Crusoe CEO Chase Lochmiller: Paraguay data center project, modular 'Crusoe Spark' product, and global expansion to Iceland and Norway
Nov 17, 2026 · Full transcript · This transcript is auto-generated and may contain errors.
Featuring Chase Lochmiller
Speaker 1: Sorry, Rashid. How are Good to see you.
Speaker 3: Good to see you.
Speaker 1: Welcome back to TBPN. I'm gonna throw that headset on. The microphone goes on the left, and we're gonna want that microphone as close.
Speaker 6: If you had it
Speaker 3: right, John, and you may you may have seen us. Sorry.
Speaker 1: My bad. That looks great. There we go.
Speaker 3: There we go.
Speaker 1: How's life?
Speaker 3: Great to
Speaker 1: see How are doing? We got cowboy hats? Yeah. We got cowboy hats.
Speaker 7: But They
Speaker 3: don't They don't fit.
Speaker 1: Fit over the headsets. Oh. You gotta wear them this way, I guess. I don't know. Anyway, are you a big cowboy hat guy right here?
Speaker 13: I got a couple cowboy hats.
Speaker 1: It's a functional product. I I... It's so aesthetically styled.
Speaker 3: I wear a cowboy hat quite a lot Yeah. Purely for the, yeah, purely for the function.
Speaker 1: Yeah. The sun protection. Sun protection. Is fantastic.
Speaker 13: Yeah. Nice wide brim.
Speaker 1: Yeah.
Speaker 3: You know,
Speaker 13: high high quality sun protection.
Speaker 1: Are you out here in Texas often? Is this the biggest project? Are you all over The United States now, all over the world? What what is the footprint of Crusoe these days?
Speaker 13: I'm in Texas a lot. We have a big footprint here both in Abilene and other, you know, large campuses that we're developing in Texas. But, you know, we're we're building all over The United States
Speaker 1: Yeah.
Speaker 13: And, you know, expanding globally. So, you know, doing a bunch of stuff in Iceland Okay. With geothermal and hydropower. We just announced something big in in Paraguay that's being powered by the 15 gigawatt Itapu Dam down down there. 15 gigawatts? Wow. That's huge. It's it's a lot... So our project is 280 megawatts. Sure. But but
Speaker 1: but you can draw the power from that because they have access.
Speaker 13: Interesting. Exactly.
Speaker 3: That's cool. Yeah.
Speaker 13: And, you know, we have something in Norway. We're doing something in Israel. Mhmm. You know, maybe some plans in Asia. We'll see. Sure. But... Yeah. So we're we're we're sort of expanding globally right now.
Speaker 3: International business, man.
Speaker 1: What what what is the what is the shape of the the team that you show up with? What are you... Are are you handing off to a government? Do you need to bring a hyperscaler or a lab on board with you to one of these projects? Or or can you show up and just say, hey. There's power. We're gonna go build it. They will come. Is it if they build it... If you build it, they will come problem at this point?
Speaker 13: Sometimes we we structure things that way.
Speaker 11: I think,
Speaker 13: you know, typically, we we like to have, you know, at least some some anchor off take in hand to, you know, catalyze some of these bigger project investments. But, you know, broadly, when you think about Crusoe, it's a vertically integrated AI infrastructure platform. Yeah. The key three products that we sell to customers are typically data centers at one layer, where we'll typically work with hyper scale or, you know, very large investment grade customers. Manage clusters of GPUs where we're serving a lot of the innovators in the economy and, you know Yeah. Quant trading firms and folks like that. Okay. And then managed services. So managed inference, managed fine tuning, and all sorts of managed AI services that are that are mostly serverless.
Speaker 1: So How is managed fine tuning growing? Because, like, open source, there's a lot of excitement. A lot of people support it. But it's very hard to actually comp it against, like, lab revenues, which are, you know, leaking every week at this point. And you can see that there's, like, something like a $150,000,000,000 of revenue across our ARR, across Anthropic OpenAI Google. What's actually the scale of that industry?
Speaker 13: You know, I I think it's it's much smaller than Frontier Lab revenue today. Yeah.
Speaker 1: But it's like a Jevons paradox thing where people are using it a lot, but they're saving money, so it's a small revenue base.
Speaker 13: Yeah. Well, I I think there's that, and I think there's a sense of data sovereignty. Yeah. You know, people really control their
Speaker 1: Okay.
Speaker 13: Their models, own the models Sure. And use what's unique to them, which is these private repositories of domain specific data that they can apply to the use cases that they care about. Sure. And what we're seeing is there are customers that are able to get better results on things that they care about. It may not be the best benchmark result like for some general purpose thing, but for very domain specific things, can get better results with fine tuning. So, and we think that's gonna be a trend that's gonna continue to grow and expand as people think about using these rich private data sets to apply to general use cases.
Speaker 1: Yeah, that makes sense.
Speaker 3: Talk more about geothermal and what you guys are working on in Finland.
Speaker 13: Yeah. It's in Iceland.
