How Applied Digital Became the Landlord for Big Tech’s AI Buildout

Thanks for having me. I appreciate it.

Sure. So, we started the company in 2021. What I would say about Applied Digital is we've never been a Bitcoin miner. We were always a data center company. So when we started in 2021, what we did was when China cracked down on crypto, we saw the opportunity to build data center capacity for Bitcoin miners in the United States. So we built, over a two-year period, we built 500 megawatts of data center capacity, but we didn't mine the Bitcoin. We actually leased it out just like a data center model would. We leased it out to Bitcoin miners, Marathon being our largest customer there. So, we signed a big contract with Marathon, built out, helped them become the largest miner in the world through that lease of our data center capacity where we ran their equipment for them. And then a year into that, so we're still building it out, in 2022, we decided to build high performance computing on the same campuses. That was the idea, is build high performance computing. So, during 2022, we hired some guys. They were previously building data centers for Meta, and we designed a high performance computing data center that we started building in Jamestown, North Dakota. And at the time the market was very different than it is today. So we designed a 10 megawatt build in Jamestown, North Dakota. It was purpose-built for high performance computing. And if you remember, AI wasn't the workload then at all. It was a niche market, and so it was a much smaller build, but we thought it would be a very nice addition onto our Bitcoin data centers. So we started building that in 2022. By the end of 2022 we had GPUs running out of our facility in Jamestown. And at the time we were just proving out that the data center capacity worked. And our customers were mostly universities, so people who were paying very, very little for the GPUs, but again we weren't trying to sell the GPU capacity, we were just trying to show that the data center itself worked. And then ChatGPT was introduced in December of 2022, no big change for us, but when the Nvidia H100 was introduced in March of 2023, we were out marketing that data center capacity, and what we had built in Jamestown actually was a good fit for the H100. And so I was introduced to some AI labs, some frontier models that had demand for these GPUs. They didn't want to deal with data center capacity. They just wanted GPU capacity. They just wanted to use the GPU. So we actually ended up purchasing GPUs, deploying them. We had one of the largest clusters of H100s running in the first week of July of 2023. And through that process, as we started speaking more and more to these companies, I started to see the demand that was there. We went into the market, looked for data center capacity. We ended up leasing data center capacity from three different data center companies, and we were deploying GPUs. But what I realized pretty quickly was there was not very much data center capacity. And so what we decided at that point, I went back to my team, my data center design team, and I said, "Hey guys, I know we're talking about 10, maybe 15 megawatt builds. We need to go back to building hundreds of megawatts of data center capacity." So during the summer of 2023, we designed a 100 megawatt facility. We worked with Nvidia's data center design team. We worked with the Crayon over at HPE because we were trying to figure out, hey, what do we design? And what we designed to was, it wasn't called Blackwell at the time, but it ended up being named Blackwell. It had a numbered name if I remember correctly. And so we designed to that, we broke ground in October of 2023. And my idea at that time was we want to serve hyperscalers. How do we get to the serving of hyperscalers? I think they're going to need this capacity. If we build the capacity and when they need it, we have it, we'll get into that business. And that's really what has happened. So we spent 2023, 2024, and some of 2025 building that capacity. We ended up leasing that campus to CoreWeave. We've started another campus with an investment grade hyperscaler. Now we have a third campus with another investment grade hyperscaler. And what we've done in that time is we've just really, really refined our process. So we changed design, we refined the design, we refined our manufacturing, we refined how we deliver this capacity. And so we've spent the last three years really tuning up what we call a franchise model where we can just repeat this build across many different sites around the country. We have our core team in the company and then we hire a project manager at each site, and that's a key hire for us, but something that we can scale and make it repeatable. So, as we were building all of this, there was a ton of stress around getting that CoreWeave lease, but we were already into contracting long-term for our supply chain. How do we scale the build because we thought it would just keep going, and it has. But that's really the origin of the company and how we made our way into the AI data center market. We really were, you know, one of the first, if not the first, building extremely large scale, and we did it on spec. It was super stressful as you could imagine. It was the 19 most stressful months of my life from starting that build until we actually signed the contract. But we've kind of been in the right place at the right time, and now we're scaling it in a big way.

Yeah. And that's exactly right. And what my thought was at the time when we really started leaning in on data center capacity was they're going to fix, you know, and I have a long history of being a tech investor. But I said Nvidia is going to fix the wafer throughput issue, the advanced packaging issue, because GPUs, when we started this process, GPUs were really the thing you couldn't get a hold of. And I said eventually they're going to fix all of that, and you're going to have plenty of supply of GPUs and other hardware. Because those supply chains know how to adjust really quickly. They've been through many, many, many cycles. But what we're not going to be able to fix once the GPUs are fixed is data center capacity. You need the permitting, you need the power, you need the land, you need to build it. You know, that takes a long time. So I thought to myself, while the GPUs were the bottleneck in 2023, if I look out towards the end of 2024, 2025, 2026, data center capacity is going to become the bottleneck. So that's why we decided to lean in big on building large scale data center capacity.

