Hello, and welcome to our webinar that we've titled "Powering the Future." Today we're going to be focusing on strategic data center and digital infrastructure investment in a capacity-constrained market. My name is Rich Hill. I'm global head of real estate strategy and research at Principal Asset Management, and I am your host for today's webinar. It's nice being on the opposite side of the screen for a little bit and asking questions of people rather than having questions asked of me.

So more importantly, who's joined with us today-- I'm pleased to introduce my colleagues Matt Hackman, managing director of portfolio management for US data centers; Sebastian Dooley, senior fund manager for European data centers; and then Jeff Matthews, managing director and head of originations on the private infrastructure team.

During this webinar, Matt, Seb, and Jeff will join me in a discussion that unpacks a couple of different things-- first of all, insights into data center and center and infrastructure sectors. But this importantly includes identifying what true demand versus AI hype looks like, how investors can manage for optimal outcomes, and then finally discussing the importance of infrastructure constraints, such as power, water, and more.

Before we jump into the webcast, I did want to spend just a couple minutes talking about our platform. We have been investing in private markets for over 60 years and currently manage over $95 billion of private assets. That consists of a combination of private real estate equity, which is around $55 billion, private real estate debt, which is a little bit more than $20 billion, and then private infrastructure debt, which is around $4 billion.

We are one of a handful of managers globally with a longstanding experience and a robust platform for investment solutions across private equity and debt, infrastructure debt, direct lending, investment-grade private credit, and real estate operating company investments.

A brief plug-- we did recently win PERE's 2025 Data Center Firm of the Year in North America. We view this as a great recognition of the work that we've been doing in data center space for nearly 20 years. For anyone that doesn't know, we did our first data center investment in 2007. And I think it really speaks to the strength of our market insights, our longstanding partnerships, and our disciplined approach to investing.

I would like to invite everyone to submit questions at any time during the webinar using the Q&A box on your screens, and we'll be glad to address those after our prepared comments. And with that, let's get started.

So when we were preparing for this webcast, I thought it was important to ask and address, why are we having this call? We think there is a lot of questions and even confusion about the data center and digital infrastructure market. It's OK to acknowledge that because the asset class has rapidly evolved over the past several years, and it's still evolving with increased nuance.

While many investors were either broadly bullish or didn't know the asset class existed 5 to 7 years ago, there's a lot more focus on it today. We want to discuss 5 key themes that we think investors should focus on to help separate fact from fiction.

So what are those 5 themes? First of all, is this an AI bubble? Number 2, how are data centers powered? Number 3, are all data centers created the same? Number 4, what are the debt markets telling us? And finally, are we actually seeing the emergence of a new third asset class?

My challenge as a moderator is fitting this all into a 60-minute webcast, because I think we could probably have Q&A for 45 minutes and maybe even an hour and a half on many of these topics. So I'm going to jump right in and start with the elephant-in-the-room question. Is this an AI bubble?

Matt, I'm going to start with you. Can you walk us through. Spending and CapEx forecast for the hyperscalers? I think probably your answer 12 months ago, 24 months ago, maybe even 6 months ago, is maybe a lot different than where it is right now. But I'd love to level-set with you. How are you thinking about the CapEx spend from the hyperscalers?

Thanks, Rich. That is something that's very much in focus for us at all times. I mean, we're always listening to those hyperscalers earnings reports on a quarter-over-quarter basis and keeping track of how much CapEx that they're reporting. I mean, when we're talking about the hyperscalers, we're really talking about 5 groups. We're talking about AWS, Google, Meta, Microsoft, and Oracle.

And following the recent earnings calls and guidance here during the first quarter, their 2026 CapEx projections are $715 billion combined across those 5 hyperscalers. This represents a 67% increase relative to the $430 billion that they had spent in 2025. While that's slightly below the 80% year-over-year growth that they had from '24 to '25, it is still quite a bit above where their targets had been previously.

And in fact, when we actually go back to the first estimates for 2025, their actual spend for 2025 ended up 27% above their initial guidance. So we might be seeing the same thing here again this year. CapEx spending is definitely something that the hyperscalers are putting a lot of time and effort into, and it's something that the analysts are very much watching as well.