Speaker 3: Sorry, Iceland.
Speaker 13: Similar area of
Speaker 6: the world.
Speaker 13: Yeah. Close enough.
Speaker 3: A lot of snow.
Speaker 13: Iceland's kind of like this geological phenomenon in like a lot of ways. It sort of has this ultra low cost geothermal power. There's a massive population there. So it's less than 400,000 people that live in Iceland. But they're rich in energy resources and obviously a very attractive environment from a cooling perspective. So you get very low PUEs when you're
Speaker 3: Yeah. Managing data what is it about the environment there that that makes that geothermal energy so abundant? Is it more... Is it like effectively more shallow because there's there's
Speaker 13: heat in the center of the underlying geology of
Speaker 3: it. Yeah.
Speaker 13: It's it's just a a very active, you know, volcanic. Yeah. Yeah.
Speaker 1: Yeah. We all saw the explosion.
Speaker 3: And that means there's more heat closer to the surface whereas, like, we could go here. As you go far enough down anywhere, you... There's... You're gonna get a lot
Speaker 13: of heat. No. That... No. That's right. It's it's it's all about how how how deep you have to drill to get to very, very hot
Speaker 11: hot rock.
Speaker 1: Rock. So... And then you're and then you're piping water down. That's turning into steam, that's turning a turbine, then you get electricity. There's a couple of
Speaker 13: different ways of doing it. But yeah, that's sort of one
Speaker 6: way of
Speaker 1: doing Everything's boiling water. Yeah. We're always boiling water.
Speaker 3: That's a human How? I'm assuming you guys have spent a bunch of time with the local population understanding how they think about this. I can imagine people there being somewhat pragmatic and being like, we have a lot of land. We have a lot of energy. We have a low population. This is like a new industry that can create a bunch of economic activity here that doesn't necessarily disrupt our daily life. But, you know, what is sentiment like in the country? Look,
Speaker 13: I I think I think Iceland, the sentiment's, you know, still reasonably positive. It's not it's not... It hasn't been overwhelmed in the same way, you know, I think the the total grid in Iceland something's like three gigawatts, something like that. And... But, you know, you look at the Texas grid is something like 90 gigawatts and they had, you know, 400 gigawatts of large load interconnection requests and, you know, it was just sort of like a bit of overload there. You know, I think energy intensive use cases have have sought out Iceland as a as a home in the past. It's it's it's actually been a very popular location for aluminum smelters. So a lot of aluminum smelters have set up shop in in Iceland where they're shipping raw aluminum ore across the ocean. Mhmm. Doing the very energy intensive process of the smelting there and then shipping finished products elsewhere. AI is a much better manifestation of that trade. You're moving data over subsea cables. We have three redundant subsea cables that interconnect into Iceland. And then you're sort of doing the energy intensive compute workload on managed clusters of GPUs and then shipping the end and and result Yeah. Whether it's inference or or training elsewhere to, you know, where where where you end up consuming it.
Speaker 1: I know I know Greenland's a different country, but are there any similarities between Greenland and Iceland? Because they are in a similar area, Similarly, leave a
Speaker 3: round of applause for John.
Speaker 13: There are. There are. You know, they both end in land. Yes. So that's like a big big similarity. You know, I I... We we haven't spent much time
Speaker 7: Okay.
Speaker 13: Thinking about Greenland
Speaker 1: as well. Industrialized enough.
Speaker 13: Right? See a very funny, like, data center or a very funny, meme on X. Sure. It was like, you know, Greenland, and then was like data center land.
Speaker 6: Data center land on top of me. Nobody's there.
Speaker 3: We had Hollywood land. No. They chopped off the land
Speaker 1: too. Thirteenth letter.
Speaker 3: It was
Speaker 1: bad luck. Bad luck. What what about Canada? We... I mean, we heard that, like, there were some politicians that saw the American midterms as an opportunity to jump in. Are you seeing that on the ground?
Speaker 13: Yeah. I mean, look. Canada has always been a very compelling market. It's a very energy rich nation. They have a lot of hydropower. They have a lot of oil and gas in Alberta and have invested a lot in renewables. So we think it's a very interesting market and then obviously has attractive cooling characteristics. So we think it's an interesting market. We've gotten close on a couple of things, but haven't quite quite gotten something across the finish line there. But Are you definitely a market we're paying
Speaker 1: Are you starting to think about what the compute campus of the future might look like before they go to space? Oh, yeah. Yeah. Something like... I'm just thinking if you if you check the boxes on it's underground or it looks beautiful and it's quiet and it's solar powered, you're you're you're checking off a lot of boxes that people have been, rightly you or wrongly critical of. What what do you think? Where do you think this goes?
Speaker 13: Well, look, I think there's a couple of characteristics when you think about the build of a data center in a neighborhood where people have problems with. Number number one is like, you know, when it's a big construction site Yeah. Like anybody who's ever had their neighbor do construction Totally. You know, on their on their house
Speaker 9: Yeah.
