I'm not sure if we're still in the morning or the beginning of the afternoon, but it's fantastic. It's the middle of the night in California, so it's cool. Cool. Cool. So maybe Chase, you want to give us like a quick introduction word on Crusoe? Obviously, I'm sure 99% of you know Crusoe, know Chase, but I think it's kind of interesting to from your own words to describe what is Crusoe and how you have evolved in the last 12 months.
Sure. So, you know, Crusoe is a vertically integrated AI infrastructure platform that really is an enablement layer to, you know, what's happening across the AI ecosystem. And, you know, we there's a couple of key aspects of vertical integration that we, you know, we do at Crusoe, ranging from, you know, at the very, you know, base layer, you know, land, power, and energy development that ultimately is sort of the core input to going from, you know, the electron to the token. You know, above that is sort of like data center development and these next generation data center designs that are enabling these very large clusters for both training and inference, you know, building large campuses like Abilene, Texas that has, you know, become, you know, one of our more famous, you know, projects out there, to, you know, our high performance managed AI cloud platform, Crusoe Cloud, that is really where, you know, we're able to deploy and manage these large clusters of GPUs and enable AI innovators to, you know, train and operate their workloads. And then we have a developer tool set that, you know, is really our managed AI services to give people critical endpoints for things like our managed inference product and really enable them to scale those applications in a cost-effective and high performance capacity.
Great. And maybe if we take like a step back, Crusoe started with a very contrarian bet and basically bringing the computer to the energy, not the energy to the computer. What did you understand about the physics of compute and the physics of producing intelligence at scale before everyone else?
Yeah, I think, you know, our philosophy has always been taking this energy-first approach to developing compute infrastructure. It's actually, you know, why we named the company Crusoe. It was named after the character Robinson Crusoe who was stranded on a desert island and had to be innovative with his resources in order to survive. You know, we sort of try to channel that same sense of innovation towards unlocking value in stranded energy with computing. And, you know, the way we sort of saw it early on was that, you know, you sort of had these big hubs of data centers, right? You had, you know, this whole Northern Virginia corridor that had been built out for many decades and was fundamentally powering internet applications. And, you know, it was important to be in that corridor because latency really mattered in terms of how you were serving those internet users and those internet applications. When you move to more computationally complex workloads, you know, things like AI, you know, training models and actually serving models where you actually have more time for feed forward, you know, through the network, the latency to get to the data center is actually a very, very small percentage of the overall time of compute. And so that opened up a lot of new geographies. And, you know, if you look at the scale of AI infrastructure, it's actually dwarfing, you know, this historical data center hub of Northern Virginia that had been built over many decades. So, you know, putting some numbers in place, you know, end of '24, you know, this is kind of like early on into the AI boom, the total data center capacity in Northern Virginia was, you know, roughly 4.5 gigawatts. You know, today Crusoe, you know, we announced recently that we've contracted more than 5 gigawatts, you know, just for our own platform and we have more than 40 gigawatts in the development pipeline. So, you know, we as a single company have been building, you know, a Northern Virginia worth of infrastructure. And so, you know, I think that the physics of this is that, you know, there just aren't that many, you know, ways to get that much power into the Northern Virginia corridor. So, it actually you need to by design open up new energy sources and new energy technologies to actually energize the compute needed to power innovation.
Super insightful. And for everyone in the audience, I invite you to watch. I think it's an exceptional documentary on your data center in Abilene, Texas. I think it really gives you a tangible understanding of what massive AI-ready infrastructure looks like in the 21st century and basically how can we use intelligence at scale in our daily life within our enterprise or whatever organization. Because often it's pretty abstract but that project, can you like walk us through like how many months it took to have the first megawatt online?
