Crusoe’s Big Bet on AI Infrastructure and Energy
Cully Cavness is the Co-founder, President, and COO of Crusoe, an energy-first AI infrastructure company.
In this live episode recorded in Austin, Texas, Cully shares how Crusoe evolved from capturing flared gas for Bitcoin mining to becoming a leading developer of hyperscale data centers. He discusses the company’s pivotal role in Project Stargate—a $500B AI infrastructure effort led by OpenAI, SoftBank, and Oracle—and how Crusoe is building a 1.2 gigawatt data center campus in Abilene, Texas. Cully reflects on the decision to divest its original Bitcoin business, the company’s vertical integration strategy, and how energy abundance will shape the future of AI.
Thanks to our sponsors for this event: Gunderson Dettmer, and J.P. Morgan.
Episode recorded on April 8, 2025 (Published on April 24, 2025)
In this episode, we cover:
[00:24] An overview of Crusoe
[01:08] Its role in Project Stargate and Abilene data center
[03:41] Shift from outbound to inbound interest
[06:17] Company pivots and existential startup bets
[09:09] Sale of Bitcoin mining business to NYDIG
[11:40] Flared gas capture and climate impact overview
[14:57] From digital flare mitigation to stranded wind use
[17:27] Cully’s personal energy background and worldview
[22:14] Why AI could drive climate and fusion breakthroughs
[25:47] Details of the 1.2 GW Abilene campus for Oracle
[36:42] 3,500 skilled trades supporting data center build
[44:42] Natural gas as a bridge fuel + CCS investments
-
Cody Simms (00:00):
Today on Inevitable, we are live in Austin, Texas, in front of an amazing audience at the Sunset Room for a conversation with Cully Cavness at Crusoe. Cully is the co-founder, president, and COO of the company which started in 2018. Crusoe is all over the headlines these days as it is the anchor developer of a 1.2 gigawatt... I feel like I need to channel Doc Brown and say 1.21 gigawatts. But 1.2 gigawatt data center facility being built in Abilene, Texas. That is the first known activity of Project Stargate, a multinational artificial intelligence joint venture created by OpenAI, SoftBank, and Oracle that plans to invest up to $500 billion in AI infrastructure in the USA by 2029. A project that would rival the US interstate highway system and China's Belt and Road Initiative among the largest infrastructure builds in human history.
(01:08):
Crusoe recently announced that it has secured 15 billion in a joint venture, including Blue Owl, to develop the Abilene campus, which will house 400,000 NVIDIA GPU chips. Crusoe also operates a substantial cloud AI offering, providing GPU infrastructure to companies wanting to easily run and scale machine learning models. MCJ is proud to be a three-time investor in Crusoe via our venture capital funds. Thank you, Cully. And we're thrilled to be having this conversation here in Austin, which has a booming startup ecosystem in energy and climate tech. And finally, thank you to our partners at Gunderson and JP Morgan Chase for their support with today's event. This is where our intro music will play.
(01:51):
From MCJ, I'm Cody Simms, and this is Inevitable. Climate change is Inevitable. It's already here, but so are the solutions shaping our future. Join us every week to learn from experts and entrepreneurs about the transition of energy and industry.
(02:18):
Cully, welcome to the show.
Cully Cavness (02:19):
Thanks for having me. Excited to be here.
Cody Simms (02:21):
So how do you describe Crusoe?
Cully Cavness (02:24):
Crusoe has definitely gone through a few iterations over the years. We can talk about the history. But we are a vertically integrated AI infrastructure company and we take an energy-first approach to building that AI infrastructure. So, fundamentally, that means assembling land and power, building the data center infrastructure itself, putting in the GPUs, the networking, all the IT hardware. On top of that, the software layer and the productization of that compute power. We can operate on a couple of different levels of that stack. We don't necessarily have to do all of that together in every single transaction. But we have the capability to do really everything in there. We can also talk a little bit about our manufacturing capabilities where we've actually vertically integrated a lot of the manufacturing that goes into that
Cody Simms (03:07):
And the name comes from Robinson Crusoe, the book?
Cully Cavness (03:10):
Yes. The most legendary story of being resourceful and creative with scarce resources. And when we started the business, it was about capturing flared natural gas, wasted resources, and doing something creative with those, something innovative with those. Robinson Crusoe was doing innovative things with what he had at his disposal.
Cody Simms (03:29):
Hopefully, minus the cannibals part.
Cully Cavness (03:31):
That's right. No cannibals so far.
Cody Simms (03:33):
I can only imagine, you're busy right now than I've ever been in my life, what does a day in your life look like right now running a business at rocket ship scale?
Cully Cavness (03:41):
It's different every day. I would say there's no standard day. It starts pretty early. And then I'm working on my calendar management, but I'll be honest with you, it's a lot of 15 and 30 minute meetings for very quick decisions on a whole range of different topics from financings and strategy things, to tactical, operational things, personnel issues, recruiting, all that startup stuff.
Cody Simms (04:05):
Well, you've really hit, I think, an inflection point I would say in the last... From what I've observed, in about the last six months where you've gone from a company that has to knock on other people's doors to one where everyone's coming to knock on your door. And I feel this because I get asked multiple times a day from other startups, can I get an intro to Crusoe you? Whether it's a company wanting to talk about how they do water resource management or a company that talks about how they do carbon capture, point source carbon capture, or this, that, and the other. How has that felt as a founder seeing your company get to that point where everyone else is now wanting to talk to you?
Cully Cavness (04:46):
That's a really special moment. You spend so much time as a founder hoping that people will take you seriously and not sort of laugh you out of the room. And I certainly remember when we were starting the business and we thought, okay, we're going to go pitch oil and gas companies that we will capture their flared gas and we're going to mine Bitcoin with it. This is back in 2018 when it was embarrassing to talk about Bitcoin mining in proper business conversations, right? It was not a widely accepted industry at the time. And we would try to almost camouflage that as we'll do computing or it'll be a data center but not say the Bitcoin word. Hoping we could just get-
Cody Simms (05:24):
High capacity compute. You guys use that phrase a lot.
