If, Grace, you want to put up the disclosure slide for us while I do introductions. My name is Brian Kinstlinger. I am the Director of Research at Alliance Global Partners, where I also publish research on technology stocks. Joining us today from Keel Infrastructure is their CFO, Jonathan Mir. For background, Keel's legacy business is Bitcoin mining, and it is slowly winding that business down. With its portfolio of power capacity, the company is shifting to an HPC/AI focus with the development of data centers. For investors, feel free to type in questions if you have them as we go along. If not, we will be having a Q&A here between Jonathan and I. Welcome, Jonathan.
Good morning, Brian. Thank you for having me. We appreciate it and appreciate the chance to catch up with all of the investors on the line. I will take the liberty of assuming that you all are familiar with the story. Keel started, like many of our peers some years back, eight years ago, as a Bitcoin mining company. That is an element of our past. We are winding down and decommissioning right now our remaining Bitcoin activities. We have some Bitcoin on the balance sheet. We have committed to investors to liquidate the entire Bitcoin position by the end of the year. To us, the focus of the company right now, for investors, the focus of the company right now is in our HPC/AI data center strategy going forward.
That is built around the idea that we have the most constrained element of the supply chain, which is power. At Q1, we announced that we were going to market to lease three sites, Moses Lake in Washington, Sharon in Pennsylvania, and Panther Creek in Pennsylvania. Each of these sites, which are fully zoned, have ESAs in place. That is to say, we have firm capacity and power available for our customers. That is typically the most complicated element, the most time-consuming element of a data center strategy. We solve that already for our potential customers. With that, I am going to allow Ben Gagnon, our CEO, who just joined, to go through our strategy and our objectives in the months ahead.
Ben, welcome. How are you?
Thank you, Brian, sorry I'm late. I had a couple of internet issues at the.
No need to apologize. You got a busy time right now. I'm going to ask you a question to help you. We've got a lot of investors on. Some are quite familiar with the story, and some might not be. I want to start. We hear about so many companies all the time. They've got energy capacity. They have 0 - 100 MW, 500 MW, all sorts of long-term large leases. What separates your sites? Why are they so attractive in this time where power is so important? What separates your sites, and what should investors be excited about?
Yeah, I think the easiest way to answer that question, Brian, is right power, right timeline, right locations. Just because you have power doesn't necessarily give it a tremendous amount of strategic value, right? We always say a MW is not a MW. Because if you have a MW in a place like downtown Manhattan versus a place like in the Yukon or in somewhere in Central East Asia, it's just worth a different amount of value. Where we have our power, we have focused on areas that are outside of major metro areas. We focus on high barrier to entry markets, and we focus on areas that all the hyperscalers have underwritten for a while and a long time as kind of their area that they want to focus on. They're having trouble growing in those areas because the barriers take years to overcome.
When you look at power and power secured, I think power pipeline is probably one of those definitions that has the most wide-ranging amount of potential answers across the industry. We like to say there's probably 20 companies doing this transition from Bitcoin mining to HPC and AI, but there's probably 30 different definitions of power pipeline. Everyone has a different view. For us, it's a combination of the power that's currently online and running through a meter, the power that's secured through an ESA, as Jonathan just mentioned, or power that has significant work behind it from the company. It's going through a detailed load study, for instance, at a place like Panther Creek, which already have months of work to complete the conceptual load study, confirm the power is there.
It gives us a really high confidence that we're going to be able to continue to secure those megawatts in the future as we execute. When you look at our power, it's 2027 power. When you look at our locations, it's outside of N.Y., it's outside of Philadelphia, it's outside of Seattle, it's outside of Portland, it's outside of Montreal. These are really major areas, really high-value markets, and really hard to grow. When you see our timelines of 2027, most of the market is already kind of moving on from 2027, thinking that they're not going to be able to get 2027 power anymore.
This becomes a very strategic opportunity for companies who are looking to grow, looking to scale, especially in these markets that have very high barrier to entry, and the best way to do that is with a company like Keel, who has it secured and can cut years off their timeline.
Great. Now, as it relates to Moses Lake, Sharon, and Panther Creek, what has Keel done thus far to prepare for leasing of these locations? What needs to happen for you to move forward with lease agreements?
