Private equity has aggressively poured billions into data centers and digital infrastructure in recent years. But today, some investment assumptions are being stress tested.
For instance, AI demand is accelerating and yet nobody can perfectly forecast where capacity will be needed three years from now. All the while new campuses can take 18 to 24 months to deliver, add into the mix, rising interest rates, supply chain constraints, and one bottleneck that towers above the rest power.
In this episode, we explore the assumptions behind big PE data center investments, how the landscape is shifting, whether creative deal making is a good thing and where things might head next. This is the financing and infrastructure story behind AI told from the ground up by Goncarlo Bernardo from Palistar Capital, McDermott Will & Schulte's Carl Fleming, and Compute Labs' Nick Filichkin.
Catch the video at top, listen to the audio edition and read our transcript below, or watch this and future episodes on YouTube.
To learn more about the topics in this episode, check out:
- What you need to know about the $40B Aligned Data Centers deal
- PE firm Brookfield launches $100B AI Infrastructure Fund
- KKR, Oak Hill Capital commit approx $2 Billion to Global Technical Realty
- Beware 'cowboy' investors looking for fast data center returns
- AI investors want to own their own compute power
- Private equity sees dollar signs in the cloud
- Fierce’s coverage of the data center power problem
This podcast is written and hosted by Diana Goovaerts. It is edited by Diana Goovaerts and Matt Rickman. Liz Coyne is our executive producer. Special thanks to guests Goncarlo Bernardo, Nick Filichkin and Carl Fleming.
Diana Goovaerts, Fierce Network: Private equity has poured really aggressively into data centers and digital infrastructure over the past few years. So what fundamentals are drawing capital in and which of those assumptions are now being tested?
Goncalo Bernardo, Palistar Capital: Number one, I would say it's attractive returns. I think private equity really is driven by the risk return metrics, and when you see attractive returns on an infrastructure or infrastructure like segment, that clearly calls up a lot of attention. And so I'd say that's number one. People are finding attractive returns.
I think number two is the large scale of the deployments. So the ability to kind of really put that much money to work without being in kind of a much more competitive or commercially competitive environment, I think is highly attractive to most, uh, to most private equities.
The next factor I think is really important, so number three, is also the counterparties. So a lot of the investment is getting done with, uh, Microsoft, Google and Amazon, that kind of counterparty. So you're talking about some of the largest companies in the world with high investment, uh, great ratings that really kind of provide investors a level of certainty that the money will be there when it's needed.
So the combination of all of these really makes the space super attractive to private equity.
Nick Filichkin, Compute Labs: From the conversations that I've had, I think it starts with the fact that these are fundamental assets. This is infrastructure rather than speculative equities or multiples into new emerging businesses. It's easier to understand that there's cash flow tied to the usage of a tangible physical device equipment rather than, ‘Hey, I think this company is worth 10x in two years, if they can reach these types of OKRs or, or milestones.’
So in my opinion, number one, from what I've been hearing, just the fundamentals of investments and cash flow. Then it comes down to, of course, this is the next future of a revolution of technology. This is AI, this is going to be powering the next hundred, 200 years or so forth. This is just infrastructure and everybody loves the idea that they have assets powering technology.
And I would say the, the last piece, of course, is the yields, the high revenue that you get to receive from the usage of these devices and equipment.
Carl Fleming, McDermott Will & Schulte: The group that I lead here, you know, straddles both the data center world and the energy world.
I think one of the key things we've seen really growing recently is, you know, a number of funds that are looking not only to develop data centers, but also all the ancillary pieces that go into turning the light switch on. So we've seen a lot of funds invest into energy developers or into data center developers that have a unique play for a power platform.
So a lot of the work we're doing now is around doing unique joint ventures to bring powered land portfolios or power teams into the conversation with the data center developers or the hyperscalers. And you've seen in the market recently some, uh, some larger investments where private equities come in, or even tech companies have come in, and invested heavily into power generation platforms for data centers.
