All right. Perfect. Good morning, everyone, and welcome to our 54th Annual TD Cowen TMT Conference. If you don't know me, my name's Michael Elias, communications infrastructure analyst here. Today we're joined by Core Scientific, and from Core Scientific, we have their CEO, Adam Sullivan. The last two sessions I said I would take questions. I don't even think this go -around I'm going to even try. We'll get into it and we'll talk. Adam, thank you so much for being here. Really appreciate it.
Thanks so much for having me.
This is what we'll do. We'll kick things off by starting with strategic priorities. I remember you came out in November. You said what your priorities were.
Yep.
As we get into the second half of the year, what are the updated priority set for the business?
Yeah, right now it's really the expansion across the five new sites that we have under development. We did our capital raise just about a month ago now. Putting the $30 billion on the balance sheet was incredibly important in terms of being able to pre-fund a significant amount of equity that is required as part of the project. Building out that breadth of portfolio, putting the capital in the ground, getting GC secured, long lead equipment secured, those were all incredibly important and put us in a position to win contracts here. We're working to diversify our customer base, and I think that's a lot easier with the position we're putting ourselves in with the capital deployment we've been able to do over the course of the past six months.
As you think about the demand set, from an industry basis, we've seen a big acceleration in demand. How have the conversations, and this isn't a question about specific sites or anything like that.
Yeah.
This is more of a question about the customer's own requirements. How have the requirements evolved since November of last year to today? Are you seeing a shift in terms of densities? Are you seeing a shift in terms of what is underlying it at the hyperscaler or AI lab level? Any observations there?
Yeah, there's been significant changes in the past six months, which has been pretty incredible to watch. I think you saw the hyperscalers executing a lot of demand through neoclouds in Q3 and Q4 of last year. Realizing they have a larger structural demand. They've come into the market, obviously, very strong. What we're seeing is the first data halls that we're building out across each of the five sites, essentially for GB300.
Yeah.
I think that part two of that is future buildings shifting to 800 V. That's going to become a requirement for a lot of deliveries in 2028. I think those are all things that we are trying to be at the forefront on and being as accommodative as possible in terms of how we're thinking about design for future developments. That's a big change in supply chain. It's a big change in purchasing. We're putting ourselves in the best position to be able to sign long-term contracts that have multiple years of deployment. To do that, you have to be accommodating to a significant amount of design changes that are going to be coming down the pipe here very soon.
You have a portfolio of sites. Some of which you've leased, others you're looking to lease. There's a flywheel where you need to continue to add capacity, buy land, and so on.
Yep.
When you're thinking about going and expanding the portfolio, what is it that you're seeing in the conversations with the customers, be it roadmaps or what have you, that gives you the confidence to go out and say, "Hey, we need to acquire X amount of land, and we need to keep going. We see visibility and demand through 2028, 2029, so we need to restock the shelves"?
I think for us right now, we have the power secured. I think for us, over the past four months, we've been very aggressive on land purchasing. The big change that I think is just starting to get recognized in the market today was the shift from a hard cap in the very low double digits as it relates to behind the meter for hyperscalers. To now that shift moving to greater than 50% that they will allow in behind the meter. Requires a significant amount of land, a significant amount of engineering planning as it relates to getting distribution pipelines to the sites, and securing pipeline capacity of natural gas. I think those were changes that we foresaw.
We've been very aggressive in terms of our planning as it relates to that. We have enough power right now for our development pipeline for the next three years, which we feel very good about. Right now, it just comes down to having five sites, first RFS in 2027. That puts us in the best position to be able to go out, win contracts. For us, the great part is, and I think this is something that isn't well understood, is the fact that our RFS dates don't slip if the conversations drag an extra 30 days with a customer.
It frankly just puts us in a stronger position where that site gains more value every single day we continue to put dollars into the site.
Mm-hmm. As you think about the business that you're building, your first deal is with, obviously, CoreWeave. You have incremental capacity that you're looking to lease. We've seen a bit of shifts in the market in terms of financability and different contracts with the demand for capital. I'm curious, how do you think about engaging with customers who are not investment grade yet?