Yeah. It's a great question. Each site is a little bit different. We've taken a different strategy at the three sites. I'll just go from west to east or from smallest to biggest. At our Moses Lake site in Washington, 18 mw. That's a relatively smaller site. That's a site that has a very different kind of customer profile. That's more like an emerging neocloud or probably an enterprise customer. Could also be maybe a government agency. They tend to take a lot longer to negotiate and to work through the steps to getting an agreement in place. For a potential customer like that, we took a different strategy.
There, we underwrote a lot of the long lead item equipment to go from a piece of dirt to a fully functioning data center, minus the actual compute racks themselves. Our understanding was with an enterprise customer or an emerging neocloud, they wouldn't have the specificity that a hyperscaler would have that says, "This is my exact build, this is how I want it. I'm not going to vary from this, and I'm not going to be able to accept your backup generator choice," or something like that. Emerging neoclouds are much more flexible. They don't have those requirements. They also don't have these long supply chains secured like the hyperscalers do. We've gone and underwritten all of that off of our own balance sheet, and we're executing that as kind of a turnkey package.
The last things that really are in place is clearing the remainder of the last few permits, Brian. That's something that we've outlined for investors a few different times. It should be done really mid to late summer timeframe for all three sites. Moses Lake, realistically might be the first one fully permitted and out there breaking ground, moving shovels, and should be the first site fully online. Really between now and that site coming online, the only thing that we're waiting to clear as kind of a milestone that investors would be aware of, is just clearing the final permits. The next site that we have, Sharon, which is in Western Pennsylvania. We've got zoning and preliminary development cleared. We're really far advanced on the conditional development cleared for the development buckets. We're very far advanced at Sharon for the permits.
For that one, the market is very different. At 110 megawatts, you're really looking at established neoclouds, hyperscalers, and large scale enterprise recently, who are starting to integrate AI into their businesses in a much more significant way. They are looking to take control of the compute themselves because this is becoming an increasingly crucial part of their business and their corporate DNA. Those sites, we've taken a different approach. It's advanced the sites all the way through permitting, and once you get through the more controversial permitting steps like zoning, which we've already cleared, that enables us to have a lot more confidence going into negotiations. Same thing, for investors to look out for is clearing the final permits, which is expected similar timeline, mid to late summer timeframe. This site is actively under commercialization. Everything is moving forward there.
We'll put out press releases when we clear all the remaining permits. That will be the last noticeable timeline for investors. At Panther Creek, which is our flagship site. The 350-megawatt site outside of New York and Philadelphia. We've done basically the exact same thing. We've cleared all the permits and trying to de-risk the site through the planning and the engineering and the design to give us as flexible of a development package as possible. We've also cleared zoning and preliminary or conditional development at Panther Creek. We're just waiting on the last few permits to clear in the mid to late summer timeframes, active under commercialization. For a site like this, the reason why this is our flagship, not just because of the size, is because of the location.
When you've got 350 megawatts outside of New York and Philadelphia, close to Virginia and Data Center Alley, that's a really, really hard to reproduce site, based on its location, scale, and timelines. That's attracted a lot of interest from the hyperscalers, the largest language labs, and those are really the kind of customers that I think most of the investors really want to see land as tenants, and those are the customers who would be actively competing over a site like Panther Creek.
Great. As you mentioned, hyperscalers, they're spending $ billions, time to power is so important. Sounds like permitting is close. Once you have a lease in place, how much time will it take for you to become operational at those sites? Well, you already mentioned the second half of my question, so sorry. Just maybe what is time to readiness for each of these locations?
For energization commissioning of the data centers, Moses Lake should be the first site fully online, and that should be done probably in the first half of 2027. With Sharon, that will be the first site fully online in Pennsylvania. Power should be coming online in the Q4 , when we commission the first building. At Panther Creek, the 350 MW, the first building we plan to commission in the end of Q4 2027, with subsequent buildings being commissioned in 2028 to enable kind of a smooth and scaled ramp-up schedule over time.
It kind of ranges based on the size of the facility and how quickly you can get to readiness. Is that right?
