‘This is the internet build-out in the late ‘90s and early 2000s on steroids,’ Applied Digital CEO Wes Cummins tells CRN. ‘It’s just happening much, much faster … That was laying cables, and this is building the structures that the cabling connects. So it’s power-driven because using compute equals using power.’
As the AI gold rush takes off, billions of dollars are being poured into the construction of data centers across the country and with early investments in land, power — including a two-gigawatt wind farm in North Dakota — Applied Digital CEO Wes Cummins has set the table for a feast in 2025.
“This is the internet build-out in the late ‘90s and early 2000s on steroids,” Cummins told CRN. “It’s just happening much, much faster … That was laying cables, and this is building the structures that the cabling connects. So it’s power-driven because using compute equals using power.”
During its most recent quarter reported this week, the Dallas-based Nvidia Elite partner said revenue grew 51 percent year over year to $63.9 million. That growth was driven by the expansion of the company’s cloud services business. The company reported a net loss for the second quarter of $139.4 million or $0.66 per share as Applied Digital finished construction of the first of three facilities in North Dakota. The firm’s stock is up 10 percent since the Jan. 14 earnings announcement and nearly 40 percent in the last six months.
The company also won $5 billion in funding from Macquarie Asset Management, with up to $900 million being directed to fund Applied Digital’s Ellendale HPC data center campus, its high-performance computing facility that is currently under construction in North Dakota. Another $4.1 billion of those funds is reserved for Applied Digital data center pipeline.
Cummins said the investment took about seven months to come together.
“I’m very happy with the terms of this structure with Macquarie,” Cummins said when asked about the deal during the company’s earnings call on Tuesday. “But we did a lot of diligence to make sure we were getting the best partner, and the best terms.”
Cummins said the investment is part of the “last piece of the puzzle” for Applied Digital to completing its Ellendale, N.D. facility, which turned on the site’s main substation transformer for the first time last quarter, a milestone to opening the 100-megawatt facility and eventually increasing its capacity to 400 megawatts.
Cummins told investors on the company’s earnings call is in late-stage negotiations with multiple hyperscale customers for a leasing agreement at that facility.
“I don’t know of anything of this scale available in 2025,” Cummins said during the call. “I’m really optimistic about signing a lease on that facility and moving on to the next one.”
Applied Digital said its cloud services business generated $27.7 million during the quarter, about $23.2 million more than a year ago as demand for high performance computing power took off. Meanwhile, its data center hosting business is running at capacity across two facilities powering 286 megawatts.
CRN spoke with Cummins ahead of Applied Digital’s earnings about the demand signals in the data center space, the progress of the company’s massive North Dakota build-out and the opportunities he sees in 2025.
Demand for data center seems to be reaching another high this year. There’s a bunch of investment coming to the data center space, including the DAMAC announcement. We talked last year about some of the supply chain constraints. It would be great to hear what’s top of mind for you this year.
I would say when you watch demand from my seat in real time, you get these ebbs and flows. The demand is definitely up and to the right. It goes from strong demand to there feels like there’s panic buying.
That happened last year multiple times, where all of a sudden just within four days of each other, we get pinged by three very potentially large customers that are looking for X amount of capacity, and they needed it by this date.
Microsoft was out with a blog post saying they plan $80 billion in data center spend in 2025. I saw an interview with someone from AWS, they’re going to spend $60 billion in 2025. So just those two by themselves, that’s probably as big as the data center market was, four years ago, the entire market. And so that’s just them.
You had all their traditional data center infrastructure investors that have been around the hoop for a decade or more. And now you’re seeing everyone trying to get into the space. So I’ve spent a lot of my time educating people about this space, because they just know that there’s a lot of money being put to work and it’s this big infrastructure revolution.
This is the internet build-out in the late ‘90s and early 2000s on steroids. It’s just happening much, much faster.
If you remember, it was all about connectivity. It was connectivity build. There was a fight over whether DSL or cable modems were going to win but either way, there was a huge amount of infrastructure built. That was laying cables, and this is building the structures that the cabling connects. So it’s power-driven, because using compute equals using power. So power is still the big bottleneck, and I think it just keeps getting worse.