That's been kind of a newer development.
Diana Goovaerts: Are there specific types of energy that are being favored at this point? I know I've reported on everything from natural gas and nuclear. People are exploring hydrogen and then there are renewables, but renewables have hitherto kind of been hindered by the intermittent power availability and the expense of battery technology. So what are you seeing from investors in this space?
Carl Fleming: We've recently done the largest gas-powered joint venture for a data center campus, a very large one. And that has been years in the making.
You had a large gas-powered developer that was able to come to the table and provide that, and that was a pretty immediate sort of source. However, most new gas projects are gonna take five years to get turbines in the door and the backlog for this stuff. So there's a bit of a gap there.
So while we're still setting those up, those seem to be a longer-term play if you don't have one already lined up here. So renewables kind of fit really nicely into the interim.
You know, you can bring that to someone and say, ‘this thing could be turned on very quickly’ versus the five-year plus timeline for gas or nuclear. You know, we're seeing some of that. The SMR conversation is still gonna take like 10 years to come to fruition, but we see some nuclear developments.
But really, I'd say a lot of the activity as of late, especially after the administration came out with its BYOG or bring your own generation mandate for tech companies, it's kind of like, co-location is the name of the game right now. How can you bring your own energy quickly? A lot of that is slap some solar or wind and storage together and do that on the quicker side, as long as the regulatory regimes allow for that in the jurisdictions you're trying to develop in.
Diana Goovaerts: I wanna follow up on the part about testing assumptions, right? Because, you know, there have been some pretty significant layoffs at a certain hyperscaler recently. Well, several. And we've also heard about data center leases being shifted or paused and all of this.
So talk to me about are some of the returns or some of the assumptions around how good of an investment this is starting to be tested?
Goncalo Bernardo: I think the staffing I would say is less related to the data centers themself and more related to kind of general business reasons.
There's one CEO that says, before you put a new position out there, you kind of need to check can AI do it? So I think that's a very important part of the equation. So what I would say is yes, there have been situations where leases have been paused, demand has been shifted from one market to another, but ultimately I think it's a market that we're all kinda learning a little bit as we go along.
No one knows how fast the adoption of AI will be. No one can really perfectly forecast where they will need capacity in the next three years. Maybe they can do it well in the next six months or next year or year and a half, but in the next three to four years, it's a lot more difficult and you're really committing to capital which is gonna take at least 18 to 24 months to actually go into the commercial market with a deployment.
And that's in a very positive scenario where you've got everything lined up and everything goes well. So I think ultimately it's a very fast-growing market. It's really a matter of people are growing, they're growing fast and they're adjusting their expectations, then they're growing fast again, and then kind of they're rethinking as we go along.
We're talking about hundreds of billions and if you look at the industry as a whole trillion dollar CapEx budgets, so you can't kind of assume that everything's gonna go straight line and linear.
So we were talking to a couple people [about] markets where they were saying the U.S. has been great last year. But other markets have slowed down and now we're seeing people kind of saying, ‘oh, I was too slow in this market. I need to play catch up.’ So I think that to me is more a reflection of the speed at which the market is moving and less a reflection of the risk.
What we're seeing is in some cases, again, investments getting delayed. The campuses that people expect it to be leased up and occupied, kinda taking longer to occupy, but overall, that doesn't mean they don't, they won't get it.
But at the same point, the only thing I’d add is it's very important to think not just because you have a piece of land and you think you have potential access to power does not mean you're gonna have a good data center.
So everyone who's got a piece of land these days kind of comes out and says, ‘well, I have access to power. I had a conversation with the person at the utility and it's gonna be great. And there's a fiber cable that goes a couple miles away from here. So right now I have a perfect data center location.’ It doesn't work like that.
And so there's a lot of people that are gonna put money into these projects and unfortunately some of them will never be data centers.
There will be bad projects in good industries where everything seems to be going great for everyone, but there's a couple of projects that just never made any sense. So I think, again, what's the expression of a rising tide rises all boats, but there's still some boats that sink whenever there's something that goes wrong, right?