It's definitely smaller is easier and faster to construct than larger. There's also just deployments for the tenants, right? Most of the tenants don't necessarily want to drop down 350 megawatts of equipment on one day. They want to have a more normalized schedule of, this is a month-over-month, this is a quarter-over-quarter general ramp, because that's how they secure their supply chains as well.
Got you. When you say 18 megawatts for Moses Lake, 350 for Panther Creek and 110 for Sharon, I'm assuming this is energized capacity. What is expected load capacity? How should we think about economics, how they're different at each location? Are there any proxies for competitors who have announced deals of how investors should think broadly about what that means for Keel?
It's a multifaceted question. In terms of market data for the locations that we have, I'd say it's very few and far between, right? Most of the market data that we've seen in the industry so far has been in Texas and a few other locations. In Pennsylvania, Washington, and Quebec specifically, I don't think there's been any market transactions that I can point to. There's been a number of hyperscalers underwriting the area. Notably, Amazon has invested in two sites within about a 45-minute to an hour drive of our Panther Creek site. They acquired the Susquehanna site from Talen Energy as well as another site nearby. I believe it's CoreWeave who invested on a 300 megawatt site kind of southwest of our Panther Creek location.
There's been a lot of developments happening, but I don't think there's been a lot of tenant contracts that I've seen in Pennsylvania or Washington. What we've seen is that there's kind of been a segment in the industry where everyone is really focused on training, and I think the emphasis at the beginning of the industry, maybe two years ago, really starting on this ride, has been how do you get as much training online as fast as possible? Because there's a race, and if AI is improving at an exponential rate, maybe if you don't start now, you're never going to be able to catch up. Right? I think that was what was driving the industry at first.
What the industry is going to eventually be driven on, we're seeing that shift taking place now with the enterprise, is actual utilization of AI and implementation of AI. These businesses are not making money training AI. It's just a huge area of expenditure for them. They make their money through the inference of the AI. As we see the market shift over from training demand over to inference demand, I think that's going to probably change the economics. I think it's going to provide greater emphasis on the locations. We'll see that probably coming over the next year or two as the industry starts to shift over. The best example that I can point to right now is just the enterprise demand that we're seeing, where we're seeing more and more large-scale enterprise figuring out how to implement AI into their businesses.
For them, it's a very different set of economics, right? For hyperscalers and for clouds, there's kind of a ceiling on what they can charge for their compute capacity. For a business applying AI, there is no ceiling on the value that they can create by applying AI. Right? They can improve efficiency. They can reduce headcount, they can reduce costs, they can drive revenue, they can create whole new business lines. They can find correlation across 10 different asset categories that nobody's ever been able to understand the data on. They can prevent massive shoplifting through real-time detection of shoplifting through CCTV cameras. There's just no limit on that value.
I think that inference is going to be the dominant player, and I think that's when it does, and it's going to be a gradual transition over the next couple of years, the sites that we have based on the location should have more and more value.
Right. If I'm hearing it right, if I look at data center announcements and lease agreements in Texas and other locations, maybe it's not as centrally located to some of the biggest cities as yours in Pennsylvania. In addition, you've got the Vera Rubin, where they're using the older Blackwell technology. If I'm hearing you right, those economics may be a little bit lower than you'd hope to achieve.
I don't want to point to any specific level of economics that I think we'll achieve through our leasing efforts. What we've said for a long time is that the economics continue to improve for landlords, and I think that trend has been really clear for the last two years. I don't see that trend changing. The landscape is such that we are solving really high-value problems for the tenants. The tenants really want these problems to be solved. I think what we've seen is an evolution of the industry and a maturation of the industry where terms have continued to improve. Creativity has really been abundant, where people are taking really different approaches to how do you solve the credit problem? How do you solve the financing of these really, really large programs? I think that's played out.
I think that's continuing to play out in our favor. I think as the movement of the industry over to inference continues to take hold, I think that's going to continue that trend as well.
Just to the other question I asked, how do I think about efficiency? Generally, I think the economics of any agreement is based on load capacity. How do I think about the efficiency of your locations, and do I have it right?