Yeah. So, I think it's a very cool piece of the stack because, you know, people using AI, there are like these inherently digitally native experiences, right? You're using on your phone, you're using on your laptop, you're maybe building an application on cloud, and, you know, there are these like digitally native experiences but what's happening on the back end is it's running in these very energy-intensive, you know, large-scale data center projects that need to be built and built by people and they're big construction projects. They're big infrastructure investments. They require a lot of capital, a lot of people, and a lot of resources to bring to life. And, you know, I think the demand that we're seeing across the ecosystem is such that, you know, people want this infrastructure yesterday, right? It's like the demand curve has been, you know, truly exponential and, you know, that's put a lot of pressure on the industry to be able to deliver at speed and scale that's, you know, completely unprecedented. You know, so for us, you know, we tried to, you know, we had this advantage of being a bit of an outsider and like a bit of a newcomer to the ecosystem that really enables us to, you know, sort of leapfrog some of the incumbents in terms of, you know, how we would do this and what speed we would do it at and what scale we would do it at. And, you know, so when you look at the Abilene project, we started construction, you know, June of 2024 and we actually delivered the first two buildings in basically 11 months, or, you know, the first megawatts were online in 11 months and, you know, there was a competitive process in that initial phase and I think the next fastest bid was two and a half years. So, you know, if you look at it from, you know, what's being demanded from the ecosystem, I think, you know, speed and scale are like the two big vectors here.
Just before we were on stage we were having the conversation with Chase and Chase basically, the DNA and the Crusoe is owning from electrons to tokens. Why is that the winning model?
So, you know, we think about it, you know, from a vertically integrated perspective because you sort of have to. It's like, you know, what we're building are these like data center scale computers and really being able to understand the end use case and the end application enables us to build better data centers and because we're really good at building data centers that enables us to deliver a better cloud platform that's more purpose-built and high performance for our end customers. And, you know, I think in the last like 24 months I think people have come to appreciate how important, like data centers were kind of a boring category. I think people have come to appreciate how important the data center is to the actual scaling of the infrastructure and a lot of that has to do with, you know, the rapidly changing chip architectures that are increasingly high density that, you know, have different cooling criteria than, you know, historically people have used. You know, liquid cooling I think we all take for granted today, you know, after sort of the deployment of Blackwell and this big shift to liquid cooling and, you know, but historically like liquid cooling has been like this very niche piece of computing and, you know, that was like, you know, deployed at small scale for kind of like these small experimental, you know, supercomputing clusters but, you know, now we're seeing that be the lion's share and that sort of the future of computing infrastructure is having these large scale liquid cool deployments.
And so Chase, you were here last year and you just told me again backstage that that's where you closed the deal with Atero, which company you acquired. Maybe you want to give us like a bit more insight on the acquisition and why it was like pivotal for your inference and orchestration layer.
Yeah. Raise is a very special summit for us, you know, because of that actually. So, you know, last year, actually up on a balcony over here that most of you probably walked past, there was like this back door area where me and Alain, the founder of Atero, were actually negotiating and coming to final terms on this acquisition. And for us, you know, the way we were thinking through this acquisition, they've been, you know, an incredible team to welcome on board and, you know, we're sort of building and hiring a lot more with our team in Tel Aviv. But, you know, where it's been incredibly instrumental is that we had sort of made this bet a few years ago that we were going to bring online an inference platform. And, you know, where that stemmed from was the fact that, you know, as a business, we sort of think about our platform as like, hey, we're like long infrastructure, you know, we're long energy, we're long data centers, we're long chips. And the more ways, you know, as a shareholder of the company that we have of turning those investments into value and those investments into value for our customers and money for our shareholders, the better off we're going to be through like multiple different market cycles. So, you know, about two years ago we sort of made this decision to, you know, invest in a managed inference offering that was a different way of sort of monetizing these investments and chips that we had in terms of, you know, turning the chips into intelligence and chips into value for our customers. Where Atero really came into play was, you know, one of the key things that we were focused on and one of the key aspects of inference is actually how you manage memory and how you actually manage the KV cache across different layers of memory and, you know, that had been something that Atero had been very, very focused on as a business and as an early stage startup and so bringing that team together with Crusoe, we've been able to, you know, embed this technology we call Crusoe Memory Alloy, that's our sort of distributed memory caching layer that stitches together both HBM, DRAM, NVMe, object storage and we're able to actually effectively distribute the KV cache across different layers of memory and help accelerate a lot of the inference workloads and, you know, we're kind of helping serve some of the leading application companies that are scaling inference today and creating a lot of value for them by being able to produce tokens in a more efficient way for the infrastructure underlying it. So, the Atero team's been awesome and, you know, I have you to thank for, you know, hosting this great event where I was able to sort of finalize that transaction.