So, Matt, I want to ask you a followup question to that. I think it suffices to say that valuations in the public markets are under pressure this year, especially AI-related stocks and AI-related stocks that are sort of AI-adjacent.

But interestingly enough, data centers have done really well in the public markets, with data center REITs up more than 20% this year. I know you're not a public equity analyst. You might like to play one on TV from time to time. But what do you make of all this, and what is it telling you about your business?

Yeah, that's a good question. So I think one of the ways to read into this is I think the data center REITs were a little bit more upfront last year about how much CapEx they were spending and what the intent of that CapEx is. Your Mag Seven stocks, obviously, are going to have a lot more analysts that are digging in on them and very much focused on how much CapEx is going in on those.

And so actually, this year, those big tech stocks that we highlight are actually down 8% since the beginning of the year, whereas some of those data center REITs are up quite a bit. Really, the biggest reason for this is that investors are concerned about the scale of the CapEx that they're putting out and the timelines for return on investment.

A lot of this CapEx is going into servers and builds of data centers that really aren't going to have a ROI in the next year or two. It really is going to take more time for that CapEx to turn back into profits for these big tech stocks.

Over the last 10 years, hyperscaler CapEx has averaged about 40% of cash flow from operations. And the 2026 guidance actually pushes that figure to near 100%. So this reduction in free cash flow for the hopes of future earnings is really causing a little bit of concern in some of those investors' minds when they're looking at those big tech stocks.

And just to tie it back in again, that's where we think that those data center stocks were a little bit more upfront and forward about the idea of we have to spend this capital now to have profits in the future. They really push on the idea of having the capital here for these facilities that will be AI-ready can really generate the profits on a go-forward basis.

I think we're going to get to that point in a little bit. But I like the contrast you were making, which is, as a former equity analyst myself, I think this is all about predictability of earnings. Today, right now, the market is valuing predictable earnings and income-driven total returns. And sort of boring REITs offer that, whereas if you have a spend out into the future, maybe there's some more questions about that.

Jeff, you and I sit in New York City together. We've had a lot of conversations over the past 6 to 9 months about this exact topic, AI bubbles. And I sort of liked the way you framed supply versus demand. There's still a lot of demand for data centers, but we are in a supply-constrained market, for various different reasons. So I'd love for you to just elaborate a little bit how you think about AI bubble risks in a supply-versus-demand context.

Yeah, thank you, Rich. Yeah, I think, overall, there's obviously tons of demand for hyperscale data centers. And that's not going away anytime soon. The supply, though, is structurally constrained primarily by power-- power supply and power generation.

And just naturally, the first point to make is when demand outstrips limited supply due to things like power, that's the first thing to point out, which is what type of bubble are you looking at when you have demand outstripping supply just by that one factor-- power constraint.

When you unpack it, you sort of have discussed it as sort of having a bifurcated result in the event of a bubble burst, meaning for all the facts that Matt just listed out, all the risks, if there isn't enough return on investment of the massive CapEx spend, we think that the Mag Seven and the big tech stocks could be revalued.

Worst-case scenario, there's a lot of off-balance sheet debt that's being used to finance the data center buildout and all that CapEx promise and historical spend. And that could translate, in the worst-case scenario, to rating agency downgrades on the bonds of the big data companies, which could affect their bondholders. This is all compared to debt and equity investors in physical assets-- so halo assets, hard assets with low obsolescence.

And data centers that are in tier 1 markets or tier 2 markets, in strong markets that have strong counterparties and contracts or tenant contracts that are long-term fixed and cannot be canceled, there is a floor to what an equity return would be in the event that those data centers are no longer needed, meaning the contracted fixed payment stream from the tenant to the data center itself is not subject to any variability. It's not subject to credit risk downgrade, per se.

So there would be a floor basically on what the-- floor to an equity return based on what the residual value would be at the end of those initial tenant contracts. And for debt investors that help finance these data center projects, that translates to low probabilities of default and especially high recoveries in the event of a default, but really well structured in a capital structure that only leverages against the more insulated cash flows that come from the sector.