Yep.
As part of that, the requirement to have some form of backstop and I appreciate there are different flavors.
Yeah.
Is that a hard line requirement? If so, what flavor?
It is a hardline requirement. There has to be some type of backstop. I think as we look out over the course of the next 10 years, when you think about the evolution of the neo cloud business. Which is essentially where the non-IG customers are. It's thinking about the long-term stability of certain businesses and picking winners. I think CoreWeave, obviously far and away market leader, the talent they've been able to bring into that business, and I think the sustainability of that business is incredibly strong. I think some are more unproven, but you also have to recognize that some of the chip makers want other winners other than CoreWeave and Nebius in the market.
There's going to be a number of new, and we'll call them hyperscalers, that emerge out of this class. I think there's going to be a lot of consolidation. I think we've spent a lot of time negotiating credit wraps with a number of different investment grade counterparties for deals we're negotiating with neoclouds. I think some of them are very attractive. I think others are frankly a lot more challenging than I think. They have a lot tougher sledding than they expect on the neocloud side. They don't have the talent necessary to execute at the scale that they're looking to execute on.
At the higher end of the neocloud space, I think a lot of those have acquired such incredible talent from across the hyperscalers and traditional cloud businesses that they have the capability to be successful.
One of the things that I struggle with is that I think of the hyperscalers, obviously, they have established recurring revenue businesses. As I think of some of the chip companies, it's like, yes, the financials have, and the revenue has gone crazy because we're in this cycle. It's a non-recurring revenue business structurally. As you think about backstops, do you also feel comfortable with the chip companies backstopping recognizing that there isn't that same recurring revenue piece in their business? Does that make sense?
Yeah. No, it does. I think the thing that we've thought through a lot is it's kind of vendor lock-in at this point, right? It's like the discussions, whether it's NVIDIA, whether it's AMD, whether it's Google, they're so focused right now on how do they secure not this sales cycle, but the next sales cycle.
Yeah.
Giving the credit backstop is essentially a forcing mechanism to not only sell that first 40 billion in chips, but to lock in the next $40 billion sale six years from now. I think, there's a significant amount of credit that's going to be extended on that basis of how do you vendor lock in a lot of these bigger companies. Especially locking in certain neoclouds into certain vendors. Obviously, we have winners on the NVIDIA side. We have, obviously, CoreWeave being far and away, I think, the market leader on that side.
There's companies like TensorWave that are being chosen by AMD to really be a market leader. I think, you're going to continue to see Google pick new folks for them to try to lock in folks into the TPU infrastructure. This is all part of a broader plan on how do they secure essentially recurring revenue out of selling GPUs every few years.
Oh, makes sense. Okay. I want to pivot and talk about Pecos and Muskogee, but just more at a high level. You've talked to different customers at this point regarding the site. Just from your conversations with customers, kind of general feedback, thoughts, has there been anything that's a particular obstacle to getting a deal done? Conversely, what are the elements of the sites that the customers find most attractive?
Yeah. I think it's always labor, number one. Pecos, it's a challenging location from a labor perspective. I think the good part about Pecos and what we've de-risked over the course of the past, since November, is the fact that we have GCs, we have subs engaged. That's always the first question. It's like, how are you going to get concrete to the site? Well, we had to build a concrete plant on the site. Right? Because you can't ship concrete that far from the nearest concrete plant.
Yeah.
For us to be able to deploy that much concrete in that location required us to mix on-site. We've de-risked a number of different factors from that perspective. You look at a site like our Muskogee site, incredible opportunity. We're obviously already building there, with a 70 MW data center for CoreWeave. You have Google building a 600 MW data center in the property next door to us. Just from a labor perspective, it's incredibly competitive. I think those are two isolated cases.
