There's probably no one in all of AI who understands power and power supply and compute and compute supply at the level of depth that Chase does right now. And so it's incredible seeing him talk about how he's bringing those two things together.
So today's guest is Chase Lochmiller. He is the CEO of Crusoe. Crusoe most recently raised over 1.3 billion in their series E. They are worth over $10 billion. The company has been on an absolute tear sitting at the intersection of two of the most important things in our world today, AI and power.
The Compute 100 is a podcast about the most consequential market in AI. These are the conversations with builders and investors about the technology and the capital powering the greatest infrastructure buildout ever. I'm Brian Shector.
I'm Gabby Lorenzi. We're investors at Primary. Let's dive in to the Compute 100. This podcast is for informational purposes only and does not constitute investment advice or an offer to buy or sell any securities.
Welcome to the podcast. We're thrilled to have you here.
Chase, what is the path to Crusoe becoming a trillion dollar company?
I mean, look, I think sometimes there's a right place, right time element for a business and we've sort of, you know, Crusoe's always taken this energy first approach to developing compute infrastructure and building things at, you know, really large scale. We happen to have found ourselves at, you know, the convergence of two of the largest industries that have ever existed in human history, energy and computing. And to me it's actually not that crazy of a path to think about Crusoe as a trillion dollar company. It is frankly being able to ride this very large wave and being a critical infrastructure provider to managing intelligence for the global economy. And I think there are going to be numerous, you know, trillion or multi-trillion dollar companies built in this ecosystem. And I think Crusoe has a very good path to becoming one of them.
And so I mean I'm sure at some point we'll get into the origins, but I'd love to just kind of continue on this. But what is the unique energy insight or insights that's different from CoreWeave?
Sure. I think as a business, from the early foundations of the company, I felt that at scale energy becomes a bottleneck and therefore we should be taking an energy first approach. And if you look at previous generations of GPUs that actually had lower cost, lower Nvidia gross margin, the key operating expense for operating large clusters of GPUs is actually power. And so if you could actually dramatically reduce your overall power costs, it actually reduces your total cost of ownership of computing infrastructure and intelligence infrastructure. And so that led us to saying, okay, well for most of these use cases you don't need to be co-located exactly at the same place where some web application is being hosted. In fact, for the more computationally complex things, latency becomes less of a bottleneck because you're compute bound, you're not network bound. And what that leads to is a philosophy of saying, okay, where can I build something where I can access low-cost, abundant energy. And we've taken an approach of trying to have a bend towards cleaner energy as well, and that ends up in us building in remote locations that haven't historically been data center markets but are very rich in energy resources, places like Abilene, Texas or Amarillo, Texas or West Texas broadly and a number of other markets throughout the United States.
We can go somewhere where there's a ton of energy that's a lot of wind, right?
Yeah. So look, I think there's areas of the world that just naturally are windier or sunnier when you think about the renewables. Abilene, Texas for instance, the reason we went there was that a lot of renewables had been built there and the problem they had in that market was actually that power prices were frequently negative because there was so little demand for power and there's been such a large investment in wind generation because it's so consistently windy there and wind developers were incentivized by production tax credits where they get paid to produce and sell renewable energy independent of there's real demand and that led to kind of this negative pricing. And we said, look, have I got a solution for you buddy, we have something that actually requires a ton of demand for power which is this AI infrastructure. And I think it makes sense in the sense of like if a wind turbine costs X, wouldn't you want to be building in the place where you're producing power 55% of the time versus 20% of the time? Similarly speaking, like with solar, we want to build solar where it's sunny. It sounds dumb, but it's as simple as that, right? You don't want to build rooftop solar in San Francisco where it's cloudy and overcast all the time. Building it in a market like West Texas where it's very consistently sunny and you get high capacity factors from that infrastructure investment in solar, it makes a lot of sense. And so that's what's led us to various markets. I mean there's other constraints that come into this in terms of permitting and where generation is available, where gas is available and low-cost. And ultimately that led us to building in these types of markets and we've been building some of the largest AI infrastructure in the world through these very rich energy resources.
Is that going to run out as an insight? Like how do you continue to stay ahead with this energy focus?