OK. Seb, we're about 10 minutes in. I haven't come to you yet. I promise you I'm not ignoring you. You're going to have your hard questions thrown at you, including this one.

And maybe it's not such a hard question for you, because I've been in some meetings with you, and I think you have an elegant answer to this. How do you think about capturing AI upside within data centers, but also simultaneously mitigating against this bubble risk? I'm ultimately asking you, which we'll talk about a little bit more, but are all data centers created the same, and how do you think you maximize that risk-adjusted return?

Sure, Rich. Thanks. Good question. I think exactly as you say-- well, hinted at. Not all data centers are the same. There's a number of different workloads that tend to get done in these types of assets. And those different types of workloads actually have quite different requirements in terms of the physical attributes of data centers, namely location of these assets, but also how they're built out, what cooling systems are in place, what connectivity infrastructure they have got.

If we look at AI workloads, generally, these fall into two categories. The first one, which is where you're seeing the very significant investment currently, is on the generative AI model training, the creation of these AI algorithms.

These types of facilities tend to be much less restrictive from a locational perspective. And so you actually see that these assets are being built out in slightly more nontraditional data center locations. By investing in those types of facilities, you are exposing yourself, in our expectation, more to oversupply risks.

In case there is, say, a bubble scenario, demand for these types of assets drop off. What's your alternative use? It's going to be more difficult to change use for these facilities, whereas the next phase of AI to come through is the AI inference.

And this is as people start integrating AI algorithms into their corporate workflows, into their normal IT processes. And that type of workload, we're expecting to start to move into locations that work well for more traditional longstanding data center use cases, specifically, say, for cloud use cases.

Cloud data center assets have to, by their nature, work in very tight clusters. And you tend to find these assets sit in clusters, say, 15 to 20 miles in radius. And actually, it's in those locations where companies are already storing a lot of their data. They already have a lot of their corporate IT workflows going on. That's the most logical place to add additional applications, which are these AI applications, and take on the AI inference workload.

So from an investment perspective, you can invest in those types of facilities targeting that base cloud demand profile. That's what's earning your base case returns. But then as AI comes through-- which I think is going to be a rocky journey, but people are definitely expecting AI to come through in a more meaningful way-- then you start to see that additional demand profile layer in, and you can start to get some upside style returns from your investments.

So in short, play cloud data centers. They're predictable. They have stable cash flows. And then they have optionality upside if AI becomes a thing, which I think we all agree that it will. It's just a matter of when it becomes a thing. I think that's actually a really elegant way to think about it.

Hey, Jeff, I want to come back to you. You've been exceptionally patient with me over the past 12 months that I've been at Principal Asset Management-- in fact, to the day-- on how power is being used to power data centers. And the more I talk to you, the less I feel like I know.

So let's just unpack this a little bit more. Can you differentiate between the grid and the power draw? Because I think there's a couple different issues here that people like myself sometimes conflate. And I do think that there's maybe two things that investors need to understand. But if you can talk about how data centers power themselves from the grid, and then maybe the power draw, and separate fact from fiction for us, that would be helpful.

Sure, absolutely. It's certainly a complex issue. And Matt and Seb obviously have lots of experience in developing data centers and dealing with this exact question. The first thing is baseload power is critical. These data centers need reliable power at-- I think it's 5 9's, 99.9999, 6 9's.

So all these data centers today need a grid interconnection. They need that substation. They need to work with utility.

And as power comes online, it may not be connected to the grid yet. There needs to be investments in transmission to get intermittent power onto the grid. That can be there mixed with baseload power.

And on top of that, the transmission grid needs to be modernized. So I think, along with the big CapEx spend for data center development, there is equal or larger CapEx spend that's needed by utilities and private investors for transmission, grid modernization, new transmission, but, most importantly, new power generation for the grid. And there are several different grids in the United States, but for each grid in particular.

So that is a massive endeavor to source, develop-- source the equipment for and develop CCGTs, or Combined Cycle Gas Turbine power plants, that, in the US, is the cheapest source given the low price of gas and fracking nature of the United States.