When you think about load capacity and efficiency, where that really comes out is in something called power usage effectiveness.
Right.
PUE. Basically what that means, Brian, for all the other investors, is that let's say I've got 1,000 megawatts. If I have a PUE of 1.5, that means for every megawatt I'm using on compute, I'm actually spending half a megawatt on all the other support operations. That could be cooling, it could be lighting, it could be the bathroom automatic sensors. It's absolutely everything that goes into that data center. If you have a PUE of 1.5 on 1,000 megawatts, you don't have 1,000 megawatts available for compute. You actually have 666 megawatts available.
Right.
You've got 333 megawatts available for everything else to support the 666. When you're in a place like Texas, where it's naturally a lot hotter, the temperature is a lot more extreme. That is kind of where you should probably expect your PUE to be. I think an efficient PUE in Texas is probably going to be like a 1.4, and more of a normalized is going to be around a 1.5. If you look at our sites in, we have sites, Pennsylvania, Washington, Quebec, nothing is below 40 degrees north. We expect that we should have probably closer to a 1.25 PUE, ±, just because the natural environment is so much colder.
If you take that on 1,000 megawatts, that means that we would have 800 megawatts available for compute capacity as opposed to 666. That's a huge improvement. The 800 over 666, that's a 20% increase in the power available for compute. No one's creating value on the PUE megawatts, right? You're only creating value on the compute megawatts. Minimizing the PUE, maximizing the available capacity for compute is what's going to be really valuable for tenants. All of our sites, because they're so far north, should have kind of an estimated PUE of 1.25 ± 10%.
That's great. That's super useful. You mentioned de-risking. I'm sure there's investors who are clamoring for you to announce a tenant. You made a strategic decision a long time ago not to rush to sign a tenant and instead to de-risk the site. Could you explain to some that might not understand why that's so important, why that's critical long-term for your company?
Yeah. This is a strategy that we embarked on 18 months ago, approximately when we acquired the sites. We acquired the Pennsylvania sites from Stronghold in March of last year. I think we closed on the transaction on March 14th. The sites weren't properly zoned. The sites didn't have the permits for development. The sites are current existing Bitcoin mines and power plants. That doesn't mean that they have the permitted status to develop a data center. What happens in these leasing negotiations is, the tenant doesn't want to underwrite any risk. If you have a site that doesn't have a clear path forward, you're going to pay for that in the lease.
You could go out there and potentially sign a lease, but you would, one, get a much lower level of economics because the tenant doesn't want to underwrite the risk and the uncertainty associated with permits and development timelines and actual delivery dates. Two, the business would actually face a liability that we would have to deliver by a certain date, then we wouldn't even be able to secure the permits to be able to deliver by that date. The best way for us as a business to create value out of these sites was to de-risk the sites through continued design, engineering, permitting, which is exactly what we've done, bring them to the point where our confidence on the sites rolling forward and our timelines for the sites became incredibly high and verifiable. That's exactly what we've done.
Now when you go into these lease conversations, we're really in the sweet spot because if we're starting too early, we're going to undersell everything that we have. If you start too late, you're going to lose out on that timeline to energization, which is so valuable for the tenant. If you start in kind of this Goldilocks phase where you're through the more controversial pieces and you have a very high confidence clear path forward, that's your optimal, and that's exactly where we are. I think that's played out well from the business perspective. Then we also had a general macro perspective that the lease economics continue to get better. The supply capacity is going to continue to be constrained, and all of those things should result in overall better economics than trying to sign today.
Our primary goal when you're looking to sign leases like this is maximizing our return on equity, maximizing our net operating income, because we're really going for that margin expansion that comes from the huge increase in the value that we create for the megawatt, the contracted revenues for a long period of time, and that multiple expansion that goes from a Bitcoin miner to an HPC and AI infrastructure company. Maximizing on lease economics was a key part of that strategy. I think we communicated that early. I think we were unique in that approach. Maybe the market didn't really like it necessarily when we first announced it. What that means is now, where we are right now, is an absolute Goldilocks phase where everything is lining up and everything is coming into place.
Where a lot of companies have all of this in the rearview, we have all of this in our front view.