Cully Cavness (05:27):
And it's gone from there, obviously, through all these iterations where we built one of the biggest Bitcoin mining businesses in the world in an off-grid capacity using flared natural gas and wasted natural gas. That led us to deploying GPU compute in the field. And then ultimately, in other data centers. Which then brought us to the opportunity of realizing we actually need to build our own data centers and we need to build them at scale. And that there's actually a large demand from third parties for those data centers. And now people are actually coming to us asking, how did you do it this quickly? Can you help me do it? Can we do it together? It is a really kind of special transition to just step back and look at that. We had an all-hands meeting just yesterday. Crusoe turned seven years old this month. And Chase, my co-founder, put together a slide where we had some of our earliest text messages and photos of the deployments and it's just...
(06:17):
Seven years doesn't sound like a long time, but I feel like this experience has aged me 50 years. It's just so intense. And just to kind of look back at those early days was pretty special. And these text messages of like, oh, I got it, we should name it Crusoe. That would be a lot better than Direct Gen LLC or Crypto Gen LLC. These were the other candidates for the name. Funny to look back at how much we've learned since then.
Cody Simms (06:44):
I love that you did name it after... You mentioned Robinson Crusoe being resourceful. Because you just talked about these three very sort of distinct phases of the business that you've built. And to some extent, I'm sure you had the foresight to see the world was going to evolve in some of these ways, but you also were very good at seeing the wave that was about to crest and knowing when the right time was to jump on said wave. Maybe describe how the company has evolved and how you navigated those different points in the business along the way.
Cully Cavness (07:18):
Those are the scariest points for sure because almost every time they required us to make an existential bet on something that we thought we could do, but we weren't 100% sure we could pull it off or exactly how it was going to work. It wasn't the kind of thing you could put in a nice pretty DCF financial model and explain clearly to an investor at the time. It was sometimes more of an intuitive vision for the business. I'll give Chase a lot of credit. I mean, he had early instincts that GPUs for machine learning would be a second market for us. And so we started way before anybody ever heard of ChatGPT. We were deploying clusters of GPUs in the oil fields powered by these flared gas resources. And there were significant CapEx investments for what was then a series A, series B kind of business.
Cody Simms (08:06):
The Bitcoin mining is totally different hardware, right? That's ASICs, it's a completely different computer.
Cully Cavness (08:10):
Totally different hardware, totally different set of requirements around networking, and reliability, engineering. Just even the air, the purity, the cleanliness of the environment. All that had to be engineered anew and then deployed and actually physically built. This is a very capital-intensive business. I can remember when we went from rented generators to owning and operating our own power generation assets. And then when we decided, okay, we should get into the data center construction business, these require us to take huge capital risks and then really burn the ships and just force the team to execute. And we have got to get this right because we've taken such a big bet on this. It can't go any other direction. I don't think every business goes through those kinds of dramatic pivots necessarily, but that's been our experience. We've had some really big chapter changes in the story.
Cody Simms (09:00):
You have a really big one that you sort of announced a few months ago, but it sounds like there may be some kind of official news about it around the Bitcoin business itself.
Cully Cavness (09:09):
Yeah, we announced a few months ago that we had entered into an agreement to sell the Bitcoin mining business. And about three hours ago, we finally closed the sale of our Bitcoin mining business. So the entire digital flare mitigation business, all of our Bitcoin mining assets and operations have been purchased by NYDIG, New York Digital Infrastructure Group. Awesome team, awesome organization. And they're going to invest in that, and grow it, and continue the vision in a format that's very focused on Bitcoin investing. And what that does for us is it allows us to focus on vertical integrated AI infrastructure.
Cody Simms (09:43):
Let me just ask the sort of obvious question there. Bitcoin is around its all-time high. It feels like it's having its own kind of moment at the time that you guys decided strategically it was a distraction from a bigger opportunity that you're going after. That had to be a hard decision to make I would think.
Cully Cavness (09:59):
It was very hard. I mean, first of all, these are... 135 of our team members are going over in the transaction. And many of these are our earliest, most dedicated long-term team members. So on an interpersonal level and personal relationship level, really hard. I mean, we have somebody who has the Crusoe logo tattooed on his shoulder that's going in this transaction.
Cody Simms (10:19):
Please tell me they're keeping... No, they're not keeping the name, are they?
Cully Cavness (10:23):
I offered to pay for tattoo removal and he was like, we'll, see how you treat me in the process. We'll make that decision later. Respected that. What it does for us though is it lets us focus on what we see as the vision for the future of the company, which is this AI-focused infrastructure. And it was getting to the point where we had a tremendous internal competition for capital. I mean, Bitcoin mining, it's ASIC hardware, it's data centers, it's copper wires and steel, and it's power generation systems. It's massively capital-intensive. Cloud business buying GPUs, building or renting data centers, and then building our own hyperscale data center campuses, this is measured in tens of billions of dollars of CapEx. And so there was a team getting pulled in many directions element, and then there was just simply a we need to sort of pick a direction and focus on that with our financial resources as well.
Cody Simms (11:14):
And with that transaction, just so I understand, you've built this whole digital flare mitigation technology, which maybe you can explain a little bit for folks who aren't familiar with it, that was the calling card of the company, right? That was the thing that made Crusoe unique, that helped you leverage stranded energy. And I think today is the bulk of your energy management platform today as of more than three hours ago, but most of that is leaving with the Bitcoin transaction, is that right?
Cully Cavness (11:40):
Yeah. I mean, the original logo of the company was a flare in a circle representing capturing this flare. So digital flare mitigation is a business model where when an oil well is drilled in an area where there isn't gas pipeline capacity available or it's full or it just doesn't even exist, historically, the oil industry's standard was to just light the gas on fire as a flare. It's a big waste of energy. It doesn't fully combust the methane. Estimates range, but on the order of seven, eight, 9% of the methane will escape to the atmosphere. It has a global warming potential of 84 compared... So that's 84 times as potent as CO2. And we would deploy power generation systems. Basically reciprocating engines or turbines. And we would capture the gas and route it over to those systems as fuel, generate our own electricity on site right next to the pump jacks, and then we would wire the electricity into our modular data centers which were loaded with Bitcoin mining computers mostly. But then eventually, we got into GPUs as well.