So that's a 5-or-6-year lead time to build. Renewables are faster. Wind and solar with batteries are faster but intermittent power and not as reliable.

So generally, every data center now needs to have interconnection. And then, over time, they can build behind-the-meter solutions. And in the long term-- perhaps some now, but a larger market share in the long term-- behind the market solution-- sorry, behind the meter solutions can be the end-all, say-all, perhaps.

So you can be disconnected from the grid if you have reliable baseload power coming from a captive CCGT or other power source. So I do think that it's a tapestry of power generation solutions, starting with the grid and utilities and private investors building out the grid, modernizing the grid, second, adding on behind-the-meter solutions that is going to be-- speed to market will be renewables and batteries.

But the ultimate solution will be gas-fired CCGTs. And in the long run, everyone hopes that SMRs, or Small Nuclear Reactors, will come online at a cheap enough price point to be deployed modularly. That really solves the behind-the-meter problem. But it is certainly complex.

So there's a lot to unpack there. But you alluded to something that I think is really important and was somewhat eye-opening for me. And maybe I'm going to show how naive I am.

But I think it's really important to understand that the US does not have a grid. It has multiple different grids. And you can't necessarily power a data center on the East Coast with a grid on the West Coast.

Now, I won't go through each of them, but there are several distinct grids in the us that are not interconnected. The only one that is not regulated nationally is in Texas, which is ERCOT. That is within the state of Texas. So it's the only nonfederally-regulated one.

But different grids have different regulations-- and some are unregulated, some are regulated-- different system operators. And they have different reserve margins, low demand. And overall, what all these grid operators are dealing with all at once is from 2005 to 2020, power demand in the US grids, plural, was relatively flat.

And because of decarbonization-- because of-- sorry, electrification across-- first one was mobility, but more importantly, industrial electrification and onshoring of advanced manufacturing, and then now the data center explosion has just caused massive demand for new power.

So every grid is racing to add new power to the grid and to modernize the grid itself. So that's the consistency across the different grids. But you can't build a power plant in Texas and have that directly connected to a data center in Ohio or Washington, for that matter. It's just the way it's set up.

So I've heard 2 things from you. First of all, I think I'm hearing from you that everything's on the table in the near term. And over the longer term-- maybe I'm going to put words in your mouth-- maybe nuclear is the long-term solution with small nuclear reactors.

So my question for you, is that right? And then, number 2, is there a scenario where-- my words, no one else's-- we're both killing the energy transition over the near-term but accelerating it over the long-term?

Yeah, I really like that last question, Rich. Just on nuclear, you mentioned everything's on the table. I think everything is on the table. I think developers and financiers are being very innovative in how to solve the power constraint problem. I think it's a very complex and large problem that has its limitations. But the innovation is there. So everything is on the table, literally. There's talks about bringing back coal, which you can take your own positions on, as well as gas-fired and renewables and behind-the-meter and SMRs.

Also, nuclear power plants, utility-scale, are coming back online. That's a 10-year proposition. But if that comes to pass, that would add a lot of baseload power. That would be a huge solve in the long-term. That would be 10-plus years out.

I mentioned here that you are seeing the start, and, I think, just the beginning of consolidation between Big Tech and power. So IPPs or independent power producer platforms are starting to be looked at, to be acquired by Big Data and entities. And that may be a trend that continues.

In terms of nuclear over the long term, we talked about utility scale. SMRs is a new technology. But most importantly, it's about, can you get the cost down on a modular basis in a safe way by manufacturing?

Then the last question you asked about-- decarbonization versus digitalization. I think it's true that with all this new power focusing on gas primarily versus renewables, and just more power gen overall, is pulling away from the lofty goals, 5, 6, 7 years ago of going 100% net zero.

However, ironically, it is the data center pole itself that is, 1, raising power prices and PPA prices that make the renewables actually levelize cost of power to gas-fired. So it actually is more economically feasible with PPAs to build renewables with PPA prices rising because of the demand of data center builds.