Yeah. So, I call us a highly technical landlord. And so what we do is we have our sites, so we secure the power, we get the permitting, and then typically we'll start to do some of the build process, right, dirt work, get the foundations moving, all of that, pretty low cost. And then what we like at this point, you know, I built that first data center on spec. I don't want to do that again. But what we do is get the site build-ready and maybe do a little bit of the work so that it's a really fast process, and then we show it to the hyperscale customers. And we really focus on seven customers at this point, and that's really all of our focus. So it's, you know, it's Microsoft, it's Meta, it's Google, it's Amazon, Oracle, CoreWeave, and Nvidia, right? And Nvidia has been in the market recently for data center capacity. So we focus on those, and we are, you know, qualified with the vast majority of those technically. We've been through all the processes with them. So we just show them what we have. Hey, here's, I can build this, first capacity shows up at this date, and then full capacity by this date. And then they'll say, "Yeah, that area of the country, that location is interesting, that size is interesting, that timing is interesting." And if those things match, then we'll go through a diligence process. And then eventually, you know, we'll sign a lease. And then these leases that we've signed, so we've done three campuses so far. They're all 15-year take-or-pay leases, so guaranteed leases. So it gives us a really long visibility. Now, the buildings themselves, those are 50-year assets, right? So the initial 15-year lease gives us a great return, but I expect, you know, to continue. You look around the world, you don't see data centers sitting empty at this point, right? We've never really seen that in the history of data centers. So I expect that to continue beyond that 15 years, but to have that type of visibility and security in our business is a great business model that I love. So, you know, I grew up investing in, like I said, in semiconductors and hardware, and so I know that industry really well. So I'm accustomed to big cycles then. So when the downturn comes, the companies, their revenue goes down significantly, and you wait for the next cycle up, which is historically what happens. So our business model, on the other hand, when the downturn comes, if it comes eventually, what happens in our business, our revenue doesn't go down. It could go to as low as flat growth, right? We won't grow year-over-year, but you go flat, and then you wait for the cycle to come up, and then you start to grow revenue again. So, I love that business model for us. And then we typically do maintenance and operations on the data center. So, we know how to operate. We built it. And then our customer, you know, sends the equipment. It's typically GPUs, networking, and then they'll have a few people on site because we typically don't go in the data halls. They'll do smart hands and repairs and maintenance inside the data hall. We just make sure the power is on and it's cool. And that's really our job. And we have to do that at five nines reliability. So that lets us have downtime like 23 seconds a month or something like that. So...

You know, so we build these super resilient, redundant facilities with backup power, multiple paths of power, backup cooling, all of these things because, you know, that equipment that goes inside is super expensive and it needs to run basically 100% of the time. So that's our job. But you can, I always tell people, please think of Applied Digital as being a boring business model where I just keep signing 15-year massive leases, and then we build the buildings, and then that revenue and those earnings start to flow through our income statement. You shouldn't expect anything more from me than just more of those leases and more of those campuses and more of those customers. And that's what we're doing. And we've even simplified our business where, you know, just this week we completed the spin-out into another public company of our cloud business. So that simplifies our business even more. We have an IP that we own 10% of called Base Electron, where that's going to build power generating assets. But Applied Digital itself will have ownership in those. So our shareholders benefit, but we sign data center leases, we build data centers, we deliver them, we operate them, and that's what we do.

Yeah. So, one, the cloud side of the business, terrific. I think huge amount of potential. You see CoreWeave, you see Nebius, you see what CoreWeave is doing. So I think there's a huge amount of potential there. However, two items here. It's a capital-intensive business. Our business is a capital-intensive business. So trying to do both of those just requires so much capital. So this spin-out, it's its own separate company now, has its own capital stack, can raise its own capital to grow its business, and isn't Applied Digital just constantly feeding, you know, all of this capital in. So separate capital stack, big, big improvement for us and I think for our shareholders. Two, a lot of the customers that we look to have or potential customers, so all of the ones that I mentioned, some of those, we've went through the process, some of those have a sensitivity to us having a business that looks like their business, right? So they have a cloud business, right? So AWS, Microsoft has Azure, right? They contract GPUs and TPUs and all of that, but they also have that business. So this keeps us from having that conflict when we're trying to sign data center leases with them. Sometimes that is viewed as a conflict. Now that's gone. So the conflict is gone. The capital stack is there. Applied Digital right now owns 97% of Kronoscale. So, our shareholders are going to benefit from the growth that I think is there. And the reason we did spin it out is because I think there's a massive opportunity in that market, and that business can grow in a huge way outside of Applied Digital and maybe be a data center customer for Applied Digital. So there's a big, big opportunity in my opinion for the Kronoscale business. Applied Digital shareholders get to participate in a big way because we own the majority of Kronoscale. But it lets it go and grow. We've hired new management there. I'm super excited about the talent that we're attracting. We're going to announce some more people that are joining Kronoscale. We had a really good experience on attracting really, really high-quality executives to come to that company, and I'm super optimistic on the growth there.

