last decade, it's nothing. I mean, it's less than a gigawatt.
And then you compare that to Texas, which is not the biggest grid in the United States. And Texas has more than 10 times as much battery capacity. And so it is sort of a failure to build and also rising demand.
Then if you look at some states where these data centers are triggering new gas plants, new transmission lines, you're seeing these utilities saying, hey, we're going to raise rates by $5 a month or $10 a month. And that is because of that new load. But in many cases, it's really either a failure to meet that demand.
Or if you don't think that we should even be building that, it's just a failure of rate design, where we just need to allocate the cost to data centers. It's a complicated issue where I'd say in some cases, it's raising bills, in some cases, it's not. And yeah, it doesn't quite lend itself to the sort of simple analysis that I think a lot of people might gravitate towards.
[Cody Simms] (39:55 - 40:00)
And makes it hard as a national political issue, because it's locally nuanced it sounds like.[Michael Thomas] (40:00 - 40:49)
Yeah, I think it's also an opportunity to pass federal policy that makes sure that data centers don't increase people's bills. I think almost everyone agrees that trillion dollar companies should pay for this infrastructure that they're going to make tens of billions of dollars in many cases. In many cases, I think even the tech companies themselves believe this. But I think right now, if you look at the federal policy that we have, it's a sort of handshake agreement, or it's a we'll do our best, which is the Ratepayer Pledge that the Trump administration passed, where they got all the big tech companies in a room and all the AI labs, and then all the utilities and said, hey, everybody, what do you say we do our best to make sure these bills don't go up?And I think what it requires is actual legislation.
[Cody Simms] (40:50 - 41:07)
Let's talk about water. You mentioned water is the number one concern that people are citing. I'm curious how water use also shows up in reality for people who live in the proximity to one of these large projects.[Michael Thomas] (41:08 - 42:17)
So the water issue, I think, is another complicated one where I think if you look just at the macro scale, and you compare data centers today, and then even data centers in five years or 10 years, assuming this astronomical growth, they're a pretty small contributor to overall water use. Certainly, if you compare them to other uses like agriculture, for example, we use so much water, especially in the American West where we have drought, just raising alfalfa to feed the cows, which are an inefficient source of food, right? There's more efficient ways to have protein in a diet.And so that is a much more important thing to focus on, in my opinion, if we were to rank and prioritize. But then I think at the local scale, it does get a little bit more complicated. You have to look at where they're being cited and how they're using water.
So in Arizona, there's not a lot of water. There's just new cuts announced with the Colorado River deal, the federal government, and there's an explosion of data centers going into Phoenix, specifically.
[Cody Simms] (42:17 - 42:27)
So it may be an issue where it's total consumption, not necessarily substantial, but growth of consumption very real and very noticeable to people.[Michael Thomas] (42:27 - 43:35)
I'd maybe tweak that slightly to say where the growth is happening is the really important thing. And I say that as somebody who grew up in Colorado, where we had drought, basically my entire life, and where there is water scarcity. And so yeah, if you look at just like the total water volume, if you compare a data center to say a Rivian factory making EVs in Georgia, you can look at the actual water use of that facility, the Rivian factory.And it's similar to a data center. And not a lot of people have opposed Rivian EV factories. But if you were to cite either a Rivian factory or a data center in an area that has drought, then I think it becomes a more complicated issue where it's not necessarily about the data center.
It's just about, okay, what is the best allocation of scarce resources? Where should we be using water? And you can see where people would be very frustrated if in the case of Denver, Colorado, restaurants aren't giving people water by default at the restaurant.
And then if they hear a data center that's being built, there's some headline that's going to guzzle this many million gallons of water, you can see where it would be frustrating.
[Cody Simms] (43:36 - 43:47)
Back to the power conversation, do we know for most of these projects how much of the water use is for cooling? And how much of the water use is actually for net gas power production?[Michael Thomas] (43:48 - 44:42)
If you look at, again, the global numbers, a lot of it does go to power production, but then you start to get into these debates where people say, well, how should you allocate the water from that? For example, in one of the national labs' study of it, they were including evaporation from hydro dams as a part of water use and all electricity usage. And so power is certainly a large part of it.But that's where I think off-grid or co-located power plant is going to use more water, even if the other thing we didn't talk about was using closed loop systems. So a lot of the new data centers are closed loop, which means that they're not going to use water on an ongoing basis. But if you have an off-site natural gas power plant right next to it, then it is still going to use water.