If the captain's not good, you're both not gonna survive.
Diana Goovaerts: Yeah, and I've heard that sentiment before that data centers aren't just another straightforward real estate project. There's a lot that goes into making sure that even if you build a facility that it gets used. You have to have the right connections. But I wanna stick to the capital side because there are things like rising interest rates, power constraints – you, you brought up the power issue – you know, the supply chain costs are changing. So how are all of those factors changing the way deals are structured for data centers compared to two, three years ago?
Goncalo Bernardo, Palistar Capital: So we've seen, for instance, in the last, uh, year, I would say a number of kind of more creative structures come up.
So we've seen, for instance, the likes of Google and some of the other hyperscalers basically wanting to use the so-called neo clouds, but knowing that their financing cost is gonna be too high. So they're willing to underwrite some of their debt, so provide assurance to the creditors that the debt will be covered.
So those are some of the more kinda, let's call it innovative structures that we're seeing happening, which is basically your end client recognizing that your provider data center company doesn't have the same level of access to the financial market as you do, but basically willing to kind of help them along the way and help them build up again.
One that's been very talked about also is Nvidia making investments in a number of players and kind of capitalizing them better. Again, that's another area where you see that the market's evolving. It's not everyone can get a data center and can get them financed at the same cost. There are different levels of players and different levels of financing.
So I'd say point number one is really your buyer's really smart. Your buyer will see or your hyperscaler will see a lot of what you see as well. So he is actually able to kind of adjust. We've seen that historically when interest rates starts to going up from being close to zero, we saw the yields on cost really moving.
When things move really quickly, a few of the less cautious management teams kind of get caught in the store. So we've seen this before where you have a great vintage of projects, then all of a sudden interest rate jumps or something, or the cost jumps. They had committed to a certain price with their client, which they can't renegotiate, but their cost is much higher and kind of add up their profit margin, in some cases give to a loss.
So what I'd say is the most sophisticated ones are trying to kind of lock them as much of the cost as they can. They're trying to pass on as much of the risk as they can to their client and/or ultimately to their supplier. But in a hot market, it's not always easy to do this, but again, you have sophisticated and unsophisticated players across the value chain.
We've seen a couple of them go as far as kind of build facilities together with some of the people in their supply chain to ensure they have access to product and cost. I think partnering with your suppliers ultimately is something that will really help you in that visibility and also being able to kind of allocate the risk between you and your suppliers and also your clients.
So again, I think it's a constantly moving environment. And two years ago you wouldn't have thought that some of the clients or the hyperscalers would be underwriting some of the risk of kind of the newcomers. And that's happening on a monthly basis right now.
Carl Fleming: I can say they've definitely changed from two years ago, So we're now structuring these to be a lot more focused on the energy side of things. I think the energy and the price of energy has become a major negotiating point in a lot of these data center development projects that are going on because it's kind of like going through the whole list again: What kind do you pick? Who are you gonna work with? What's the pricing gonna be? Is it gonna be intermittent? Do you have backup power? What's your connection or how long is this gonna take? What's the regulatory environment for this? Can you even do a co-location project in this state? That kind of stuff has really become a lot of the conversation with data center folks.
So there's a bunch of ways to structure that, right? You do see the same large private equity funds. They typically are, you know, already looking into or invested heavily into some form of the energy value chain, whether it's renewables, gas, what have you. They're also the same folks that are often investing heavily into these data center platforms.
But what's been unique recently, and I’ve had a number of conversations, is folks are really looking to do more joint ventures.
I think everyone sees that the hockey stick demand for energy is really outpacing the ability to flip the light switch on for these data centers. And honestly, energy has now become the biggest CP to getting the power to flow and the funds in the door.
Diana Goovaerts: We're talking about new funding vehicles here. So Nick, I know you have, uh, some experience or some information on ABS. What is that? And why does that matter in this space?