And secondly, data centers need the power now. And wind, solar, plus battery is a much faster-to-market solution for power than a 5-, 7-year lead time to build a new CCGT. Now, some have been developed for years and are coming on sooner. But if you're starting now, wind and solar is actually financially feasible and reasonably fast speed-to-market solution to support the data center buildout.

So in some ways, it's a push-pull. There's more power generation, and that's not great. But it's keeping the renewable sector continuing through some of the subsidy being rolled off in the latest policy news.

So I want to transition to our third theme, which is, are all data centers created the same? I've spent a lot of time on this topic over the last 12 months or so. And I love having conversations with investors to unpack this. And by the way, my knowledge is nowhere close to the knowledge you have.

But, Matt and Seb, I want to start with a really simple question about US versus Europe and if they are, in fact, competing for the same megawatts or, in some cases, gigawatts?

Seb, you and I had a conversation yesterday that I thought was really interesting. Demand for European data centers seems like it's increasing at a time that demand for US data centers remains really high. So why is Europe getting more interest right now?

Sure. I think Europe has generally been very reliant on infrastructure sitting in the US markets or in the US geography. And I think increasingly that is becoming something that is being focused on by governments, by businesses, by individuals as well, who are ultimately storing data.

There is a much greater sovereign push currently going on within Europe to look to house European companies, European governments, and European individual data within European boundaries. And a lot of that is purely around self-reliance. At the end of the day, you can see how, in certain sectors, Europe has grown up being very reliant, especially on the US.

Increasingly, that's a more uncomfortable position. And there is that greater focus of, yeah, we're at a relatively early stage, still, in the development of this sector, to make sure that it's being built out in a way so that Europe can be much more self-reliant, which I think is a positive thing for everyone going around.

In general, though, the demand for the sector is just growing at such a rate. So the incremental shift away from the US is relatively de minimis compared to the incremental growth in demand for Europe, which then does come as a much larger percentage.

So, Matt, we're a developer in the United States. And I'd love for you to unpack what we're focused on, and specifically, maybe, drill a little bit more into cloud versus AI inference versus generative AI. These are all terms that I probably had to ChatGPT it at some point and say, what is Matt and Seb talking about?

But I think I have a better understanding of it now. Could you maybe just drill down what you guys do in the United States and where you're focused on across these 3 different pockets of data centers?

Yeah, that's a good question. So there's definitely different pockets of your traditional workloads, your generative AI, and your AI inference. And so your traditional workloads have been your cloud, your enterprise-based solutions, stuff that really has been the majority of the data center industry up until, effectively, the release of ChatGPT.

I mean, at that point in time, everybody started to hear about AI training models. And then ChatGPT was one of the first widely-known AI inference models. And so where we're focused, majority is going to be on primary, strong secondary markets.

These are going to be where your traditional cloud-based facilities are. This is going to be close to the population base where latency matters. This is always going to matter from your cloud-based facilities, because they always are going to have to be able to transmit that signal from the user to the data center and back to the user as quickly as possible.

When we get into that next phase of generative AI, this is here where we're now hearing about these multihundred megawatts or these gigawatt campuses that are being built in Bismarck, North Dakota, or Cheyenne, Wyoming, or really in the middle of nowhere where they can get access to cheap land, quick access to power.

And then, really, they can just put up a lot of data center capacity here. So those facilities, though, are going to be the ones where people have concerns about the multibillions of dollars that are going into these campuses. The models that are trained from these facilities, though, that goes back to the AI inference, then. These are now put closer back to the consumer.

And so really, what we're seeing is AI inference really kind of integrates back into your cloud-based facilities. And so if you have cloud-based facilities that have any flexibility at all of being able to handle the added capacity from inference, those are going to be the facilities that are going to benefit the most from the AI surge.

And really, those are the facilities that already have a very extremely tight and low vacancy rate due to all the demand from cloud. So continuing to focus on latency-specific locations in primary and secondary markets for us is going to be a big driver of added risk return.