(12:41):
And we would monetize that with a satellite internet connection. So we could sort of deploy this anywhere, even in the most remote, rugged oil field locations. When we turn on the generator, the flare would shrink. We get 99.9% combustion efficiency in our generators. We're avoiding waste. Basically removing abating methane emissions. And then achieving a pretty competitive power price that we could use for Bitcoin mining or other computing. So that was the digital flare mitigation business. We built that to about 250 megawatts, which was sold to NYDIG as part of this transaction. We ended up eventually getting into some renewable-powered projects that are part of that portfolio as well. And yeah, across that operation, we operated more than 110 power generation systems. Hundreds of megawatts of power generation we operate ourselves across seven states. And we had a partnership in Argentina, the Vaca Muerta oil field. So, in total, we abated about 39 billion cubic feet of gas in total in the history of the operation. It's the equivalent of taking many hundreds of thousands of cars off the road.
(13:49):
We do a scope 1 through four carbon accounting on that business for the last three years, which showed that there's the CO2 emissions that we produce from combusting the gas, last year, that was about 1/2 a million tons. And then we, by virtue of the avoided methane emissions, abated about 750,000 tons of CO2 equivalent. And then there's another about 500,000 tons of avoided grid emissions if those computers had been deployed to the grid by some third party. So we sort of add that all up. If you take scope 4 into the equation, it's a pretty deeply carbon-negative operation compared to the status quo of just persistent flaring. That was the original business model. Happy to talk about all the chapters since then, but I think it's good to just go back and understand where it all came from.
Cody Simms (14:33):
When you started the company, it seems like the idea was how do you monetize stranded energy and try to avoid emissions as part of that. It feels like that was sort of the origin idea of Crusoe. Maybe you can elaborate on that or say that in a better way than I just did. And I'm curious how you view where the company is going as being aligned with that same mission if it still is.
Cully Cavness (14:57):
Energy is really the core of how we think about all this computing infrastructure. You see it more and more in the press these days, but it is the main constraint for the advancement of AI. We started with a strategy around monetizing stranded energy, and that was flared gas originally. Then we started thinking about renewables. The insight is basically that there's also stranded wind. I mean, in Texas, it's sort of famous that in west Texas there are congested parts of the grid where you have effectively excess generation that's been built chasing the production tax credit, and power prices can go negative. But by virtue of the tax credit, those producers can actually get a net positive cost per kilowatt-hour for their power that they're producing. In some cases, when those windmills age, they get to the point where they no longer qualify for that tax credit. And so the rest of the market is forcing the price negative, they can no longer produce at a negative price.
(15:50):
And so in that case, you can see a portion of the fleet is actually aging towards either an unnecessary or uneconomic repower investment or curtailment and significant amounts of curtailment. There are some nodes where you have up to 20-plus percent of the hours of the year, you'll see negative pricing during the hour. And the curtailment process isn't start and stop. So if you have negative pricing within the hour, you're going to have a significant amount of curtailment around that negative pricing moment. And so our insight was basically this is another place where putting load, putting, in this case, baseload data center demand could potentially put in a better pricing mechanism for the producers, reduce curtailment. We'd have attractive power producing. They would have less risk that of a price floor depending on the structure of the deal. So we started pursuing that.
Cody Simms (16:38):
It's still not baseload though, right, it's wind, it's intermittent.
Cully Cavness (16:40):
It's intermittent. But for us, for the data center, the requirement is baseload. I guess I'm kind of skipping ahead a couple of chapters. We started with a Bitcoin mining approach, which could have really followed the load profile. That's what led us to-
Cody Simms (16:53):
Oh, because you could turn it on or off as you needed.
Cully Cavness (16:55):
Exactly. If you want to, we can talk about Abilene. This is sort of what led us to Abilene. And that might be a good opportunity to talk about what we see as the future of the energy mix for AI data centers and where this can all be going.
Cody Simms (17:08):
Before we do that, I want to understand more your background and relationship with energy. Clearly, you think about this a lot, you've spent a lot of time navigating it. You've worked in a few different forms of energy. Maybe share a little bit about your background and how you ended up building this company in the first place.
Cully Cavness (17:27):
I would say it's been a long and passionate love affair with a lot of strife along the way, my relationship with energy. I was born in a family of oil and gas entrepreneurs. My dad, my grandfather. And I kind of grew up around the dinner table hearing about supplying our nation's vital energy needs. That was my dad's turn of phrase that he always said at dinnertime.
Cody Simms (17:50):
Energy abundance.
Cully Cavness (17:51):
That's right. And so I was like, okay, I admire my dad, I admire this whole industry, I grew up around a lot of the characters in the industry and understanding how it worked. And then I went to Middlebury College and it was a very different chapter of my relationship with energy.
Cody Simms (18:05):
That's the home of Bill McKibben.
Cully Cavness (18:06):
Bill McKibben was getting arrested at climate protests in major cities around the world. We had An Inconvenient Truth, came out my freshman year. The whole conversation in the student body was very climate-focused. And I ended up feeling like I'm going to have a crisis of conscience if I sort of take this path into oil and gas without really thinking about this climate issue. So I applied for a program called the Thomas Watson Fellowship, which is an academic fellowship that was funded by the founder of IBM. It's for 40 students a year from around the world who want to pursue some topic that they're personally passionate about, but there's no official program for them to learn about this. And so, for me, it was basically the morality of energy. And I wanted to spend a year thinking about that. And I was lucky enough to receive this fellowship. And I got to basically travel around the world interviewing and interning for power project developers and financiers.
(19:03):
So I ended up working with the Icelandic Geo Survey around geothermal power. I worked with the CFO of one of the largest renewables developers in Spain. I worked with a hydroelectric engineer in Argentina. And I spent five months in China working with, at one point, a family that was a coal importer from Mongolia. And then also a company that had licensed American waste heat recovery technology.