On the other side of that is you're seeing that nationwide as well. It's an extremely competitive labor market. I think on the other side of that, Pecos' ability to expand is just something that is incredibly enticing to customers right now. That shift I talked about earlier in terms of willing to push further into behind the meter and percentage of the overall site makes Pecos incredibly attractive. We couldn't be closer to more natural gas pipelines in West Texas. From that perspective, Pecos is an incredible opportunity for us. Muskogee, we're incredibly excited about, continuing to expand.
Obviously working closely with the governor out in Oklahoma as it relates to behind the meter as well, and just the amount of power that we're going to be able to achieve at that site is just incredibly large and a great opportunity for us.
One of the things when I think about both those sites, correct me if I have this wrong, there's an initial utility. You have utility power-
Yep.
... up to a certain amount, right? From there, you're supplementing it with onsite generation. Recognizing that there is actually a utility piece on site and understanding that the ideal pathway for the hyperscalers to get power these data centers is through utility, does utility represent a meaningful differentiation in the customer conversations, or is it solely about speed to market, the source of the electron is less relevant?
It's really less relevant. This is all about speed to power at this point. We can solve via behind the meter the first nine. We can solve via batteries and generators the next two nines. I think they're getting comfort around it. I think the big part is the first buildings that are going up, at least the first building, in some cases multiple buildings, are utility, which gives them a lot more comfort. The fact that we have additional power coming at those sites later also gives them more comfort just as it relates to their ability to continue to expand and scale.
The timelines on behind the meter right now, you can deploy hundreds of megawatts inside of 18 months. The biggest lump on the 10 is really just as it relates to distribution level pipelines getting to the sites.
Like the trunk lines and laterals and.
Yeah.
Got it. Okay.
Frankly, those are two sites that are going to be incredibly large and I have an expectation that those are going to scale fit to a very similar size to what we're kind of quoting right now, which is a gig of critical IT load.
One of the things that when I think about the industry is there are two times when you celebrate. First is when you sign a deal, the second is when you deliver the deal on time and on budget. Now you're through process and I think you're halfway through delivering, or just about halfway through delivering for CoreWeave. What have been the key learnings through that process, and what do you take from that to building out at scale Pecos and Muskogee?
Yeah. We had 245 MW fully commissioned live, GPUs spinning. At our last earnings call, we had 200 going through the commissioning process, so we'll be around 450 MW at the end of the summer this year. Absolutely incredible learnings, and I think a lot of them stem from two areas. First is design, in creating a much more flexible infrastructure base. Obviously, we're designing for a lot of CoreWeave customers, which are some of the largest companies in the world. Just learning what their needs are, how they're thinking about future development as it relates to the shift into the next GPU cycle.
Being able to aggregate all of that information, design information that we've gone through and processes we've gone through, has helped us create our new standard bases of design, which is a lot more flexible. For us, it's created a much easier time for us to order parts on the supply chain side. I think part two of this is securing great partners on the labor side that have national footprints that can support you across multiple sites. One of the biggest learnings is when you're going to small communities, which is where a lot of our facilities are, partnering with certain groups that you're bringing in from out of state that aren't necessarily part of a broader national brand.
There's no allegiance there, and they're willing to reallocate labor to other projects because they're going to get paid 30% more on another project. You start to bleed some labor as it relates to that. Through the process of building these first five sites and now with the next five sites that we have coming down the pipe, picking really strong partners on the labor side and making sure that the labor doesn't walk off, or parts of the labor pool don't walk off throughout the construction process. That's a huge learning here for us because small tweaks on schedule can create much broader impact.
Yeah.
Or the 18-month build process and, frankly, we've chose being able to choose and work with a lot of really strong national partners that have really helped us as we've continued to grow these new sites.
Would you say, as you think about expanding the portfolio, would you say that the leasing of these sites is a priority before continuing to expand additional capacity? How do you think about the balance between the two of those?
What was the first part?
Was like, how do you think about the balance between adding net new sites that you can lease to the portfolio versus leasing it first, then going and refilling the tank? Is there a priority with which you are going to grow? Does that make sense?
No.
It doesn't?
I'll answer your question anyway.
Sure. All right, cool.
Why not?