Nick Filichkin: ABS is simply asset-backed security. And what that really means is that you are investing into a physical, tangible asset and you're securitizing that. And when we're saying securitizing, what we really mean is giving the ability of multiple people or multiple entities to own that same asset.
So you won't necessarily have a security – there's only one owner. I mean, you can call it security, but it's really more of a security when you have multiple people, multiple firms owning that single security, that single asset. And it's asset backed, meaning there's a physical collateral. If the deal was to fail, you can always go ahead and liquidate or sell that asset.
That is an asset-backed security in a nutshell. The ability for multiple people to own one physical thing. T he way we do it, Compute labs, is through securitizing the asset, plus the cash flow and usage of the equipment on a revenue share.
Meaning that if a company, the borrower is doing really well, then the investor and the borrower is doing really well with it.
Diana Goovaerts: I have to ask, we're talking about some of these new unique deal structures, about trying to underwrite but then also offload some of the risks. And I don't want to compare apples and elephants here but I'm getting shades of the subprime mortgage crisis in terms of like getting creative to get deals done.
And so I have to ask the follow-up question: Are these creative deals a good thing for the industry?
Goncalo Bernardo: I think the creative deals, if they're structured appropriately, are a good thing for the industry. I think ultimately, again, there's good deals and bad deals and you gotta be cautious on which one you're going into. It's the job for the private equity capital to kind of really differentiate that and be willing to kind of underwrite its risks.
I think when we look at some of these deals, they're enabling companies to kind of get built up. Again, if you think of one of the largest players in the space, CoreWeave. CoreWeave was in there in the early days with the creativity before Nvidia and kind of supporting the player that they wanted to rise up.
Now, again, you've gotta be cautious. There's a lot of commitments going around. There's a lot of CapEx going around. There's a lot of risk effectively going around, so these risks don't disappear. So you need to enter these deals with open eyes and kinda really know what you're ultimately getting into.
And if you do that, I think ultimately you'll be okay. If you're in the mood which is ‘I gonna invest in data centers 'cause everyone is and at any cost’ then you're probably gonna get hurt.
Diana Goovaerts: So how do these kinds of financing or deal structures benefit non-hyper scale players? And I have to ask kind of a similar question to what I asked earlier: Are these a good thing? Are they more risky? Are we gonna end up with, you know, I hate to beat a dead horse with the analogy to previous interesting financing vehicles crashing the economy, but are we gonna end up having a bunch of data centers foreclosed on and GPUs sold off at auction? I have to kind of poke at that just a little bit.
Nick Filichkin: So the question sounds like to me, like, are we in a bubble? Is there enough demand out there? To my opinion, the short answer is yes. We're seeing so much demand and we're only seeing more and more increase of demand. We've seen the value of the equipment go up both on the primary market, meaning when it's being first sold by an OEM or manufacturer, and then also the secondaries market when people are selling it to one another.
We've seen the value of these assets go up astronomically, and what that tells us is that there's more and more demand. A lot of it is contracted, meaning that there are AI companies, small businesses, large businesses, enterprises that are looking to rent out this equipment for years at a time. It's getting really difficult for more businesses to access more compute, and thus the cost of them is just going up with demand.
Diana Goovaerts: So demand isn't going away, but are these kinds of financing beneficial to smaller players or are we kind of in a very, very roundabout way via the demand cycle just kind of finding new ways to fund hyperscalers if they're the ones who are buying that capacity?
Nick Filichkin: I love that question so much because it really hits at the point of ‘are only the few largest companies benefiting from this or are smaller players also benefiting?’ And that's actually the entire reason that Compute Labs built its business, is to make compute more accessible, both to small players, neoclouds, and also of course investors who want ownership and the rights to these assets.