And then after that, you have supply chain constraints. In my opinion, we’ve moved on from in 2023 and the early part of 2024 the supply chain constraint was primarily GPUs, and now you’ve moved more to where do I put those GPUs?
That’s the piece that was so clear to me in 2023, when we started going hard at building these types of data centers.
Look, they’ll fix wafer throughput. They’ll fix advanced packaging, like co-ops packaging.
They’ll fix those types of bottlenecks much faster than we’ll fix, ‘How do we get more power generation?’ ‘How do we get permitting, land, buildings built, all the equipment that needs to go inside of data centers built?’
I thought that would become the bottleneck, and that’s really what we’ve seen.
How are you approaching some of the constraints? How are you finding success there, whether it’s power, whether it’s supply or whether it’s real estate?
What we’ve done is we have focused on stranded power in North America. And so stranded power is created generally in two ways in North America. One, is a really power-hungry industry goes out of business. So think of an aluminum smelter, steel mill goes out of business, and leaves all this power infrastructure behind and doesn’t really have any use for it.
So if you look at the big bitcoin campus in the center of Texas, I believe that was an old Alcoa Aluminum plant that had all this power-gen and infrastructure that went away.
The other way is, typically through renewable energy. The incentive programs in the U.S., especially around renewables, is you’re generating renewable energy credits, RECs. And so if you’re a wind farm developer, ours happens to be mostly wind. So I use that example, if you’re a wind farm developer, you want to go for the most bang for the buck on generating renewable energy credits. You want to go where the wind blows the most, and the land is the cheapest, and so the center of the country has a lot of this.
North Dakota where we’re building our big campus now. That substation is surrounded by about two gigawatts of wind power, and it’s much faster to permit and build generation versus transmission. So you can permit and build (power) generation in 18 to 36 months. New transmission takes 12 to 15 years to build.
So you end up with locations like ours, where there’s more power generated than can be used. And so hence the term stranded power. And so we have several of those locations, and that’s the route we have went. Now, I think that’s a short-term solution, right? We’ll build our capacity up. We’ll lease it out. But really, when you look into the future, to feed this demand, you need new generation.
And, I think the gap in between is mostly going to be filled by a lot of natural gas. So if we compare natural gas with more renewables. Natural gas, you can take up and down pretty easily to pair nicely with renewables, because renewables generate either when the wind is blowing or the sun is shining, and so if you can match those together.
But, over a long period of time, you’ve seen a lot around nuclear power. You’ve seen contracts with Microsoft. You’ve seen announcements from Amazon. You’ve seen a lot with the hyperscalers around nuclear power. It’s just that it’s going to be slow and take a while.
But I think that’s kind of the solution, if you’re looking seven-plus years out.
When you hear that $20 billion is going to be spent on data center — between Texas, Louisiana, Indiana, Illinois — does that affect kind of one way or the other what Applied Digital does or your strategy for the year?
For us, our strategy is pretty locked in for the year. We have our power that we have locked in. We’re building our facility at Ellendale. We expect to start building more in 2025.
But when you look at the market at large, the data center capacity available for 2025 is largely baked in at this point.
It’s really 2026 planning, and then out into 2027 and beyond planning at this point. So our strategy for this year is staying on time and on budget for what we’re building for ‘25 capacity, and then starting to lock in ‘26 capacity from a customer perspective.
We’ve done a good job of locking in supply chain as well. Supply chain is really difficult, as you can imagine. So backup generators’ really long, lead times, switch gear, UPS, step-down transformers, chillers — all of these things that you need in a data center have extraordinarily long lead times, because of the unprecedented demand we’re seeing.
Dell and Lenovo have both introduced liquid-cooled servers in the last year that promise to remove much of, if not nearly all of, the heat from the compute. With the power constraints in the data center, are you seeing demand for those?
I think that’s going to be one of the biggest stories of ’25: liquid cool at scale. Because liquid cool has been done for a long time, but not at the scale of the deployments that we’re looking at for with AI.
So we’re going to see a lot of stories about it this year, good and bad. There’s going to be some snafus with deploying at scale. There’s going to be some great success stories.