We're always kind of staying at the cutting edge of a lot of different things, whether it's, you know, energy generation is one aspect of things, but there's so many different elements that go to this, right? I think new energy technology we're always continuing to stay at the forefront of. Whether it's helping bring to market new forms of gas generation. We actually announced a big partnership with Boom Supersonic who was building supersonic aircraft and they basically took a lot of the innovations that they had built in developing these supersonic jet turbines and we said, hey, what if we use that jet turbine to produce power for AI data centers instead of supersonic jet travel. And they said that's awesome. And we partnered with Blake Scholl, who's a great CEO and amazing entrepreneur, to help them bring to market a power product that's using that same supersonic jet engine to produce power for data centers.
It just so happens that that's like a great power producer.
Yeah. I mean it's the history of power production, right? I mean that's what GE does, GE Aerospace produces the jet turbines for big aircraft is the same reason that GE Vernova came out of that same class of turbines. It's the same playbook with a new type of company. It's actually more so the case with supersonic jet turbines. And one of the reasons for that is the engineering process they went through for these supersonic jet turbines is meant to actually have a higher temperature air intake because of the speed of the planes. And what that results in is in a market like West Texas where it's very hot, traditional gas turbines will actually take a derate. So you actually have to say like if it's a 20 megawatt turbine, when it's really really hot in Texas, if it's 110 degrees, you will only be able to generate 16 megawatts of power, 15 megawatts of power. But these supersonic jet turbines actually are designed to have very high temperature inlet because of the high pressure and we actually take less of a derate. So it's actually beneficial in certain ways.
And that's also combining with real supply chain constraints on just traditional turbines.
Exactly. So our goal in partnering with them was like how do we help bring a new turbine manufacturer to market that can help drive down costs for us and expand the amount of power that can be produced for AI infrastructure. So part of the DNA of Crusoe is like we know how to identify power opportunities that are non-obvious that will allow us to create an efficiency on top of something that's already available. I think the company has an energy first DNA. We did something with Redwood Materials who listeners may or may not be familiar with but Redwood was a company started by the co-founder of Tesla, JB Straubel, with a focus on recycling batteries, like end-of-life EV batteries. How do you handle when you have these huge fleets of EVs and over time as they lose their charge capacity, what do you do with the batteries? Trying to recycle that lithium, all the different components. Well, we did something with them where we actually built the largest microgrid in the United States. So it's off-grid solar plus these end-of-life EV batteries that still have a valuable amount of charge in them. And we're able to basically power a data center fully off-grid with solar plus batteries. And because they're these end-of-life EV batteries that have already been used, already been monetized, already been advertised once, the overall cost of this is actually cheaper than the grid power price in Northern Virginia.
So, we're able to do this at sort of a pilot scale, but again, it's the largest microgrid in the United States. But our hope is to scale it up quite a bit. So, it's examples like this where we believe that we're doing things that other people aren't even thinking about. And part of it is just, frankly, it's in our DNA.
There were three things you just talked about. Abilene has lower, if not negative, power prices that made it really opportunistic to build. You have Boom Supersonic and using those turbines. And then you have this Redwood Materials collaboration. Those three things seem like you just walked into the most perfect storm of opportunities and all the other things that Crusoe is doing to be energy first. If you look back five years, could you even have imagined that this would have worked out as well as it has?
Maybe that... you know, I think my answer is yes. You know, as an entrepreneur you're sort of hopelessly optimistic in all scenarios, you believe even if it seems completely irrational to other people, you believe the future is going to exist in some way. And this was honestly the future that I believed would exist is that energy would become a bottleneck and AI would become embedded in all aspects of the economy and the compute scaling needs would be enormous. I was actually reflecting on, there's a blog post by this guy Tim Urban. I don't know if you guys ever read his blog, Wait But Why? But he had this whole blog post on the road to AGI. And this was formative in my view of the world in the 2015 era, reading him, reading The Singularity Is Near by Ray Kurzweil early on, just all of those philosophies had this notion that compute, in order to get to super intelligence, compute was the bottleneck and the bottleneck for compute was power. And so those two things went hand in hand, compute, power, and data. Do we have enough of those things? If we have enough of those things we can sort of achieve superhuman capabilities through digital intelligence.
There's this additional move which is about the efficiency of power and finding opportunities. I would love for you to talk about the first data center that you got going with the offshoots from the gas.
