We're continuing our series of interviews here for the All-In podcast. And we're very lucky to have Chase Lochmiller with us. He's the CEO and co-founder of Crusoe Cloud. C-R-U-S-O-E, not Kuso, Crusoe Cloud.
Which is or was the first NeoCloud really. You were kind of...
We were part of an early contingent of NeoClouds, you know, between us and CoreWeave and Lambda.
How did that occur? Because you had anticipated buying a bunch of H100s before everybody else and you had the inventory? How did that start that you got such a good lead?
Well, so we had started well before Hopper came out. You know, Crusoe is a business that's always taken this very energy-first approach to developing computing infrastructure and really changing the motion from trying to co-locate compute in these network hubs and really focusing on where we can access abundant energy. So, you know, we were experimenting a lot with the Ampere generation which preceded Hopper, so a lot of A100s, and we built out this high-performance cloud platform that was meant to enable AI innovators to do incredible work. We launched before ChatGPT came out and so sometimes you're in the right place at the right time and...
And your biggest customer by far is Oracle and OpenAI.
Correct. So, you know, we're a vertically integrated business which means we're focused on not just the compute layer but also the energy development, the data center development, as well as the application layer where we're running things like our managed inference service. And as this boom really took off, the scarcity of data centers became very apparent and our ability to build those large-scale AI data centers very quickly became a very in-demand skill set.
So why West Texas? Is it because the politicians there are pretty permissive in terms of giving permits and you can build quickly like Elon experienced with his big factory in Austin, or is it because of energy?
Yeah, it was really an energy-driven decision. So, in Abilene, Texas, we had... it's an area where a lot of actual renewables had been built out on the back of production tax credits. So, there was an abundant amount of wind and solar that was there that actually had issues with transmission. So, power prices were actually frequently negative. There were renewable energy producers that were having to curtail that could be producing power but actually were shutting down because there was no marginal demand.
So the grid didn't need their energy.
That's wild to think about. It's also wild to think about that Texas has the largest solar base of any state in the union.
Yeah, absolutely. And, you know, we went there and we said, 'Man, you guys have too much energy. Hey buddy, I have an idea for you.' And so we have a 1.2 gigawatt substation there. We've also built a 350-megawatt gas plant on site to energize one of the largest clusters of GPUs in the world.
So, was there gas under the actual data center or are they shipping it?
No, no, no. We had access to a pipeline that feeds into 10 gas turbines that are on site there.
Amazing. And the gas turbines, those were a blocker for a while too, huh?
They still are. I mean, gas turbines are a massive supply chain constraint as people look to energize compute infrastructure. I think a lot of people... we were really one of the first groups to be focused on natural gas-powered data centers, doing things behind the meter, off-grid. And I think a lot of people have sort of followed this pathway which has created a lot of supply chain challenges with some of the major producers, folks like GE Vernova, Caterpillar has a company called Solar that we work with very closely, Siemens, Mitsubishi. And we actually recently did something with an incredible company, Boom Supersonic.
Yes, Boom Supersonic was making the Concorde replacement, Blake...
And they're making their own engines that go pretty fast. And so what was the idea there? Yeah. So, you know, they had re-engineered this turbine for supersonic jet travel and, you know, Colleen, my co-founder and I were talking to Blake and said, 'Hey, you know, could we use that to generate power instead of, you know, transport people over the ocean at supersonic speeds?' And now we're their first large purchase order for $1.2 billion of gas turbines to power AI-critical AI infrastructure.
It's wild to think about. The gas turbines and jet engines are not super dissimilar.
No, I mean I think this is the full playbook of GE historically, right? I mean, the reason GE Vernova exists is their leadership in terms of air transportation as well as power generation.
Yeah. And I wonder if it's really interesting about that opportunistic thing for Blake to do from Boom Supersonic is that now that will allow him to fund his Concorde replacement.
Yeah, absolutely. You know, power is a great business and, you know, we think he can make a lot of money as the scaling up of AI infrastructure occurs. And hopefully, you know, he's able to build those incredible supersonic jets that get us to Tokyo much faster.