Great. The company just raised $458 million in a convertible note. Maybe talk about your liquidity today and what's the total remaining CapEx for the three main sites that you still have remaining? How should we think about that?
I'll go through our liquidity guidance as of Q1, and then how that is impacted by the convert we just did.
Great.
As of Q1, our guidance was we had $533 million of liquidity on the balance sheet, and that was enough liquidity to get us through leasing and then have all of our cash SG&A fully funded for 2027 and 2028. We're assuming cash SG&A of about $100 million a year. That'll move around because of various divestitures and winding down certain businesses, but it's a reasonable assumption for now. That guidance remains completely unchanged. Even if we hadn't done the convert, that would still be our guidance, that we're fully funded on a cash SG&A basis through 2028.
We issued the convert the week before last because we had some very specific uses of capital that were going to be available sooner or later, and we wanted to raise the funds at an opportune time, particularly during a benign market environment. The use of those funds will be around power capacity expansion, some of it at Panther Creek, some of it around natural gas infrastructure at Scrubgrass. These sorts of expansions are extremely attractive to shareholders because it is adding additional capital spend to de-risked projects. Rather than taking de novo development risk, we're simply adding to the numerator on which we earn a cost of capital. One has to be ready to make these investments as soon as they come up and lock them in.
At Panther Creek, we've discussed publicly our objective of increasing the capacity available on our ESA from 350 to potentially as much as 500, although I think we'd be quite pleased if we could actually achieve that. Nevertheless, increasing that sort of capacity involves building additional substation capacity, calling away custom transformers, additional transmission upgrades and significant LCs with utilities. Similarly, at Scrubgrass, we're looking to enhance the delivery potential of natural gas to the site so that it can support behind-the-meter CCGT scale generation that we would do with a generation partner. Again, some tens of millions for CapEx build-out, as well as a variety of LCs. We wanted to have that money immediately available because of the attractiveness of the investments rather than trying to urgently scramble around at the time it was needed.
Again, this second convert has specific use of funds in mind, appreciating money is fungible. As to the cost, the full FX costs of each site that we really think of as taking on after a lease, suggest folks use industry rules of thumb as a reasonable convention for their own modeling purposes.
Okay. What do you think the rule of thumb on cost per megawatt is, the range?
11 - 13 is probably.
$11 million-$13 million. Yeah.
Rule of thumb, or at least how we think of where most folks come out.
We got some questions, I'll get to them. We've talked about the three main sites. Maybe you can talk about other pieces in your portfolio. You said, Scrubgrass is an exceptionally large, potentially game-changing location. Maybe talk about that. One of the questions we got is there's been some news in Scrubgrass. How does that impact Keel? Maybe you can address that at the same time.
Sure. We continue to move forward on all the sites. What we've outlined with Scrubgrass is that this is a pipeline site that we're focusing on securing the energy, and this is probably our longest out project. This is probably a 28, 29 timeframe at the earliest for the first commissionings of the first buildings. When it comes to the things that are happening legislatively and kind of regulatorily, those might actually establish some potential moats for us and make these higher barrier to entry markets even higher barrier to entry. The reality is that a lot of what they're recommending and proposing through these legislative frameworks, which right now nothing is in place and nothing impacts project scope or timeline or scale. A lot of it we're already doing for sites like Scrubgrass, right? Bringing behind-the-meter generation in addition to a large interconnect, that was already the strategy.
For us, that doesn't impact our plans. Maybe it potentially helps to speed up the process. Maybe it's a neutral, is kind of how we're thinking about it, but we're going to continue to keep an eye on all of those different developments. I think the reality is that we've got a really supportive community around our two main projects that we're focusing on in Pennsylvania, in Panther Creek and Sharon. I think that's been a huge part of our strategy going into this, is engaging early and trying to be transparent and trying to be answering as many questions and putting the face time there. I think that's paid a lot of dividends, and I think that really ensures that we're going to have a successful project at both of those sites and high confidence on Scrubgrass as well.