Yeah. So that was another example of taking this energy first approach to things. And honestly, the Bitcoin ecosystem is a very efficient ecosystem in terms of seeking out low-cost energy resources to power computing infrastructure. It was the first manifestation of this trade, right? Go seek out low-cost power and then use that power to produce SHA-256 hashes to secure and validate transactions on the Bitcoin network. And when we started the business, frankly, I didn't have a background in energy. It's really my co-founder that has the background in energy. I sort of came from this background of AI and computing and cryptocurrency. And when we started the business, it was actually after the two of us were on a climbing trip together in the Rocky Mountains. And we spent pretty much the whole day with him educating me on this process of natural gas flaring. And I was so mind blown by this notion that a company would basically burn off this waste product and it would happen 24/7 and it's something that has value in one context but in this other context it's a nuisance. It's a waste product. It's a thing that you have to manage. It's a liability. And really what's happening there, just for the education of the listeners here, is an oil company will drill a well for oil and what comes out of the ground is a combination of three things: natural gas, water, and oil. And the oil and the water can be separated into these tanks and you can bring trucks to the site that then truck the water to a water treatment facility, truck the oil to an oil refinery. But the gas, unless you have access to a pipeline, because it's in gaseous state, the best and most economic thing to do with it is to actually just light it on fire. You just burn it off. And that whole concept really blew my mind. And it was only when I first saw it in action when I went out to the oil field with my co-founder and I saw these giant fireballs in the middle of nowhere and I'm like, 'Holy crap. What? They just do this 24/7?'
And you were just like, 'Let's put a data center here.'
He was like, 'Yeah.' And I was like, 'Man, this is insane.' And exactly. I said, 'Let's, the key metric of profitability in Bitcoin mining and for computing more broadly is going to be energy costs.' And so our first application was basically building these mobile and modular data centers that would be co-located alongside this waste energy resource. And then we brought power generation to site. So we were actually managing our own power generation, our own data center infrastructure and operating it in these remote, very hard to operate conditions in the oil field. And we built a business that scaled that to more than 200 megawatts and serving some of the biggest companies in the world, folks like Exxon. And it was a very successful growth of an energy first computing platform.
Fascinating. I want to know the... so right now Jensen is saying that compute equals revenue and we're moving into a world where the connection between compute tokens and economically valuable productive output is getting more and more clear. That's an incredible framing for why we want more and more data centers. You said a few things that point to you sort of being on this from the early days, the 2010s and thinking about AI. What about AI then? Were you seeing that compute would be so central and so important for the world?
That's a good question. I guess my... and you were at MIT, is that right? I was at MIT and then Stanford for grad school in computer science and AI. And then I was working in the quant trading space. And the quant trading space is this interesting ecosystem of researchers that are not often publicly publishing things. They're sort of behind the scenes consuming a lot of cutting edge research and trying to adopt new cutting edge research to produce better, more valuable predictions than the rest of the market. And in that domain, my original role, we were building a lot of forecasts for security returns and my job was honestly a lot of feature engineering. So we would build these features, oftentimes built on economic relationships and using a lot of physics-based models that would say, okay, if this security goes up, we sort of expect this security to do that and here's how we're going to model it. We're going to look at a lot of data historically and then we're going to use more classical machine learning techniques to try to fit a model to this economically driven idea. Things like linear regression, logistic regression over these complex engineered features. When deep learning, the boom in deep learning that happened after the publishing of the AlexNet paper and convolutional neural networks and recurrent neural networks and some of the time series based mechanisms of thinking through predicting the next token in a sequence, that really re-architected a lot of our thinking and we started experimenting a lot with these deep learning based approaches to quant. And that led me down this path of both exploring deep learning as a mechanism to produce alpha for the quant finance space as well as exploring a lot of other things in my personal free time around computer vision and natural language processing that I just found to be really, really fascinating problems to work on. And then we shifted these workloads from CPUs to GPUs and I was like, wow, we're starting to consume a lot of power with these GPUs, they take a lot more power than a CPU and you need more data center infrastructure to make this happen. And it was sort of at that point where I felt like with pretty high conviction that AI was this meta-science that it's not one thing that it helps with, right? It literally improves every industry. Every industry that produces data, which is to say every single industry, can be made better and can be improved by AI and artificial intelligence driving efficiencies, creating better predictions, aiding in the human aspect of a lot of these industries. And so I had this pretty strong early conviction that AI was going to be embedded in every aspect of the economy.
Cool. Maybe taking a step back, I'm thinking about this discussion of the flaring piece and there's maybe one thing I want to hit on.
So I think there's two important characteristics when you think about our early days of taking this energy first approach to computing with flare gas.