Speaker 13: You know how annoying it is when there's a big active construction site goes
Speaker 4: toward you.
Speaker 6: We're
Speaker 13: very empathetic to that. But a lot of the other big criticisms like water usage, job creation, energy usage, I think a lot of them are oftentimes misinformed by just things that are spread through social media that are misleading or oftentimes untrue. So the modern data centers are typically closed loop systems. We use almost no water in these big AI focused data centers.
Speaker 1: Dylan Patel had a good quote. Was like, the same water is three In N Outs, which I thought was more grounded.
Speaker 7: Yeah.
Speaker 1: Golf course or something as abstract. In N Out is a funny
Speaker 4: one.
Speaker 13: Mean, of our giant buildings in Abilene uses 10 single family homes worth of water. Mostly like the employees that go to the bathroom, wash
Speaker 1: their plants.
Speaker 13: Oh, sure. That's the water usage. It's not, the GPU's boiling all of Abilene's water.
Speaker 3: So
Speaker 13: I think there's a lot of misinformation out there. But we think there's other ways of building the data center infrastructure needed to power AI that's driving economic results for the world that doesn't need to look like a giant gigawatt scale campus. Yeah. Which is something that we're focused on and that product, that gigawatt scale campus is very uniquely suited for training next generation frontier models.
Speaker 1: Sure.
Speaker 13: I think that's gonna continue to be a thing. Yep. But as we serve these models, you don't need a gigawatt of compute in a coherent cluster. You can do it with much smaller clusters of compute. So Crusoe's actually built this product called Crusoe Spark. Mhmm. It's a small modular manufactured AI data center, an AI factory that we ship to site. It's basically a prefab data center. Sure. So it eliminates all of the construction that's going on on-site. Is huge
Speaker 3: part of the pushback.
Speaker 13: Huge part of the pushback. And also, oftentimes where these are being installed, there's limited labor supply to actually do the construction work. So oftentimes, you have to bring in labor to support the the big builds. Yeah. In this case, we're actually concentrating the labor in a high volume manufacturing environment where you can access the labor very cost effectively and efficiently and you don't have to heavily impact local labor market with short term construction work. And it's also very flexible in the sense that there's actually a lot of areas of the grid where there is excess power. And there may be limited transmission. There's too much generation on individual nodes within the grid. And it actually goes to waste. But they may not be a gigawatt, it may be 10 megawatts, 20 megawatts, 50 megawatts. And this is a very well suited solution to basically say where can we access power that's stranded on the grid. And we can plug in five of these, 10 of these, 20 of these modular data centers.
Speaker 3: Back to the original thesis.
Speaker 11: Yeah, that's the Exactly.
Speaker 5: Bring it
Speaker 13: compute to where you can access low cost energy and mitigate a lot of the concerns around data center builds with these small modular manufactured approaches. We also think it's an important thing from a national security standpoint. When you think about these giant campuses and if you imagine a future where a lot of the economy's powered Running things. By AI. If you have 10 data centers that are running the whole US economy, they become these like sitting ducks for like Sure. Attack point.
Speaker 3: You see that when AWS goes
Speaker 13: down. Exactly.
Speaker 3: If Yeah.
Speaker 10: If have a
Speaker 13: widely distributed ubiquitous AI infrastructure that's serving all of the inference needs that are powering the economy. It's way more resilient. You can fail over to many different locations. And we think it's just a better, secure, distributed infrastructure solution.
Speaker 1: That's great.
Speaker 3: Last question. You can answer in a minute. I think we have to jump for another event. Obviously, you're focused on cybersecurity. Every inference provider needs to be right now. We've seen a number of different hacks within the category. How are you thinking, though, specifically about your software supply chain? I think we're seeing more reason now for people to build their own versions of some products because you don't want to rely on an external vendor for something because you don't know if they're going to have the same standards that that you would.
Speaker 13: Yeah. I mean, you know, we we we often take the build it ourself approach. It's just like ingrained in our DNA. It's like we make our own electrical switch gear. We're not afraid to make like
Speaker 6: software. Certain
Speaker 13: software products. But security is something we're very focused on right now. I mean, all of the ongoing hacks and all of the things happening that are disrupting software systems is definitely like a huge area of focus and investment for us to harden our systems to try to get ahead of any any future issues. And part of that's evaluating some of our software suppliers.
Speaker 1: Love it. Well, good luck out there. Thank you so much for coming on the show. Great to
Speaker 6: see you. Yeah.
Speaker 1: Have a great rest of TomorrowX.
Speaker 11: Yeah. Yeah.
Speaker 1: And that is our show. We will see you tomorrow. Thank you so much for tuning into TBPN at TomorrowX. Leave us five stars on the podcast and Spotify. Sign up for our newsletter, tbpn.com.
Speaker 3: I love that you're hanging.
Speaker 7: We'll see
Speaker 1: you tomorrow. Goodbye.
Speaker 3: See you guys. Goodbye.