Yeah, why not?
I think for us right now, it's like we have the 1.5 grid connected power-
Yep.
... on the critical IT load side. I think from the perspective of what we're negotiating right now, there's going to be significant amounts behind the meter at both Pecos and Muskogee. There's a limit in terms of how many megawatts we could potentially build in a year.
Yeah.
There's going to be a significant focus on Pecos and Muskogee, obviously, is the upfront. Hunt is an incredible site. We're going through the process right now behind the meter, that's another site that's expected to scale significantly. We have two smaller sites that I would call are kind of power locked.
Yeah.
Our Dalton site, 120 critical IT load megawatts and Auburn in 32 MW. Those are still really attractive sites, those are going to be one and done situations. Focusing on the sites that are going to continue to scale while still managing the projects. Obviously, GC's been engaged for a while at Auburn. That site's moving very well. Dalton 5 is a site that, we have Dalton 1 and Dalton 4. Dalton 1's fully operational, turned over to CoreWeave. Dalton 4's also a CoreWeave site. Dalton 5's a 120 MW building that frankly is perfect for GB300 and is scaling very quickly. We're already through civil work there.
I think the focus for us is just really continuing to scale at the existing sites and finding the right clients that we can continue to scale at those sites. For us, as we think about pipeline moving forward, we're looking at really power that delivers at the end of 2029, beginning of 2030. For us, it's a lot cheaper opportunity, not having to pay up for power in 2028, because frankly, if we bought 2028 power, we couldn't build it.
Yeah. I was going to say, implicit in the answer to your question, or you answered it, which is that you can only build so much. I've been thinking for you guys a rough frame of 400 MW to maybe 600 MW.
Yeah.
A year that you could deliver. You tell me you have 3 GW in total that you could build out.
You have runway for the next few years.
Yeah.
Okay, cool. As we think about build costs, because you were the first ones to sign in this space and deal, you have a sense of, in 2025, 2024, what the build costs were looking like?
How have they evolved? I appreciate it's going to depend based on design, but just as you think of more broadly, how have they evolved? Second, where have you seen the most inflation as part of the overall build cost?
Yeah. First and foremost, we'll talk about inflation, right?
Yeah.
That's on labor. 20% nearly across the board, 30% on electricians. It's year-over-year, it's been a significant increase. You're talking about labor used to be 30% of overall build cost, now flexing closer to 40% or higher.
Wow.
Labor is obviously an incredibly important part about this entire exercise as it relates to build cost. I think the interesting part that we're going to see over the course of the next two years is we saw an inflation, right, of build cost. You go back a few years, you're looking at, let's call it $8 million a megawatt. That's slowly inflated. We're now touching $11.5 million, $12 million per megawatt. As we go through the design changes that NVIDIA's expecting to go through as it relates to Vera Rubin.
Oh.
Even, really the change to 800 V, I think we're actually going to start to see that pare back a little bit.
Interesting.
They're pulling certain parts of the design out of, or they're pulling certain structures out of that design that actually will lower the overall build cost. Frankly, some of that's just related to how do we remove the longest lead parts of the supply chain to essentially allow us to build data centers faster. That's where a significant amount of design and engineering work has gone into. I could easily see that paring back down to closer to $10 million over the coming years.
Interesting. What would you strip out? If you're stripping out long lead, we already stripped out gensets largely on the compute train. That was a long lead. You probably can't strip out megawatt switchgear. I know that's like a two year lead time item. Just curious, what do you see?
I mean, the next big thing is stripping out UPSs.
That's what I'm feeling we're going to see.
Yeah.
Okay. All right, cool. Inflation on the labor side. As you think about the ability, yes, there's price, but price is rising, I would argue in part because of the volume of demand for these items, the derivative question becomes lead times. Where are you seeing the longest lead times? I think you made a comment earlier when you were talking about gensets or doing onsite generation, that you can get them in actually reasonable timeframe.
Yeah.
Did I misinterpret that or no?
It's about a 14 -month delivery timeline right now.