One was that we were taking an inherently energy first approach where we said let's build computing infrastructure and bring computing to where we can access low-cost, abundant energy. That's number one. Number two was we actually thought through the computing problem as a modular problem. We built these mobile and modular data centers that were initially housing Bitcoin mining equipment and then we actually had our first GPU-based modular data center that was deployed in the oil field powered off of flare gas in the 2019-2020 time frame. And we were inherently thinking about the computing problem as the composition of these modular components. And I think today as we look forward into the future and the scaling of AI, the amount of infrastructure that needs to be built inherently has to happen on a modular basis. And we sort of think about that both in the development of these very large AI factories, like multiple gigawatts, like what we're doing in Abilene, Texas, where what you're doing is you're building something and then you're repeating it over and over and over again. So you can actually create these modular building blocks, manufacture them in high volume in a concentrated factory, ship those components to site and you're doing less on-site construction and more on-site assembly. It's sort of like an IKEA type model or a Lego based model where you have these individual building blocks you're putting together. The second is we believe that the future of AI infrastructure is going to have this barbell where you have these big concentrated giant AI factories of multiple gigawatts and then you're going to have smaller AI factories that are hundreds of kilowatts to a small number of megawatts widely dispersed everywhere and that's going to be serving the scaling of inference and the scaling of the inference economy as the demand for tokens grows at an exponential rate.
Man, there's a lot of things to talk about. Something I would love to know is what did your seed stage deck look like? What was the business when... I wanted to get to some entrepreneurship questions. What did it look like when we're going to just go put data centers next to gas flares?
That was kind of the pitch and we were like, look, we have this... because an important aspect of the business is we actually built our... so I funded the initial MVP, right? So basically I... and the MVP was the modular data center, a small modular data center. We worked with an electrical fabrication shop to basically build out this initial data center that we were using for Bitcoin mining.
What compute was in there? What type of hardware was in there?
They were ASICs. So they were ASICs designed for Bitcoin mining. And what we did was, I mean, we felt Bitcoin was sort of the fastest path to monetization. But from the seed stage I had always felt like the bigger opportunity was going to be AI infrastructure. It was also the thing that I was personally more interested in. Like I feel like Bitcoin and AI are like my two loves but AI I love more than Bitcoin or something like that.
Love all your children equally.
I don't know. The thesis around the seed investment was that we had executed this first piece and we actually got an oil company to say yes to like, okay, these guys can come out here and flaring is a problem for me and they can take my flare gas and run these modular generators and power their data center. And so we had that stood up by the time we were raising our seed round. And so our seed deck was basically focused on scaling that opportunity in Bitcoin and then hopefully building out an AI platform that was an energy first approach.
Were the gas companies charging you at the... or the oil and gas companies charging you at the time for that waste?
Yeah. So this is one of the interesting facets of the business. Our very first deal the gas was just free. Yeah. So it was just like, just take the gas. And that actually helped solve a problem for me. Shortly thereafter, I sort of took the approach that if this thing was going to fail, I wanted it to fail pretty fast and then move on with my life and go do the next thing. And so I said, 'Look, if flaring is truly a problem for these oil companies, they should be willing to pay us.' This gas that they're burning off, if it's an environmental liability for them or there's some constraints from a regulatory standpoint or an emission standpoint or something that's actually bad for them, they should be willing to pay us to get rid of this problem for them. And so, the initial set of contracts that we signed, we were actually getting paid a service fee to take the gas off the hands of the oil companies. And you actually have to have a handoff of the commodity because there's all these mineral rights that are owned by the royalty owners. So our contracts were typically we would pay 1 cent for the gas and then we would get paid this big service fee for deploying the equipment and providing the service of flare mitigation.
Wow. Okay. So seed stage pitch deck is we've tested this out. We've done it and we need to scale this. We need to make it a lot bigger.
We need to scale it and we're going to expand into other new frontiers like AI.
AI into AI. And so you started Crusoe with one of your high school friends, correct?
Yep. So, I started Crusoe with a high school friend of mine, Cale Cafines. It's my co-founder and known him for a long time now.
Amazing. And so, you guys raise your seed round. What comes next? What was that journey to Series A? And what time frame was this? What year was that?