The best way to solve those is to do it two years ago. And so that's what we did is when we were going through that, the first build, what we did, we redesigned, we really slimmed down the amount of critical vendors that we use really to just a handful or five. And so that streamlined our supply chain. And then what we did with those vendors is again, I'm like, hey, this is, I think this is going to be a big market. And so we went and bought out capacity of their factories. So I bought out all the capacity of my key component suppliers of their factories. In some cases, we've really helped them with ramping up their manufacturing of some of the things that we need. But we did this two years ago when the market wasn't like it is right now. And so we've locked in that supply chain, what we call MEP, mechanical, electrical, and plumbing. That's the key of the supply chain. So that's all of our, you know, medium voltage switchgear, step-down transformers, chillers, heat rejection, all of those items. So locking that in a couple of years ago is a big advantage for us. And then when you look through the rest of it, so power is always going to remain, you know, in my opinion, an issue. We started an IP that Applied Digital shareholders participate in as well to go build power generation. What that's going to do for us is we have the grid power that we've already contracted, and that'll let us build through, you know, 2028, like in North Dakota. But those campuses have the ability from a physical delivery perspective, the grid perspective, to deliver a lot more electrons, but we need to make those electrons. North Dakota has a huge amount of oil and gas in the Bakken. I like to refer to it as cold Texas. Sometimes we have a really similar setup. So we're going to build, we're going to do power generation. We started down that path last year. We've really ramped that up. We've purchased a lot of steam boilers that'll be delivered in 2028 from Babcock & Wilcox. And, you know, expect to just continue to expand that because that's one of the bottlenecks. And then, you know, the harder things that have started to happen is, you know, you're starting to see this intense blowback around data centers and, you know, AI data centers. Yeah. And just getting communities on board with the benefits that we bring. And what we do, you know, we were ahead of this. Again, I'm not trying to pat myself on the back of being ahead of these things, but a lot of the more traditional data center companies, they've never dealt with this type of opposition. I don't know how familiar you are with the Bitcoin market, but, you know, of course it was never open arms welcoming us into communities for Bitcoin data centers, right? It just never was. So we have a lot of experience in dealing with this and educating the communities in which we want to build projects on what the benefits are. There are definitely, you know, negatives. The negative with what our projects are, the biggest negative is the construction process, right? It creates a lot of traffic and there's some noise potentially. But also where we site these, back to the kind of the Bitcoin data center roots, you know, we had a lot of experience in siting these in places where we don't disrupt people, where noises don't disrupt people. So that, you know, I've purchased soybean fields, I've purchased cotton fields, right? And we build these out where there's not population for a reason. And it's because the power is there. And then also we don't want to be disruptive in thousands and thousands of people's lives. So we're not building data centers in Dallas. We're not building more data centers in Northern Virginia, right? We're not building these where you're just trying to get incremental power on the grid. We're building them in places where there's power that goes unused. You know, Ellendale, we have the perfect case study for this in Ellendale, North Dakota. We built our first campus. There's 2 gigawatts of wind power that feed into a substation. We co-located right next to that substation. The utility doesn't have to build really any new infrastructure. Anything that they do have to build, we pay for so that it doesn't go into the rate base and increase rates for the ratepayers. The electricity users outside of us pay for everything. Then we use a lot of electricity where that wind feeds in, and a lot of times they would just have to curtail those wind turbines because there was not enough transmission capacity to take that energy to someone to use it. So we're using power that wasn't being used. What is the net result of that? We have saved the ratepayers in the region like 35 plus million dollars over the past four years because the utility gets to charge us for power that wasn't being used. We pay for any new infrastructure, and then they have to share some of that earnings back to the ratepayers, and so the ratepayers actually see their electricity bills go down and not up. And that's all in how you choose the site where you go. The community there loves us. We're building, you know, at an out-of-the-way area. But they're seeing all the economic benefit. They're seeing all the tax benefit. They're seeing all of the big benefits that we bring. And then you take, you know, what was a slowly shrinking town into growth, and not too much growth, right? There's the growth from the construction, but then at the end we train locally, you know, 80, 85% of the jobs. But it's two or 300 jobs for that town is 1,100 people. So you're not overwhelming them with a bunch of jobs, but you bring jobs that pay two and a half to three times on average the average household income of North Dakota, highly technical jobs that we can train for locally. So it's all these are the things that we're doing because that's become the biggest hurdle. But it's not a new hurdle for us, and it's something we've, you know, been doing since the inception of the company. But it's something new for a lot of data center companies. They used to get to fly under the radar and never have, you know, any issues. They just apply for permits and the data center gets built, and they ask for more electricity from the grid, like I said, in these key data center markets. So we've had to go about it in a different way. But I think that's going to be a big benefit to us. It already has been, but it's going to really be a benefit in the future. As you know, data centers are facing kind of the same blowback, right, that Bitcoin mining was facing back in the day.