For investors, it's quite straightforward. You own the asset. You own fundamental infrastructure. You own rev share. You own a lot of upside for the neoclouds. Generally speaking, neoclouds face so much difficulty with accessing financing and funding to purchase these assets. Those neoclouds can't get bank rates. They can't go to private credit and find the same type of rates or financing access that hyperscaler can. So they're constantly looking for alternative sources of funding, which is why we've built our business model to allow them to use the equipment on a revenue share, meaning that it's not fixed to a tight amount.
We're not necessarily relying on them to give us a fixed monthly rate. We know that the value of these assets are strong and that they will be used, which is why we're saying we're both in this together. You, the neocloud, get to use this. We're moving your heavy CapEx to an OpEx. We're making your books cleaner. You don't need to manage depreciation on these assets. We will go ahead and do all of those things and now you have equipment to serve your businesses, your customers, and grow your business.
Diana Goovaerts: Okay. I also wanted to touch on another element, and that is data centers are kind of now considered critical infrastructure in a lot of markets. And given the geopolitical landscape, does this change their risk profile? So, are there new risks that this introduces, whether in terms of threats or some other kind of economic or financial risk, or are there other risk elements that are higher priority that we should be thinking about more?
Goncalo Bernardo: I don't think data centers being critical infrastructure is new. I think it's more, people are more aware of it right now. This was a couple years ago, I remember sitting, I was on the board of a company that had a data center that it used for its own means and kind of also provided a little bit of colocation to for IT parties.
And during COVID they were open 24/7 and they were given priority status and that they needed to be up and running. So I think they've been critical for a long time and that's why kind of private equity and infrastructure went in as heavily as it did because they saw a lot of this service as a fundamental. What is happening right now with a whole geopolitical environment is it's changing the nature of the deployment.
So you have deployments that are a lot more regionally focused. You have a concept which NVIDIA's been pushing at a few of the countries as well of sovereign AI, where you want kind of the data to be kept and the processing to be kept in your country rather than kind of going abroad. So that is creating, I'd say, a change in the deployment pattern.
And that to me is kind of the key factor that is being affected by geopolitics.
Diana Goovaerts, Executive Editor, Fierce Network: Are geopolitical factors influencing where deals are happening?
Carl Fleming: I think historically we've seen the U.S. was probably already primed for immense growth relative to other jurisdictions. Most recently we've done some work and we're starting to do more work in the Middle East. We do a lot of work in Latin America. I do a lot of work in in Africa and some work in Europe. Markets like India and Japan have grown, which I think everyone kind of knows. Spain is growing a bit.
But I think in terms of like the actual geopolitical factors, there's been a strong push I think for ‘America first’ or even security over the past year, year and a half. So I think a lot of the data center development is taking place here in the States because we're trying to bring in all sorts of our own supply chain. We're trying to win this war against China in terms of developing data centers and our AI policies.
So I feel like the geopolitics brings it all back home for this kind of work. But I'd say probably the secondary market has been a decent amount of work through Africa.
Diana Goovaerts: So I have an obvious follow up here. Given the instability that's happening in the Middle East at the present moment, do you see that influencing the volume of deals that are going on there? Because that had been a very hot market until very recently.
Carl Fleming: No, it's a great question and I feel like the data center and the AI growth that's kind of happening, it's gonna happen regardless of all the numerous things that are going on in to the geopolitical environment.
The UAE and Saudi Arabia have had a massive initiatives to develop data centers and, and advance AI. I [] think those initiatives go unchanged. There's a need for that. There's power there. There's the, the abilities and the capital to do it. So I think that will continue in those hotbeds, if you will.
I think it becomes a little more of a conversation when you get to actual conflict type of zones, right? That's gonna be much different in how that rebuilding process goes over time. But I think the two that I mentioned there are probably areas that are gonna maybe feel this, just like the whole world is feeling this, little bit of uncertainty at this point in time.
But I think those are two that are gonna continue to develop and it's probably where I see a lot of people actually putting on a capital still and talking about it constantly.
Diana Goovaerts: Yeah, and both of those areas are ones that have government initiatives around that. So that's the kind of thing that doesn't necessarily change on a dime.