So, we started the company in 2018. And that's like again, so I put up the first money. And that gave us enough to go through this design process, build out this first modular data center, buy a bunch of ASICs, run our day-to-day operations. I wasn't paying myself anything or whatever. But it was enough to get us through that initial development phase. We raised a seed round that was with Founders Fund and Bain Capital Ventures as the two primary investors in that round with a bunch of other friends and family and a number of other funds. Between that and Series A, it was really like, is there a real... they did this once, can they do it again? Is there real demand from oil companies to actually solve this problem of flaring? And we expanded the number of basins we were involved in. So our first deployment was in Wyoming. We then got into the Bakken in North Dakota which was the second largest flaring basin in the United States. And we ended up signing some big deals with big customers. And honestly, it was one of these things where if I could ever get someone to go on a field trip with me, they were totally sold. Like if I got someone into the field and they drove around the Bakken with me, they were like, 'Holy crap, there was so much gas being burned off with no beneficial use. This is totally insane. We should turn all this into Bitcoin.' Or we should turn all this into computing cycles for more valuable use cases. And I think people would just see it. It was a very visual experience and then they saw the moat which is that it was difficult. You're often operating these very challenging conditions. So if we could figure out how to operationalize hard things, that would actually create a huge strategic advantage for us over the course of time.
Fascinating. Okay. And so you started to scale up. You're starting to show repeatable value through the Bitcoin business. Today what percentage is still Bitcoin versus AI?
We sold our Bitcoin business.
We are still a large shareholder in a privately held company called NaIdig which is where we sold our Bitcoin business to. But so we basically at some point there was this whole period of, around our Series A, in fact, a lot of investors were like, 'Holy cow, your Bitcoin economics are incredible. Why are you screwing around with this AI thing?' And it was like, 'Well, that's the thing. That's what I care about. That's what I want to do. I view this as an opportunity. This is what Chase wants to do. If you want to invest at this stage, that's kind of what I'm going to do.' And so I think there was a lot of folks early on that just doubted that AI was going to be an opportunity and wanted us to just focus entirely on the Bitcoin ecosystem. And my commitment was like, look, I view there's a huge opportunity in the Bitcoin ecosystem and I want to build this whole thing in AI and it's very, very important to me. And so for a while our AI cloud platform business was sort of like Chase's side project experiment that was happening within Crusoe. And over time, the ChatGPT moment was kind of this shot heard around the world and we'd sort of been building this platform and we launched our AI cloud platform publicly to paying customers early in 2022 after years of experimentation and product development and figuring out what we wanted to build. And then of course after that moment there was just this insatiable demand for the type of compute infrastructure that we've been building for years.
No longer your side project.
No longer, became the main show. Yeah. It just scaled so quickly and folks just recognized... and sorry, it scaled very quickly and we kind of found ourselves in this position of having two big businesses, right? Two businesses with nine figure revenue streams. And there were investors that would come to it and say like Bitcoin is such a polarizing thing. I still don't understand exactly why, but it is such a polarizing thing for certain investors where certain investors would come to us and say I love what you're doing in AI but we can't invest in anything that's related to Bitcoin or I love what you're doing in Bitcoin. I don't really know anything about AI. And so we kind of had these two companies existing under one roof and we said, you know what, I think at this point we just felt like it made sense to be entirely focused on the opportunity we were seeing in the AI landscape. And as we were developing this giant largest data center in the world, we said, you know what, we should be entirely focused on this single opportunity.
Like give us a map of the Crusoe universe right now in terms of power, GPUs. I want to learn about Abilene and how complex that is but first just what's the universe?
What's the universe? Yeah. So Crusoe is a vertically integrated business. And we're a vertically integrated AI infrastructure business and the reason we chose to take that approach is that this landscape is such a dynamic landscape and it's moving so quickly. The number of different bottlenecks and what those bottlenecks are, they evolve and they change over the course of time. Being vertically integrated provides you this dynamic advantage where you can actually remove these barriers by being able to do them yourself, by being able to control your own destiny. By being your own data centers, build your own power plants, build your own electrical switchgear, power distribution centers so that you can stand up the infrastructure needed to power the intelligence demands. Being able to deploy large clusters of GPUs and manage different networking architectures on the back end. Being able to find different ways of monetizing that infrastructure has been a critical aspect to our core strategy. Where we have built out an infrastructure as a service business that's really focused on selling to folks that want to manage the infrastructure directly but they really need core infrastructure engineering teams to be able to manage large clusters of GPUs.
Just to understand that, I'm assuming that's the hyperscalers, labs who say we want to manage the infrastructure ourselves but we need someone who's going to install it, monitor the power, is that...