Cody Simms (19:27):
So you had a lot of different perspectives and inputs on how you think about these things.
Cully Cavness (19:31):
It really changed. I mean, you go from Iceland where you have 300,000 people, 100% geothermal and hydropower. It seems easy. It seems like we should be able to make this work. And then you go to China, more than 1 billion people. This was in 2011, sort of the peak of.. Just the air quality was brutal.
Cody Simms (19:49):
Brutal.
Cully Cavness (19:50):
There was massive amounts of coal fire being just commissioned every single week. They were turning on multiple large-scale coal power plants. But honestly, my experience with the people was... That was a good trade.
Cody Simms (20:03):
It was pulling them out of poverty.
Cully Cavness (20:04):
They were like adding protein to my diet. I'm upgrading from a bicycle to a scooter, or a scooter to a small car. I'm getting air conditioning for the first time. And it led me to, I think, kind of come a full circle back to what I would call a pragmatic view of environmentalism, which is human nature is a really important factor in all this. People crave prosperity, they want to advance in their lives, and that drives consumption. And energy is the basis of a lot of this prosperity. There's this chart that relates to the Human Development Index with energy consumption. Probably a lot of people here have looked at this. I studied that. And it really seeded in my memory just this relationship between energy per capita and Human Development Index, which is your longevity, your education, and your prosperity.
(20:52):
And so, for me, I came back to we need economic forms of environmental solutions because that's the only way these are going to scale. So happy to talk more about that. I ended up writing a 40-plus page piece on Medium that's called Talley's Law, which is sort of my thoughts of what this means in an equation form.
Cody Simms (21:10):
I totally missed that in getting ready for this conversation.
Cully Cavness (21:13):
Basically, the gist of it is technology and natural resources on one side of the equation, and population, quality of life, and environmental health on the other. So if we want to have a rising population, rising quality of life without sacrificing environmental health in a world of fixed or depleting resources, technology is the one variable that we can control to lift up the left side of the ledger to balance that equation. That leads you to something like digital flare mitigation, which is, okay, we've got resources that are being wasted, a new technological approach could actually unlock more from those resources.
Cody Simms (21:48):
Let's then apply that theory to the world we're entering now with AI. I would say that the counterargument to everything you just described that people might make with AI is to say, growing the energy pie has helped humanity up to this point, but we're about to take that energy pie and just give it to the computers, we're not going to use it for ourselves. Do you totally disagree with that?
Cully Cavness (22:14):
I totally disagree with that. Yeah. I'm really excited about what I think the next few years, five years is going to look like as we really see what this technology's capable of. I think at Crusoe, we're seeing a lot of just incredible examples of what could be coming here. And from an energy perspective, my view is if we're going to get there on fusion, it's probably because AI is going to help us figure it out. If we're going to get there on smart grids really and be able to integrate a lot more intermittent renewables and batteries, it's probably going to be because AI plays a role. Better battery materials, better carbon capture materials, breakthroughs with SMRs. Some portfolio companies and some companies in the room are using AI to innovate on some of the things I just named. So, I think not only are we just going to give all the energy to the machines, we actually have to make this investment to get to some of these breakthrough technologies or else we're going to be on... Right now, with the glide path, it doesn't get you to 1 1/2 degrees, it doesn't get you to two degrees. It might not even get you to three or four degrees if you don't have any disruptive breakthroughs in technology. They've actually become economic.
Cody Simms (23:22):
So you don't think we're in the mode of just deploy the solar technology, the battery storage technology we already have is going to get us there.
Cully Cavness (23:28):
I mean, I just think you can look at the raw data and see that it's not working. We're consuming more hydrocarbons every day than before. The rate of adding CO2 to the atmosphere is increasing year over year, not slowing down. These technologies are scaling relative to their own histories, but not relative total consumption. We're not in an energy transition, we're in an energy addition. And the gap is largely being made up by traditional sources. Which back to the person who wants to add protein to their diet and upgrade from a bicycle to a motor scooter, which is where a lot of this increased emissions and demand is coming from, I don't see who's going to argue with that. There's no political or moral argument you can make against that. And so it needs to be a technological breakthrough in my opinion. And I think AI is going to be the thing that accelerates basic science and technology innovation. It lets us get there faster.
Cody Simms (24:20):
I sure hope so. With the internet, the negative argument of the internet would be like we could have had all these wonderful things and instead, we got cat memes. I don't agree with that. I think we have-
Cully Cavness (24:30):
I think we got a lot else besides cat memes.
Cody Simms (24:32):
We did. We got a lot. And so the hope is that AI-
Cully Cavness (24:34):
We got this podcast.
Cody Simms (24:35):
There we go. Thank you.
Yin Lu (24:38):
Hey, everyone, I'm Yin, a partner at MCJ here to take a quick minute to tell you about the MCJ collective membership. Globally, startups are rewriting industries to be cleaner, more profitable and more secure. And at MCJ, we recognize that a rapidly changing business landscape requires a workforce that can adapt. MCJ Collective is a vetted member network for tech and industry leaders who are building, working for, or advising on solutions that can address the transition of energy and industry. MCJ Collective connects members with one another with MCJ's portfolio and our broader network. We do this through a powerful member hub, timely introductions, curated events, and a unique talent matchmaking system and opportunities to learn from peers and podcast guests. We started in 2019 and have roamed thousands of members globally. If you want to learn more, head over to mcj.vc and click the membership tab at the top. Thanks, and enjoy the rest of the show.
Cody Simms (25:39):
So let's maybe describe with the Abilene project, what's been announced, kind of walk us through what you're building in Abilene.