Carl Fleming: Correct.
Diana Goovaerts: Obviously there's a lot happening right now, but what do you see as the next phase of opportunity in digital infrastructure financing? Is it edge? Is it just AI to the max? Is it emerging markets or is there some other bigger story?
Nick Filichkin: I've seen a lot of buzz about AGI and I think a lot of people are scared about it, both on, of course, the humanitarian ethical perspective, which is very relevant and valid, and then I also think about it from an infrastructure buildup perspective. And I've heard some whispers how people are saying, ‘hey, if we have AGI, does that mean that we won't even need compute anymore? If it's so smart, it won't need the ability to inference more or train more.’
In my opinion, AGI only means that we're going to need to ramp up even more compute, because if we have something so smart, we're going to want it faster. We're going to even want it smarter, and we're gonna want it even better and more global. Just because we have AGI somewhere, an example, if we have one smart person, it doesn't mean that that person's intelligence is capped.
Intelligence can grow even further. It can get even better, faster and more people can become smarter.
So if hypothetically we do have AGI, I don't believe AGI is a cap or a limit. I think AGI is more like a middle introduction of, ‘cool, we got this new milestone, now let's build even more.’
Goncalo Bernardo: Number one that's in most people's mind is inferencing. And some people will call it edge, some people will call it inferencing. Inferencing is clearly a priority.
I think when we go from these big learning models into models whereby you are actually deploying them out in the real world and you need to be closer to the consumer and you don't need a gigawatt of capacity, but maybe a few megawatts or a few tens of megawatts are enough, I think that will be a big shift and that will be a big trend.
I think ultimately the thing that will drive the shape of the sector the most is what do people actually do with AI? And so where is the consumption? Is the consumption at their desks typing a question into ChatGPT or doing some sort of coding? Maybe. Or is the future that, again, your self-driving car talks to the network and it needs to be closer to the street?
So I think that will be the big kind of shift in deployment, and that's why inferencing is so important because at this stage. We're still early on in the AI adoption and [what] people don't know is do I need at every street corner or do I need it at every city or do I need it at every continent? So I'd say to me that's the biggest trend.
And I think the other one is really, it's really gonna be about power. The power grid has been somewhat sleepy for the last few years. And all of a sudden, the AI awakening has moved a lot faster than the power grid.
So I think to me, that's the next big that's happening already, which is people building their own power plants, whether they're called behind the meter solutions or island data centers. But really it is how do you continue to develop an AI power hungry solution in a power market that's not moving nearly as fast as you can?
Carl Fleming: I think my hot take would be somewhat biased. So I'm, again, coming from the energy angle. I think you see this ongoing as to how exactly these data centers should be powered.
You know, you have the ability to power these things on a massive scale from large nuclear generation. You can then go through and fire these things up with large gas-powered facilities. And then you have the renewables aspects that maybe are smaller footprint, but maybe a lot cleaner.
I think there's been a lot of articles printed recently and a lot of media coverage around just what are the impacts of this massive growth of data centers near so many cities and hubs. You know, is it good for us? Is it bad for us? Is anyone thinking through this on a macro level?
So I think my take, unsurprisingly, is that I'm more on the side of renewable energy co-location. Because it seems like that can often fill the gap, like I mentioned before, of this short, near-term need to get things online quickly but also in a way that's more sustainable.
As you look for other sources to come online in the next five and 10 years, this is probably the quickest and the cleanest way to do it. And I think the most recent policies out of the White House where they do say, bring your own generation – it didn't dictate what you should use or what you shouldn't use. It just says bring your own.
A lot of the tech companies already have these mandates to bring along renewable or clean power. So I think that's kind of where I see things going for the next three to five years, a heavy investment into those areas.
Diana Goovaerts: That is all the time we have for today. We'll leave it there. Thank you so much for your time and to all of our listeners and viewers, wherever you are listening or watching, make sure you like and subscribe, and we will see you next time.