And the Stargate project... is a $300 billion project. What's the number? Because there was a big announcement at the White House and there's been a little bit of, you know, wink wink, like these numbers were kind of estimates. What's the realistic footprint of this?
So, you know, I think Stargate has been a term that's been used to describe a lot of different things at this point. Initially our campus was called Stargate and then, you know, Stargate was a company for a while and then I think OpenAI sort of described it as all of their spend on compute is just sort of broadly labeled as Stargate. So, I think the number is $500 billion and this incorporates chips, data centers, energy to ultimately power this intelligent infrastructure that's running and scaling both ChatGPT and all their other core services.
One of the blockers in addition to the turbines has been electricians and construction workers.
My understanding is you're paying two to three times what they were getting paid before the data centers.
No comment on exactly what we're paying the electricians, but they're very well compensated.
Am I in the right zone that their salaries in the industry have doubled or tripled?
Look, I think it's an incredibly exciting career path for anybody looking to do work with their hands and get well... very...
Hundreds of thousands of dollars a year.
Yes. So, you know what...
I mean let's think about that for a second. No college degree. You just have to be an apprentice as an electrician and you can make hundreds of thousands of dollars a year and you need how many? Thousands? Hundreds?
Which one? Thousands or hundreds?
Thousands. So I'll get into it in a second. So, I'll give you an example. So, in Abilene today, we have 8,000 people on site every day. They're working day and night to bring this facility online as fast as possible so we can energize this intelligence.
And that's 3 hours west of Dallas.
Correct. So, it's in West Texas. You know, Abilene is a town of 120,000 people. So, you know, when you're hiring 8,000 people to work in that town, you can't source everybody locally. So, we've actually had to bring in a lot of labor from all states.
And housing, I would assume.
Yeah. Housing, you know, the market's actually fairly efficient. There's a lot of people that won't be long-term permanent workers there, that are actually, you know, they bring mobile homes, you know, they collect a stipend. It's kind of a very efficient process in that regard. But, you know, long-term, we're going to have about 2,000 permanent workers at the campus. So, there's both, you know, short-term job spike and boom in terms of demand for the construction and installation process, but there's also great long-term permanent jobs to operate the facility. You know, we have a power plant there. There's a lot of mechanics that are going to be operating this facility. But this is just one campus. We have another campus in Armstrong County, Texas, that has close to 3,000 people on site every day. We have another project outside Cheyenne, Wyoming, that we've announced we're planning to build a 10-gigawatt campus.
Wyoming is a state that is very rich in natural resources and energy. And there's been really favorable public-private partnerships in terms of making this project happen. There's a tremendous amount of gas there to sort of support what we're doing. And one of the very cool things that we're focused on is actually trying to bend the arc of energy production towards sustainable resources. So, even with the gas, Wyoming is a place where you actually have primacy on what's called Class 6 wells. They're post-combustion carbon capture and sequestration disposal wells where you can actually take the carbon that comes out of the combustion process of a natural gas turbine and you can pump it underground and permanently sequester it.
And there's actually an incentive there called 45Q where you get paid by the government to do this and it doesn't quite cover the cost of doing it. But for customers that care about the impact and the sustainability of the campus and the...
They're not putting the carbon into the atmosphere.
Exactly. They can produce power from gas in a carbon-free way, which is a really exciting value proposition.
How are batteries? I think we saw Elon bought a... and is building... he's got his own lithium refinery and he's building obviously his own batteries and they're one of the larger battery manufacturers. How do batteries play into all this currently?
There's a bunch of different ways batteries play into it. I think you sort of need batteries in these low-voltage UPS systems that sort of help operate the electrical system of the plants or...
To normalize and base load or...
Yes. Actually one of the issues that we're seeing is, you know, when you think about these large-scale AI factories, when you're deploying giant clusters of GPUs, the entire data center is acting as a single computer. It's running one single workload that's training some breakthrough foundational model. And what that results in is actually massive load fluctuations in the actual power draw because what they're doing is the chips are basically running compute cycles and then publishing the data over the network to all the other GPUs so they can sort of be in sync with one another. And this causes this massive fluctuation in overall power draw. Well, utilities hate that. It's terrible for the turbines. So, you really need a way of actually normalizing that power draw.