[Cody Simms] (44:43 - 44:59)
Which makes me, I guess, hopeful that if a lot of these projects eventually get grid connected, that maybe the water side also starts to become less of a problem, as they don't need to use their own on site gas generation as much.[Michael Thomas] (45:00 - 45:46)
Yeah, but I think then you will start to get into the issue beyond all of the facts and numbers, which is just the trust issue, where a lot of these companies are saying, hey, we're closed loop. Like Microsoft has said, all of our new data centers are closed loop systems. So when they go to a community, they should expect no complaints or no opposition on water, but they still get a lot of them.And the reason is because they say, well, I don't trust you. I think that you are going to use water. It's also kind of just an issue in our country now.
We just have a lot of misinformation. So even if Microsoft isn't going to use water for their data center, people might think that they are because somebody is saying that they will, but then they don't trust the company because of this lack of trust over time.
[Cody Simms] (45:46 - 46:18)
Lastly, you mentioned that people are concerned about the environmental footprint of these projects. And I think there's both the local pollution aspect. There's the broader carbon emissions aspect from a climate perspective.And then there's also just their noisy if you live near one as well, particularly if you're running heavy diesel or gas generators all the time. Maybe describe a bit about the impacts there that you're hearing people cite and how that relates to what's actually happening.
[Michael Thomas] (46:19 - 47:09)
We haven't touched on carbon emissions, but maybe touch on that for a second and then noise. I think the emissions impact of this is going to be very large. I think that there's no getting around that at this point.The amount of fossil fuel infrastructure that's getting built out is just very large. And the emissions per project or per plant is increasingly very large. So the Amazon project that we talked about at the beginning of this episode, that single site, if it runs at industry average utilization rates, it could become the largest single source of pollution in America more than any coal plant in terms of its carbon emissions.
And that's a single data center. We should expect emissions from the sector to go up significantly. And then I think there are these more local issues like noise....
[Cody Simms] (47:09 - 47:31)
Which by the way, it is no surprise. Power generation has been the largest, goes back and forth in the United States between power generation and tailpipe emissions, as the single largest source of emissions in our economy. And so as the tech industry moves into power generation for what they do, it should be no surprise that it is becoming a substantial emissions driver for them.[Michael Thomas] (47:31 - 48:09)
Absolutely. I think we had this period in the first part of the 21st century, where a lot of the economic growth in America was coming from really carbon efficient sources of the economy. It was tech and it was pretty efficient data centers that didn't use a lot of power relatively speaking and a services economy.And now the biggest driver of economic growth is AI. And it turns out it's a very energy intensive, carbon intensive activity. And so it isn't surprising that if that's growing at the rates that it is that we're going to see emissions go up.
[Cody Simms] (48:09 - 48:15)
Hit the noise side of things and then I do want to get your view on how this evolves over the next five or 10 years to kind of close us out.[Michael Thomas] (48:16 - 49:53)
I think noise, it's certainly an issue if you're building a power plant right next to your data center. I think that's the one I can speak to most since it's the area that I've kind of looked at the most. And if you look at the second off-grid data center built, Vantage VA2, it was built about a year after Colossus came online in Virginia.If you look at satellite images of this, you can just see this, these rows of houses right next to it. And then you see seven gas turbines right there. And these gas turbines are often what are called aeroderivative turbines, which means that they're quite literally a jet engine that's strapped to the back of a semi truck.
Aero is an airplane design that's been repurposed. And so if you just think about how someone might feel if seven jet engines were parked outside of their neighborhood and then run 24-7, and then people start to scratch their head and wonder why are people opposed to these projects. I don't get why our polling has gone down.
Now, of course, that's one community, but then there are all these viral videos and news articles where people are putting up videos of this site in particular where they're putting mattresses up in their window to try to block the noise. Because again, there's seven airplanes parked down their street. And so I think there's good and bad ways to develop data centers.
And I think anyone who wants to see AI grow should really be thinking about what does a good development look like. And I would say building a power plant in a neighborhood is probably not going to be the best fit.
[Cody Simms] (49:54 - 50:53)
Let's talk about where this all ends up. And obviously, none of us know where it's going to end up. But on the optimistic side, you also see these companies entering into agreements with small modular nuclear developers, with fusion companies, with other forms of breakthrough clean energy technologies that may not have had a market if not for this explosive growth, or at least didn't have the big dollars being thrown at them for R&D and development that they are now seeing.Do you feel optimistic that we're in a short term gas explosion that will go away? Or are you more of the mindset that we've never been very good at retiring sources of energy that we bring online? And so this is probably a permanent growth back to fossil fuel economy.
[Michael Thomas] (50:54 - 53:44)
I'm not sure exactly where I land. I think I probably land in different places every day. I would say, I understand the argument people make where they say, look at a project like Fervo Energy's enhanced geothermal project in Utah.They just signed a deal with Google. Google said, we're willing to pay a premium. We want to help accelerate this technology.