Cully Cavness (25:47):
So, our customer is Oracle, and their offtake of their computing that they're installing is OpenAI. So this is really a premier AI project on a global basis. We're building a 1.2 gigawatt data center campus consisting of eight buildings. And we're also building an on-site power plant. It's a 350 megawatt natural gas-fired power plant that will start as the bridge power source to get the site up and running. And then as the grid expands, it'll become the backup generation resource in lieu of diesel generators. And this is located in Abilene. So you're in one of these nodes of the grid that is really saturated in wind and solar and does experience this type of negative power pricing. So putting this load into that part of the grid, it helps to stabilize prices for some of these other renewables producers. There's an on-site solar development opportunity, up to 500 megawatts. And then there's basically a site that's surrounded by wind farms or VPPA contracting.
Cody Simms (26:47):
And there's two phases to the project in terms of kind of phase one and then building a much larger footprint beyond that.
Cully Cavness (26:53):
Phase one was two buildings. Then phase two we took it up to the eight buildings, which when you add in the ancillary loads is what gets us to the 1.2 gigawatts.
Cody Simms (27:03):
And what's the timing on each of these roughly?
Cully Cavness (27:05):
So phase one is the buildings are largely complete. We've really started the interior build-out, and in some cases, even the operations and some of the portions of these buildings. They're huge buildings, they're 1/2 a million square feet. And then phase two, buildings three and four already have structural steel up. The roofs are going on and the remaining buildings are in different phases of foundation construction. So we'll be basically energizing the first 206 megawatts, the first two buildings, this year. From breaking ground last summer in June to energizing was about 300 days for the first building. We're going to get it under 300 days for buildings three through six. The previous record for building a hyperscale data center was about two years, meaning more than a 100-megawatt scale data center. So we think we have broken the record by about 50%.
Cody Simms (27:57):
For people who don't natively speak megawatts and gigawatts, can you give some sense of scale of that?
Cully Cavness (28:03):
More power than Denver uses by about 20%.
Cody Simms (28:07):
Wow. The 1.2.
Cully Cavness (28:08):
Yeah.
Cody Simms (28:09):
How in the world did this come together?
Cully Cavness (28:11):
We were looking for sites that had this characteristic of curtailed wind or negatively priced wind. And we found this company, Lancium. My co-founder, Chase, built a relationship with the founder of Lancium. They were the landowner of this site and they had taken some risk early on and built a 200-megawatt substation. And they had a vision that it could be Bitcoin mining, it could be data centers, it could be batteries, it could be renewables development, but that this part of the grid with these characteristics was a good place to start developing a land and power strategy.
Cody Simms (28:40):
It was a substation, but otherwise, it was just trees and land, right? There was not an existing wind farm there.
Cully Cavness (28:47):
Undeveloped land with a substation. To Lancium's credit, they had had this insight that this piece of electrical infrastructure in this place could be pretty useful. And we originally started talking with them about our Bitcoin mining strategy. And then as the AI wave started picking up momentum, we started thinking maybe we could actually build a data center here for our own GPUs. And then maybe we could actually build a data center here for our own GPUs plus a third party that might want to have a large data center campus. And that's when we started speaking with some of the biggest tech companies that wanted these 100-megawatt-plus scale data centers and they wanted them very quickly to be able to ramp their AI capacity. We were in conversations with a couple of different companies. Ultimately, Oracle came through as the group that took this lease. Again, it was first for the 206 megawatts. Which was at the time, a huge deal for us. This is like a multi-billion dollar contract and we were extremely excited to be underway on that. And they very quickly wanted to upsize it, how we got to the eight buildings.
Cody Simms (29:47):
You hadn't built a major AI data center yet at this point.
Cully Cavness (29:52):
So what we started doing is building the team's capability with people that have built these types of data centers and this type of infrastructure. We had a lot of experience around generating our own power, right? We'd been generating hundreds of megawatts of power from these distributed sources for the past six years of the company's history. We had been building data centers in modular capacities throughout the country and operating those data centers. And what we did is we recruited in a team led by Chris Dolan and Jamie McGrath that were previously co-founders of 365 Main. It was a data center developer in the 2000s. They built up a portfolio of data centers and ultimately, sold to Digital Realty. That team was able to kind of get the band back together and form what we call today our digital infrastructure group inside of Crusoe. Which is today approaching 100 employees that have deep experience building large-scale data center infrastructure.
Cody Simms (30:46):
Is there something unique about building a data center with GPUs for AI compared to just the data center business that's been around for decades now such that you can build it out in rural Texas?
Cully Cavness (31:00):
There's a big difference. I mean, historically, data centers were sited around population centers and latency constraints. The densities were sort of like five kilowatts to 15 kilowatts in a rack. This is sort of a space that you could fit inside of as a person. That's roughly the power density of what a traditional storage-type cloud data center would look like. Today we're north of 100 kilowatts per rack on the latest and greatest from Nvidia. And there's a line of sight to 600 kilowatts and beyond and some of the things that they've announced. So the power density has become absolutely incredible. And that's also led to changes in the way we're going to cool this data center. You can no longer get enough cold air through there to actually cool the data center, so it has to be liquid-cooled. So what we did is designed a data center that was compact. We could get all the GPUs close together to each other. That's a networking constraint that you need to have. And then we have a closed loop water cooling system where we actually bring chilled water into the building.
(32:00):
Each chip gets a little delivery of cold water, and then warm water goes back up to a pipe system that goes outside to a pool fleet of air chillers. And it's a non-consumptive cooling design. So we don't evaporate or leak any water in the system, it's totally a closed loop.
Cody Simms (32:16):
The neighbors to the data center will have access to use their showers and all of that.
Cully Cavness (32:20):
That's right.
Cody Simms (32:20):
You're not stealing water from-
Cully Cavness (32:22):
We shut it up once and it should stay... I mean there'll be probably some top-off and maintenance-type events. But in terms of the normal course of business, it's not consuming water.
Cody Simms (32:29):
I'm hearing two big things that are unique that you all brought as unique capabilities to this. Help me make sure I'm following this correctly. One is as a team, you're rooted in energy first and power first. So you started with the energy requirement. You didn't start with, hey, I got a good deal on some land. It wasn't a real estate deal, it was a power deal. There was this substation that was already built out. You knew there was probably access to curtailed wind there that you could use. And you could build with that sort of expertise in mind. And then the other was essentially the architecture of the data center itself and how you're dealing with this highly power-dense racking setup that requires sort of a unique, essentially, plumbing to it from a cooling perspective. Are those the biggies?