Okay. All right.
The biggest item, though, is medium voltage.
Yeah.
Switchgear right now, you're just looking at 100 -week lead times.
Ooh.
It's something that we've been very proactive about pre-buying a significant amount of capacity on the medium voltage side. Obviously, that's something that can be easily changed between sites. If we need to pull megawatts that we were sending to Dalton into Pecos or Muskogee, we can do that.
Yeah.
That's something that we've really had to get ahead of the supply chain on, just given it's really challenging to bring that in tighter.
Yeah. Well, that makes sense. Okay. I want to shift and talk a bit philosophical with you about how you think about approaching pricing and yields and so on, right?
Yeah.
You can take the traditional real estate argument, okay, this is a spread of a cost of capital business, and I need to, what's my financing cost? It's a function of my tenant, and then I price a spread, on top of that. There are also other approaches, right? You're thinking about, all right, as I'm exiting, because I'm building it to sell, I'm not holding it over-
Yeah.
... a 15 -year period. What's my exit and what's my IRR there? When you're negotiating these deals, how do you think about structuring it? How do you think about yields? What is the governing principles that's determining how you're negotiating from a price or yield standpoint?
I think it's actually, there's limited negotiation power, right? With hyperscalers. They know what the cap is in terms of what they're willing to pay on a dev yield basis.
Yeah.
I think, some people have quoted very high numbers on hyperscale deals. I frankly think everything's going to come down to probably a 12%-14% yield. It's easy for us to model hyperscale deals. I think that where you're going to find alpha is in neocloud deals at some point. Being able to pick the right winners on that side, getting high teens is going to be the people who choose the right ones. It's nearly impossible from all of our seats right now to know exactly who's going to be the winners there.
Yeah.
I think that if you're searching for alpha in the space, you have to go to neoclouds because you have a significant amount of pricing power. Where you don't have pricing power right now is necessarily with the hyperscalers.
I think on some of the yields that I hear quoted, it seems to me like the swing factor is the build cost that they're using for the denominator is lower. Right? Some folks talk about building out, let's call it $9 million a megawatt, which just seems, for me, quite low in this environment. Would you say that if you're fully burdening your build cost, would you say that the hyperscale is reasonably comfortable with giving a 10% unlevered yield on cost?
I would say that might be the highest end.
Really?
Yeah.
10% unlevered. Okay. All right. That's interesting. The other thing that we see on the market is that with how big these data center deals are getting, right? The numbers on turnkey, and I hear from the customers, the numbers on turnkey stop making sense, i.e., this is a lot of data center lease obligations. It does seem like we're seeing an element of a desire to shift back to a powered shell. Swap your OpEx for CapEx. I'll do more of the fit-out work.
Yeah.
I'm the customer. We'll do that. How do you think about willingness to do a powered shell deal or what we've seen some private operators do is a hybrid deal?
Yeah.
A portion of this is powered shell, a portion of it is turnkey, gets you to a more attractive blended return. How do you think about that?
Yeah, I would say we're not necessarily focused on powered shell. I think we've looked at a number of different hybrid deals. Hybrid deals are about $9 million a megawatt. It's a little bit further than three quarters turnkey. I think some of those actually pencil out better on a yield perspective, albeit you're deploying a lot less capital. From a delivery standpoint. The hardest part about delivery is the last mile.
If you're eliminating that, you're actually eliminating a bit of development risk, which actually, from an overall perspective of signing customers and being able to deliver on time and meeting SLAs, that actually might be an easier pathway forward. I wouldn't say that's our strike zone right now. Our strike zone right now is full delivery GPU-ready facilities, because that's really where a lot of the market is. Albeit, this might be a flash in the pan for some of the hyperscalers who are used to signing powered shell deals.
I think we've seen that mainly from AWS is traditionally a powered shell.
Yeah.
Less so, I think from their perspective, they will probably migrate back to that at some point.