We want power to power our data centers. And so that is pulling forward this technology that could help decarbonize the rest of the West, if it can get down the learning curve. At the same time, I'm looking at what's happening with power prices or solar and wind prices.
And if the argument is, this demand can help bring these technologies down the cost curve and then that will benefit the rest of the economy and the rest of the sectors we need to decarbonize. Well, right now, the cost of solar and wind is going up because there's more demand than supply. And so that's not falling down the cost curve.
It's kind of making it even harder for utilities in parts of the country that don't have data center demand, make the justification that they should build solar and not a gas plant to meet, say, EV demand or home electrification demand. In general, the trajectory that it seems like we're on is one of more energy on top of the current system. So we were on this path to decarbonize before the data center boom.
And we were seeing coal retirements accelerate. And we were seeing a decline in natural gas generation we were building. And now we're seeing coal retirements get delayed, which is more emissions today.
And then a lot of gas plants. And so the trajectory right now is we're going to see a lot more gas electricity generation and a much slower phase out of coal. And so we're going to see a lot more emissions in the short term, which is a really important decade if you think about where we are in a warming path and climate impacts.
And I think without federal policy, I don't see a case where we don't just see this growth of emissions. I think if we were to have a policy where there was some targets or some kind of emissions standards, then I think that we could see emissions fall. But absent that, I think that we're just going to see emissions rise in the short term and have to ask the question as a country, do we feel like the benefits of AI are worth the environmental costs of data centers?
And everyone will have a different answer to that. It's not a lot different than the same question we've been asking around fossil fuels, where fossil fuels improve our lives, make us richer, provide a lot of benefits. And they also have an externality and a negative cost.
[Cody Simms] (53:44 - 54:25)
I would say the possibility of getting federal legislation on emissions around data centers right now is probably at about zero, if not lower than zero. You say, hey, we as public have to ask ourselves if the benefits of AI outweigh the potential rise in emissions. But that's back to the individual empowerment feeling that these data centers are basically just happening and the public doesn't necessarily have a lot of room to weigh in on that conversation.Where do people have empowerment to make that type of idea play out?
[Michael Thomas] (54:26 - 56:39)
I would have had a different answer to this question a few years ago. If we were talking about where can someone get involved to decarbonize the economy on an issue like this, it feels a bit naive for me to say, just show up to your city council or state legislature, advocate for good policy. I think the dynamic that we have right now is one where everyone, when we do start talking about good development policies, starts to get into this almost cold war dynamic where we talk about China and we say, even if you just propose the idea, assume you don't have the current presidency, you say, let's just do an emissions standard and say, what a great way to pull forward this investment and clean firm like advanced geothermal and nuclear and long duration storage, then to just force the AI companies, if you want to see that growth, recognize that revenue, then pay the green premium on these technologies and we don't even have to have the public pay for it in the form of federal funding. I think in a naive way, that seems like a great policy, but does it have a chance of passing this administration? No, so that's two years out.Then even in a future one, I think you could imagine people bringing up the China debate and saying, well, we're going to hamstring ourselves. Even if China agrees to solar and wind and batteries and advanced nuclear, well, they're going to be able to build way more of that than we are and so we're just going to give up the AI race. It does feel like we're in a bit of this trap and I've found myself struggling to understand it and struggling to think about it for myself and when people ask, what should we do?
I think my position right now is where I want to play a role is I just want to publish information about these projects, help people have the information to have that debate and maybe that's naive and I'm living in an Obama world where there's less information and more nuance and it's where I've landed right now is I want to provide the objective information and the tools that people can have that debate and conversation.
[Cody Simms] (56:40 - 57:21)
Michael, you're doing great work to do that. As someone who spends a lot of time thinking about these issues myself and I'm sure many of our listeners do as well, the information that you're putting out there almost every time you publish it still feels very novel and feels like you've uncovered a new way of thinking about what's happening because you're exposing information that while it was out there in public, it wasn't pulled together for people to understand.So I want to express my gratitude and thanks and I don't think any of us have the answers or the crystal ball to know what's going to happen but the more visibility we have into what is happening can help inform the future as well.
[Michael Thomas] (57:22 - 57:36)
Well I appreciate you saying that and I appreciate you having me on today to talk about all this. It's a big topic and there's no shortage of projects to cover or complex issues like the ones that we've talked about and so I appreciate you reading some of the work.[Cody Simms] (57:36 - 57:44)
Thanks for taking the time from your satellite surveillance and public records sleuthing to join here and share a little bit about what you've been learning.[Michael Thomas] (57:45 - 57:46)
Thank you so much for having me. It's been fun.[Cody Simms] (57:47 - 58:00)
Inevitable is an MCJ podcast. Learn more at MCJ.VC and subscribe to our newsletter at newsletter.MCJ.VC. See you next week!