Great. Now, your data centers, like I had mentioned, are at least mostly going to be using the Vera Rubins. The Blackwells have only been in production for so many years, but we've got better technology. What happens in three - four years when the next generations of GPUs come? What happens to the data centers that are running on Blackwells, and is there an upgrade? Who's going to bear that cost? Just maybe go through the evolution of technology, what happens to these data centers.
Yeah. It's a great question, Brian. The compute market has constantly been innovating and evolving and driving efficiencies in compute. If you look at the supply of compute over time, there's never been a reduction in the supply of compute on a year-over-year basis, right? Every year, the amount of data center capacity, the amount of compute capacity increases. Every year, the chips get better, they get more efficient, they get more productive. In parallel, so too does the utilization and the demands on that hardware. Right? There's been a fairly nice balance between the increase in productivity, efficiency of compute, and the increase in demands for that compute as the cost and the efficiency has gone down. The Jevons paradox, as everyone is probably aware of.
Most of these data centers haven't been growing at this kind of a rate that we've seen over the last couple of years. Like, the traditional data center CAGR is like a teens kind of % CAGR. The technology wasn't evolving the way that it is right now with AI. It was 10 - 15 kilowatts per rack for a very, very long period of time, and upgrades were really on the compute side, and you make an economic calculation. Okay, this is the value of the compute on the books. This is the value that we can generate from running it. Here's the margin relative to the operating cost. Does it make sense for us to upgrade and sell this or keep this one running? That's been a determination that people have run for a very, very long time.
A lot of people do this also in their daily lives with their personal laptops or their smartphones, right? They say, "Hey, there's a new one out there. It's bigger, it's faster. Do I really need it?" Some people may want to pay the $1,000 bucks or whatever to upgrade their iPhone every year. Some people may be totally fine letting it run for five years and just happy with the device. It's really specific to the customers, and how they're positioning. What I think what matters for our investors is that we're not signing contracts that expose us to a three-year upgrade cycle at our expense, right? You're looking to sign contracts for 10 or 15 years, which enable us to recover our investment in the infrastructure and make a nice return.
We shouldn't be held responsible for the cost associated with upgrading, which means that customers will just determine whether or not it makes sense. This is something that will likely take place if the power capacity continues to be in severe shortage. The more severe the shortage, the more incentivized people will be to upgrade. The less severe the shortage on energy capacity, the more likely it is that they'll just continue to let things run, and they'll just deploy their compute in a new location.
Great. I guess to end it, maybe your message to shareholders, there is a number of companies, stocks, that they can invest in, to invest essentially in the opportunity for data centers and the need for power. Why should they invest in Keel?
Yeah. It's the fundamental question that every investor needs to be asking. I'm not trying to convince people around AI or the AI trade or infrastructure or semis better than infrastructure. I think if you're looking at Keel right now, you've probably already made up your mind around the AI trade. You're probably looking for what are the best ways to get that exposure. There's been a number of companies who've had incredibly successful run, and have followed a very clear playbook around signing leases and executing against the development and the construction of those facilities and the delivery of those facilities and unlocking the value that all of that creates.
I think the good opportunity with Keel is that we've been doing all of this work to get us ready to this point, but we haven't executed any of these major catalysts that we've been working towards and that we've been guiding towards. I think all of that is in the front view. If you're looking for the rotation from Nvidia or something, CoreWeave or whatever it is that you've been running up for the last two, three years, you're probably looking at those companies who don't have the catalyst already executed against and have that in their front view. Because that's what you're looking for in terms of the opportunity set. I think there's very few companies who compare with Keel in terms of the opportunities that we have to execute against over the next two, three years.
I would direct every investor to our quarterly earnings deck, where we have a slide that tries to explain how we're creating value for shareholders and what the potential implications that could be as we execute against advancing the sites through permitting and leasing, securing additional expansion capacity, which we don't believe we're getting little or any value for, and then delivering the sites and continuing to scale to the business from there on. I think we're incredibly well-positioned, and we have a tremendous set of sites and opportunities ahead of us over the next six, 12, 18, 24, 36 months.
Great. Well, we appreciate your time. We look forward to hearing the promising news on permitting, right? We can take that next step. Again, thanks so much.
Thanks for the opportunity to speak with you.
Great, guys.