Cully Cavness (33:14):
I think that's right. I mean, speed is really the essential competitive dynamic going on in the hyperscale data center industry.
Cody Simms (33:21):
Time to power. Getting plugged in.
Cully Cavness (33:22):
Time to power. And so being able to say, hey, we actually are going to build our own power plant here in addition to we've identified some available power on the grid, and then we'll build out an expansion of the substation. Take on all that complexity in addition to building a new data center design that was informed by our experience of operating the GPU clusters. I think putting all that together, the team around that, and then the financing structure around that is just a lot to package up.
Cody Simms (33:49):
Because I'm sitting here wondering why haven't these companies that have been building data centers for 15, 20 years... I'm sure they saw this company they probably hadn't really even heard of land this monster deal, and they were like, who are these guys?
Cully Cavness (34:01):
I would say that traditionally, you could just call up your utility and say, hey, I need 25 megawatts for my cloud data center. 50 megawatts, even 100 megawatts. The grid could sort of respond to these increments. And it fit with the relatively shallow growth curve that you see in the utility market for a long time historically. When you start calling up and asking for a gigawatt and you want it next year, the supply chain doesn't react-
Cody Simms (34:24):
You get Doc Brown on the other side of the line, right?
Cully Cavness (34:26):
Yeah, exactly. So it's required sort of a different approach to how you're going to do that.
Cody Simms (34:30):
Another thing that I understand is unique about what you guys are doing is because you've built all these vertically integrated digital flare mitigation Bitcoin mines that you built, as you said in a very wide distributed manner, you also got good at manufacturing and sort of building componentry quickly, so you're not relying on some incredibly complex supply chain. Can you share a little bit about what you all do on the manufacturing side?
Cully Cavness (34:55):
Yeah. In 2022, we acquired an electrical mechanical manufacturing business called Easter-Owens outside of Denver, Colorado. It was about a 100-employee business that manufactured our modular Bitcoin mining data centers. And we sort of designed this data center with them, and they had been our contract manufacturer. And we acquired them because we thought, look, we should vertically integrate this. We know that there's sort of margin to recapture there, we're going to be doing this for a long time. It makes sense to make that investment to control our supply chain, iterate on designs faster, and just kind of have that as part of the business. As we get more into the data center side of the business, the big AI data centers, so much of the CapEx is electrical infrastructure. It's switchgear, it's power distribution centers, it's remote power panels. These probably don't mean a lot to a lot of people here, but they're pieces of electrical kit that they can have very long lead times to manufacture because post-COVID, like everything, lead times just blew out across all manufacturing.
(35:55):
And so we pivoted 100% of the facility's capacity to just focusing on these core components that go into our AI data centers. And then we've expanded it to what will be... We will end this year with approaching 500 employees across four factories. And we'll be able to make really all of the switchgear that we need to go into our data centers, all the power distribution centers, and a lot of the other components that go into our data centers. So it's going to save us money, it's going to save us time, and it gives us more control over the supply chain.
Cody Simms (36:23):
I've heard your co-founder, Chase, say that he thinks AI is the greatest blue-collar workforce development event at least of our lifetimes if not of all time. When you talk about the welders, the electricians, the plumbers, everybody that is going to need to get involved in building these things.
Cully Cavness (36:42):
For sure. I mean, we have 3,500 people employed in Abilene right now on the skilled trades labor side. And these people, they can sort of name their price. I mean, electricians just command a real premium in the market right now. It's an amazing opportunity for anybody that's trying to start a new career path. You don't have to take college debt. You get a skill and a career path that as far as I can tell, it's going to be there for a long time. And they have this big opportunity to work on something really exciting. I mean, I think they're excited to be building this facility that's going to shape the future.
Cody Simms (37:15):
And not just the data center itself, but the power stations that are going to go in around it, right, there are branches out from there too.
Cully Cavness (37:21):
Absolutely.
Cody Simms (37:22):
It strikes me that a lot of these folks, skilled labor workforce in particular was largely left out of the internet tech boom. And here you have potentially an opportunity for greater participation.
Cully Cavness (37:36):
Maybe. I don't know about the internet tech boom. There are a lot of jobs that I think were created that maybe don't get all the headlines. But we built a lot of data centers for the internet tech boom too.
Cody Simms (37:47):
Yeah, that's true. Okay. So we've talked about your original business in Bitcoin mining and digital flare mitigation. We talked about what you're building in Abilene, which I assume presumably you're going to try to do more of those over time. You also have this cloud AI business that got built in between the two. What is that for, who's using it, what's the setup of that business?
Cully Cavness (38:06):
Yeah, so Crusoe Cloud is a business model where we own the GPUs. We've purchased the... Whether it's Nvidia or other manufacturers, their hardware, we've installed it into our own data centers either built or rented. We've built a software layer on top of that, done all the network engineering. And we offer that infrastructure as a cluster for rent for anybody that wants to operate inference or training workloads on GPUs.
Cody Simms (38:32):
So what are some examples of the kind of stuff people are doing?
Cully Cavness (38:34):
Everything from enterprises, a bunch of the big names the everybody in that room would recognize, through the long tail of startups. Some folks in the room here who are trying to basically train a model, fine-tune a model, or operate inference and they need GPU capacity. What we're doing is we're bringing the sort of best versions of Nvidia hardware to market as fast as possible. Often, we're the first to market with some of these new SKUs. We're doing it in a capacity where we're tapping into our energy sources that are often cheaper. So offering lower pricing really across the board
Cody Simms (39:09):
Back to that sort of energy-first mindset.
Cully Cavness (39:11):
Yeah. I mean, our biggest cluster is actually deployed in Iceland, which is a 100% geothermal and hydropowered cluster, basically getting affordable renewable powered GPU compute quickly.