So, we've actually been able to solve that with a one-hour battery. That's actually a medium-voltage battery energy storage system. That's something that we've deployed. You know, there's a lot of other use cases for batteries. I'll give you another example. We're actually working with Elon's former co-founder and partner at Tesla, JB Straubel.
Oh, he's doing the recycling of them.
Correct. So, he has a business called Redwood Materials that's focused on recycling of batteries. And he has this massive supply of end-of-life EV batteries, batteries coming out of digital cameras, all these like, you know, this whole collection of batteries that are sort of at the end of their life. And so, as a demonstration to show people what we could do in terms of cost competition as well as deployment, we actually deployed a fully off-grid solar plus battery energy storage system that powers an AI data center 24 hours around the clock with a power price that's lower than the power price in Northern Virginia. And we're able to do that because you're basically taking these batteries that were in an electric vehicle and maybe the range was 300 miles and now it's 250 miles.
So, the owner trades it in and it's like, okay, this car is done, right? But there's still a lot of juice left in those batteries. There's a lot of useful life in those batteries.
So, we're able to really make use of them in a second life to power these AI data centers.
You're under massive pressure to deliver these data centers.
Um, yeah. Yeah, you could say that. Yeah. Compared to where you were like two or three years ago, maybe let's say three or four years ago, you were knocking on doors saying, 'Hey, do you need a data center?' Now your door is being knocked down.
I think there's just... I think it just really speaks to the demand for compute. I think people are constantly having this conversation of like, are we in an AI bubble? You know, I think there's just an incredible... nobody has enough compute. None of the leading labs, none of the leading application companies can get their hands on enough compute and that's just driving an incredible amount of urgency to deliver infrastructure.
Both the chips and the data center. So even this year going into 2026 you're seeing the same amount of inbound, 'Hey, we need more, we need more, we need more, what do you got?' Or was last year like people put in their big orders for the next couple years?
No. We're seeing things accelerate.
Who are the new customers because obviously Elon, Sam Altman, I'm sure AWS... I'm sure... Well, I think Google does their own. Yeah.
Google, I think... well, I don't want to speak for them directly.
This is All-In. If you want to be on All-In, you got to speak candid. Just tell us what you heard at the cocktail party.
Google does a mix of self-perform development as well as outsourcing.
So they do do some outsourcing. Yeah.
Correct. But, you know, I think all of the leading labs we're seeing a tremendous amount of demand. So, Anthropic, OpenAI, Google... Elon builds his own. Elon builds his own.
What was your take when you saw him build Colossus under that short time frame? Like your team must be like, 'That's just not possible,' and then he did it.
No, I don't think we thought that at all. Well, I think I was like really rooting for him.
No, but did he do... my understanding is he did it in half the time anybody else had ever done something like that.
It depends on how you sort of measure these things. I think Colossus was... Colossus one was sort of this unique case where he had this large-scale industrial building. He had power to the building. And really what he was doing was what I would call like the tenant fit-out, which is basically the data hall buildout of the cooling distribution units, the RPPs, the electrical systems, the hot aisle containment systems that really... and then you sort of roll racks of GPUs into these, right? So they were able to execute on that incredibly fast.
Yeah, Jensen said he'd never seen anything like it. He seemed to think it was like a unique thing that occurred in your industry.
Yeah. Yeah, I mean Elon's the GOAT of modern industrialist. So, you know, hat tip to him.
You as running... not competitive to him, but building the same things. You have to look at that and study it a bit. How did he get it done so fast? What do you think? How do you think he was able to do it so fast?
I think Elon does an incredible job of breaking down a large industrial process into a lot of subprocesses and understanding constraints and really taking a first-principled approach of how do I build things as quickly as possible? How do I parallelize things as quickly as possible?