So our Ellendale facility – a 100-megawatt, critical, IT load building that you can see videos of in our social media — is liquid to liquid. We call waterless liquid to liquid. We’re able to do that in North Dakota because of the climate.
I think if you look at some new stories around Microsoft specifically, as far as hyperscalers go, they’re trying to move to a waterless liquid-to-liquid solution. But I think you’re going to see a lot of that go out this year.
There’s all of the plumbing that we do building the liquid cool data center that has to happen. And then there’s a decision point for our customers where they actually choose which type of rack they’re going to use for their liquid cooling.
And so there’s a lot of different vendors for those solutions. We won’t actually choose that. That’ll be our customers that do that, but it’s our job to get basically all the way down to the CPU, and then deploying whatever solution that they choose, but you’re gonna see a lot of that this year.
On the supply chain side, do you foresee products like the GB200 easing the availability of Nvidia’s last generation of chip, or is there always going to be a flight to the highest quality chip?
We’re in this part of the market. Everyone’s really focused on Blackwell right now and getting Blackwell, and there’s not a huge amount of supply that I’ve seen. You don’t have to take it from me, you can just read plenty of news articles about the supply of Blackwell.
Nvidia is a great company, so they’ll get there on the supply but you’re seeing a lot of focus on that from the market as well. And the reason is while the chips use more power, you’re actually getting much more compute. I think it’s like 3x the amount of compute on a per-watt basis from the chip.
So it lowers the cost overall to run. As these chips get more and more efficient, even though you’re going to use more power in a dense area, you’re getting much more compute than the extra power. You’re getting, like I said, I think I saw recently, it’s about 3x the amount of flops that you’re getting out of those on a watt-for-watt basis.
People are going to move towards that because it lowers costs, right? If you’re getting 3x the amount of compute for a similar price, and it uses a third of the power, even though the power density is going up, I think there’s going to be a push to move towards those newer chips. Because it’s expensive, right? The AI infrastructure piece is expensive. The chips are expensive. The data centers are expensive. And the more you can push down, or the more efficiency you can get out of power usage, the more efficiently you can operate the data center.
In terms of chip rivalries, Broadcom gets brought up a lot these days. How serious a threat is Broadcom to the big GPU makers?
There’s a lot of great chip companies out there, Broadcom being one of the best. And so you’re going to see this. You’ve seen it in many markets before, right? It typically takes longer than you think for people to get displaced.
There was a time when you thought the BlackBerry would never get replaced, right? And it lasted for a long time. But when markets get big enough, and enough capital flows in, eventually there’s going to be competing solutions.
What I will always say with Nvidia is that I think people focus way too much on the GPU with Nvidia, and what Nvidia has done, in a spectacular way, is create a system. That’s the harder part, I think, to replicate, versus just, ‘Can someone else build a chip?’
So Nvidia has the GPU, but they have the networking, which is super important. Like when they announced Blackwell, I thought one of the biggest advancements there, obviously, you’re getting many more flops per kilowatt or per watt going in on those GPUs. But the way it works is a system with the networking and then the software layered on top. It creates a really powerful mode for Nvidia, and that’s why you see them, I think, get the valuation. That’s why you see them just continue to grow rapidly.
There absolutely are already competitors today, and are there going to be more and more? Absolutely. But there’s just this really big moat around the system, not just the chip. So I wouldn’t just focus on a chipmaker, and then what the market is going to try to do overall. There’s a lot of work around Ethernet to try to mimic what InfiniBand does to capture the networking portion. But then you still have the software piece that layers on top, right? CUDA is a very powerful platform.
Wes, what’s this year going to be like for Applied Digital? I’d love to hear what is on the horizon for 2025.
I think it’s going to be great for the market. It’s always interesting to watch, because we saw a big demand spike in September. Everyone came back from summer and from Labor Day, and then it was like time to get back to work. And we’re kind of seeing that again.
Now, post-holidays, everyone’s going back to work, and it’s like, ‘OK, we need a lot more capacity.’ But all the demand signals have been really robust to start the year. I’m optimistic about the year ahead, not just for us, but the industry at large.