Cody Simms (39:21):
So everything about the AI space, to me, to steal the name of this podcast, feels inevitable, like it's going to happen. We could sit here and talk about should AI happen, should it not happen, is it the right ethical thing. Are the reasons to be afraid of it? Are there reasons to be excited for it? But it's clearly happening. And so the market is going to do what the market does. The question I have is, where does clean power today win and where is it still behind?
Cully Cavness (39:49):
I think clean power wins in the circumstance where the grid can ramp up and down to accommodate it. And as soon as it's a requirement to be baseload is where it still has work to do. It's where you have to start adding a lot of expensive batteries and overbuilding the generation to charge those batteries during the day or during the windy period. And where it becomes uncompetitive economically in some cases. So with an AI data center, that data center has to run every day of the year regardless of rain or shine, day and night, windy or not. So it's requiring a baseload power source that is fully dispatchable. The options there are-
Cody Simms (40:28):
Just to be clear, you can't... The requirements from your customers are not, one, like it was with Bitcoin mining where you can ramp it up or down depending on the power price.
Cully Cavness (40:35):
Correct. For a training data center and for most inference data centers, that's going to be the requirement. The market may change and we're having conversations about, hey, is that really a requirement or are there some versions of this that can ramp in response to energy profiles? But for now, the when somebody comes through the door, it's like I need four nines of reliability, meaning 99.99% uptime. So what are the energy sources that can meet that today? You've got a nuclear power plant that's baseload. You've got coal power plant, you've got natural gas power plants. In some places, you have geothermal or hydro, pretty geographically constrained. And then you could have renewables plus batteries. And I would say of that mixture, the ones that can actually respond very quickly to bring hundreds of megawatts to bear in the next year, you cannot build a nuclear power plant that fast.
(41:24):
Hopefully, the folks that are working on SMRs can solve that. Crusoe's pretty committed to that path and starting to work on partnerships to be one of the first to have a nuclear-powered data center from SMRs. But today that isn't an option. Coal is frankly too unpopular politically, at least for the time being, to see that we're going to really ramp hundreds of megawatts or gigawatts of new coal production in the United States. And then renewables plus batteries, a kilowatt-hour of solar is extremely inexpensive, a kilowatt-hour of wind is extremely inexpensive when it's being generated. But if you want that to be stored and then delivered to you at another period of time, that becomes expensive. That becomes two or three times typical grid prices for power if you really are going to have 24/7 battery coverage. You don't really see this today at a utility-scale. And the supply chains and the scalability of that to deliver hundreds of megawatts next year, we don't see that that's realistic today.
(42:22):
So that leaves you with gas. That's the most mature supply chain that can actually do something like give you 500 megawatts, a gigawatt in the timeframe of a year or two.
Cody Simms (42:33):
And yet the turbines are all backlogged, aren't they? It's hard to get gas deployed today.
Cully Cavness (42:37):
We pre-ordered some of those. It is a big part of how we're doing what we're doing. And there are other ways to get gas turbines. I think a lot of the lessons learned have been around supply chain and how to get creative around how to procure critical components of the supply chain. But basically, you can actually get delivery of a gas turbine, build it, commission it, and operate it within the timeframe of one to two years. And the pathway then has to be to clean that up, has to be carbon capture and sequestration. And so what we're working on right now is we've been investing millions of dollars into the feed studies for CCS projects. My personal belief is if we're actually going to get to large-scale post-combustion CCS, the way that's going to happen is in partnership with a data center. Because you have an off-taker who has the highest willingness to pay for environmental attributes. These companies often are already paying hundreds of dollars per ton for nature-based removals or anthropogenic removals.
(43:35):
They also are really focused on their climate goals that they've already announced. They've made big statements that AI is now pulling them away from. So they're going to have to find ways to respond with the speed, but also have it be clean.
Cody Simms (43:49):
The chief sustainability officer of Microsoft recently said their own net-zero goals in 2020, and they said it was a moon shot at the time and we're still committed to them, but the moon is getting farther away. Because the amount of energy they're needing now is much greater than they thought it was going to be in 2020.
Cully Cavness (44:05):
All these ESG reports are going to take a step back because of this ramp in power and the associated emissions of that. And so the question is, okay, what are the solutions that get you from here to there and can bridge some of these gaps back to the trajectory they want to be on?
Cody Simms (44:19):
So, it sounds like there is some risk then that we're going to bake a bunch of new fossil fuel infrastructure in right now, which will last for another 20 or 25 years during this AI arms race. But you're saying you believe that it's the best way to fund and get adoption for the carbon capture part of that sort of clean baseload power around gas that is gas plus CCUS.
Cully Cavness (44:42):
I think our approach is very specifically around building gas to be the temporary bridge for the grid to then expand. The grid can then be the thing that absorbs a lot of the renewables, the batteries, the SMRs. All the mix can go through the grid to come to these data centers in the long-term, which have a short-term problem to solve.
Cody Simms (45:01):
Do you think we'll get CCUS cost affected before batteries just come down the cost curve or it's a matter of you need it now, so it doesn't matter?
Cully Cavness (45:09):
I mean, there are some projects that we're seeing where it can be cost-competitive. More cost-competitive than 24/7 batteries on a big renewable deployment.
Cody Simms (45:18):
What do you worry about? What keeps you up at night?
Cully Cavness (45:20):
We've got 800 people working on all kinds of exciting projects every day. It's a lot to keep track of. I won't try to lie and say that it's easy. But I focus on trying to execute the vision of the company, try to fulfill this vision that we've set out for ourselves.
Cody Simms (45:35):
So most of this that we've been talking about, the AI growth and arms race is a hyperscaler battle. These are big companies kind of duking it out to build the future of this. This room is full of startups, innovators, people wanting to jump into the fray. What advice do you have for folks? We started the conversation by saying your company has recently turned that corner from being one that has to knock on everyone's doors to one that is getting your own door knocked on a bunch. What advice do you have for founders who are in the room or founders who are listening?