Yeah. Here's one thing, when we think about the evolution of the data center market, I think of Digital Realty and CoreSite, they were born out of the pain of the dot-com bust, right? GI served as a consolidator, that's what led to Digital Realty, in part, in its current form. How do you think about the opportunity to be a consolidator in this space? There are a lot of new providers, not only on the private side, but I'd also say on the public side as well.
How do you think about consolidating, and as part of that, do you keep an eye to making sure that the businesses continues to be in a strong enough position where if that opportunity were to present itself, you're able to prosecute on it?
Yeah, I think right now valuations are inflated across the board on the public side. I think valuations on the private side are probably still a bit inflated. I think what we're going to see over the next 12 months is where execution actually is going to come to the forefront.
Oh.
People who are having significant missteps, valuations are going to come to a much more reasonable range and create an opportunity for relative value plays with deals that are actually accretive to the business. For us, it's about signing the next few large customers. Building out our organic pipeline essentially to take our valuation much higher from where it is today, and put ourselves in a position to be a consolidator. I think what we're going to see is essentially a replication of what we saw in the data center industry in the past. It's going to happen on a much faster timeline, though, just given how much build is occurring over a short period of time.
Is the enterprise at all a focus for you over the medium to long term?
Is it what, sorry?
Enterprise, like going after the enterprise opportunity. Recognizing that enterprises, we're seeing more of these like 50 MW, 100 MW, some even 150 MW opportunities.
Yeah.
Right. I appreciate it's a different go to market than giving your whole campus to a hyperscaler or, is that something that you think about pursuing over time?
Yeah, we looked very closely at it, including building out designs to essentially do five megawatt pods inside of a 100 MW build-out. Those are things that we've definitely considered. I think for us, that's probably going to come through acquisition initially. There are some companies on the private side that are focused on some of these smaller scale build-outs, especially as it relates to smaller facilities closer to metros that are attacking more on the enterprise channel. It is a different business model.
Yeah.
I think it's something that's going to come with time, and frankly, right now it's like you win one hyperscale deal. It takes the same amount of time as winning one enterprise deal, but the one enterprise deal is 20 MW, right?
Yeah. It's also a higher return.
Yeah.
Yeah.
You have to think about engineering dev cycles, right? We're limited in the amount of cycles our engineering team has, we have to be very cognizant of how we're dedicating those cycles. Really right now, the focus is just on larger scale single tenant sites.
Oh. How do you think about long term being a REIT? I appreciate you're in growth mode now, but is that something that's on the dashboard for the longer term?
Yeah, I think it has to be, right? As you think about, we do have a significant amount of NOLs today, we're a few years out from that. I think there's going to be a new format of essentially data center REITs that are large single tenant campuses. They consider diversification 10 clients, not thousands. I think the big question mark is, how do those get valued on a multiples basis look forward five years from now? I think there's going to be a few winners in that space that consolidate. I think the main competitors for that are actually going to be the folks like Blackstone and their new public company as it relates to buying stabilized assets.
I think it's really going to be a competitive process between those folks buying assets, which I think will be some of the largest REITs in the world very quickly. Competitors that are consolidating in this space, like us.
Last question for you. You've made your bet on the front lines for the last few years. What keeps you up at night?
I think right now it's mainly as it relates to things in the community level. It's something that hasn't necessarily kept us up in the past. We were a Bitcoin miner. Every community we went to, we were the black sheep. We did a tremendous job at going into every community. You could find a 5K in every community that's named after us. We're the largest donator to every food bank. I think the challenge right now is just there's a significant amount of public backlash as it relates to data centers in a way that we didn't necessarily even see fully on the Bitcoin mining side.
Oh.
I think for us right now, it's just how do we continue to stay at the forefront of being heavily engaged in the community. Working really closely with community leaders to ensure that we're getting the right information out about what we do as a company. It's for sure something that keeps me up at night because, frankly, I think it's something the hyperscalers have not done a good job at. I think they've created a lot more angst amongst local communities than trust.
Right.
That's something that we spend a lot of time focused on internally.
I could sit here for another hour. We got to wrap it up. Thank you so much. Really appreciate it.