Cully Cavness (46:07):
You are going to have to have an enormous amount of persistence and endurance to make this work. And I would not start unless you think you have that. It is just extremely hard. And if you can be honest with yourself and say, I'm ready to do something extremely hard for a very long period of time with a lot of uncertainty associated with it, then you're actually very well positioned to go out and do the thing that you want to do. But what I think kills a lot of companies is the 1/2 commitment, and it just requires a very full commitment. I was thinking about this on the way over here, that earlier sort of chapter of my life, I was very into these extreme endurance sports. I was into running 100-plus mile marathons and come across the finish line with no skin on my feet and broken vertebraes.
Cody Simms (46:56):
And your co-founder, Chase, has climbed a bunch of major mountains, right? You guys are kind of crazy.
Cully Cavness (47:02):
Yeah. But honestly, I was thinking about it might just take that level of abuse that you sort of have to get through that to build something sometimes. So, I certainly am not trying to discourage anyone from trying to be an entrepreneur, but I'm trying to say you should psych yourself up and be completely clear-eyed about how extremely difficult this is going to be. And if you're actually excited about that, embracing a challenge like that, it's the greatest thing you can do. But just beware going into it that that's what it's going to be.
Cody Simms (47:33):
What's Crusoe in a decade? You've had three different companies in seven, eight years. Seven years.
Cully Cavness (47:38):
Yeah, seven years.
Cody Simms (47:39):
Seven years so far.
Cully Cavness (47:40):
I think we're the biggest independent owner-operator of AI infrastructure in the world.
Cody Simms (47:45):
Any last thoughts?
Cully Cavness (47:46):
I appreciate you having me here. I mean, I think the conversation around energy, AI, climate, the human economy, it continues to be the thing that's most fascinating to me. I tried to kind of explain that I've been through this mental cycle a few times, going from climate doomer and guilt to environmental pragmatism, to supplying the energy that we need to have prosperity. And there's that tension... I certainly haven't figured it out, but I think these types of conversations are really important for us to debate and discuss exactly that issue. So I really appreciate what you do with the podcast. I've listened to a couple of episodes on the way down here to try to get ready. Grateful to have the opportunity to talk to this crowd.
Cody Simms (48:27):
Oh, I have one other question for you. There's obviously a lot happening in the federal government right now around energy policy and, in particular, tied up in what's going to happen with all the various tax credits related to the Inflation Reduction Act and whatnot. Your thoughts on how that should play out or what you hope to see?
Cully Cavness (48:45):
We're of the opinion that we need to accelerate the abundance of energy and intelligence. That's really becoming the mission of Crusoe. And we would advocate for any policies that are going to help accelerate the abundance of energy. And to remove important financial incentives or mechanisms that would restrict the amount of energy we have access to in this country and around the world, it'd be better to just try to see a world where we're having access to more and we're incentivizing the generation of more electricity. That's clearly going to be the thing that constrains our development as a species and our development of AI, which is, in my opinion, going to be the thing that unlocks the next wave of human prosperity and flourishing and success across the world. So we should have more energy and we should not do things that restrict our access to energy.
Cody Simms (49:35):
On that note, are you rethinking how you describe Crusoe now that the origin of the company, this Bitcoin digital flare mitigation business is moving out? You kind of are almost fresh-starting the business with this incredibly large thing that you're going to build. Does that change how you describe the company?
Cully Cavness (49:52):
It does. I mean, we've gone through this period just recently where we've reframed our mission and values internally. And we still have a very strong commitment to environmentalism and sustainability, and it's been embodied in a new core value. We have six core values. One of them is bend the arc of energy towards sustainability. And each core value has a set of practices, which is basically how an employee can think about living up to this value. As it relates to that one, one of the practices is to use our purchasing power to create a demand signal for energy innovation. That's the practice... One of the four of how you would live that value. When we're spending 15 billion at a data center campus, there's a lot of opportunity to invest in SMRs, to try to be the first to have a PPA with an SMR. To invest in innovative approaches to batteries and solar.
Cody Simms (50:46):
Not only you're no longer knocking on people's doors, you're getting your door knocked on, you actually get to become the kingmaker. Like we can help make these things happen.
Cully Cavness (50:53):
I'm sure we've invested as much as anyone on carbon capture and sequestration engineering. And really developing a set of projects where that actually could work. And understanding the real geology and the real requirements and the amine capture and the implications for the turbines and all this. So using our purchasing power to create demand signal for energy innovation is really one of the key new forms of our impact going forward. It used to be let's abate methane and reduce waste by generating power from flared gas. We're not doing that anymore, but what we are going to be doing is spending tens of billions of dollars around energy infrastructure and digital infrastructure. There's going to be a lot of opportunity to seed entrepreneurs and projects like many of the ones that are in the room here today. And I think the other element of the way we think about it is, like I said earlier, if we're going to have these breakthroughs, which I believe we need, I just think the status quo is not going to work.
(51:46):
We just need to have breakthrough energy technologies to really get to any of the climate goals that are widely accepted internationally. AI is, obviously, going to have to be part of the solution for that, in my opinion. So that's not as much in our direct control, but we're creating the infrastructure to build the tools that is going to lead to the fusion or the breakthrough in battery materials or whatever the case might be. And we've already published a few case studies of how that's happening. SES materials using AI to innovate battery materials. Avalanche Energy working around fusion. Just a couple of examples.
Cody Simms (52:23):
We are between this wonderful group and happy hour, so I think we will wrap it at that. Thank you so much for coming down to Austin. I was born in Texas actually, but I don't live in Texas, but clearly, you all have a huge vested interest in Texas at the moment. And I have just been blown away by everything that's going on in this ecosystem. So thanks to everyone in the room for coming out and supporting everything that you all are building. And thanks to Cully for joining us.
Cully Cavness (52:52):
Thank you.
Yin Lu (52:57):
Inevitable is an MCJ podcast. At MCJ, we back founders driving the transition of energy and industry and solving the inevitable impacts of climate change. If you'd like to learn more about MCJ, visit us at mcj.vc, and subscribe to our weekly newsletter at newsletter.mcj.vc. Thanks and see you next episode.