Yeah, one followup question for me-- can you just talk through the returns that we're achieving when you develop a data center? I think a lot of people focus on stabilized data centers and the growth. But the returns are eye-opening for me.

Yeah, yep. So the way we look at developing data centers is really we're looking for a return over your cap rate, your development yield. And there's multiple different ways to look at this. I mean, there's powered land plays, where you can actually create a lot of return by just actually pulling the power and having power land available for data center development.

We in the US do a little bit of that, but I know our friends in Europe and Seb will do a little bit more of that focus just due to the even tighter constraints on power and land availability. However, when we're starting to think about it, too, then there's 2 real different types of data centers that you can build. You can build a powered shell data center or a turnkey data center.

And a lot of that is going to be specific based on tenant requirements. Some tenants are going to want to do the fitout and the design and have control of that themselves. And so therefore, they're going to want a powered shell data center.

Those yields are going to be a little bit tighter relative to a turnkey development, because we're spending 4 to 5 times as much money to build out a turnkey development relative to a powered shell. A powered shell is really just going to be that structurally enhanced warehouse with all the access to the power and the fiber for the data center user.

A turnkey one is going to have that exact powered shell, but then we're also putting in the generators. We're putting in the cooling units. We're putting in all the electrical rooms. We're getting it ready for the tenant to just move in their servers within the data halls themselves.

So the tenants are still making significant investments within these data centers. I would say, in general, we see-- and a lot of this really is going to depend on the actual credit rating of the tenant. But we would expect a powered shell data center to have a development yield of about 100 basis points tighter than a turnkey data center.

So again, added premium to develop the turnkey due to the added specialization associated with it, but then you really need to make sure that you understand the credit of the underlying tenant and the financeability and liquidity of that facility as well.

You're on mute. Rich, you're on mute.

Oh, there you go. Your host is on mute. Great. So I wanted to come back to you, Seb, real quickly and talk about availability zones before we move to theme 4. What is the importance of availability zones, and why does make it so attractive for investing in data centers that are located in availability zones?

Sure. Thanks, Rich. So availability zones generally are a concept that's very important for traditional cloud workloads. It's a specific region, typically within-- I mentioned earlier the, say, 15-to-20-mile radius, where you have specific data center assets that are run by a specific hyperscaler providing their cloud. And they're directly linked with fiber.

And that enables them to communicate in a very quick and efficient way so that they can share workload between the different assets in a very efficient way. And they can also provide redundancy in case you have something going wrong in one of the facilities.

And that's very important because, for cloud assets, the cloud in general needs to be able to provide very efficient, scalable compute, but also in a very, very, very reliable manner. And everyone-- it normally hits the news if one of the large hyperscale cloud customers has an outage in one of their availability zones.

So you tend to see that the data center market grows out around these very, very precise geographies and these very tight regions. And then also availability zones will club together to form an availability region where you've got multiple offer.

But yeah, these availability zones in themselves then are not born equally. Some are very, very well located with a huge amount of connectivity. That tends to be in the tier 1 markets, in the longerstanding markets. And in those locations as well, you tend to find that these hyperscaler can add on additional applications. They can sell additional services in these locations too.

So generally, they have more demand, and they can be more profitable in those locations. So you tend to find that there is a hierarchy as well for these different availability zones across different geographies.

And it's what really brings data centers, and especially the cloud, which is an incredibly misleading name in terms of everyone thinking it can happen everywhere-- it really brings these assets and these data centers into real estate 101. It's location that really matters. And you've really got to understand that to be able to understand [INAUDIBLE].

I probably have another 30 minutes of questions for you and Matt on this topic, but I do want to switch to theme number 4. I started my career as a debt capital markets banker. Don't tell the other three quadrants, but debt is still near and dear to my heart.

What are the debt markets telling us, Jeff, about the data center market? There's so much focus on CapEx spend that sometimes I don't think there's enough focus on, how do you finance these things?

Sure, well, the first thing to say, I think, is that the debt markets as a whole have proven to be extremely supportive of all this buildout and, I believe, will continue to be. So just unpacking this, there is a smaller market for preconstruction, prelease powered land financing. That's a smaller market.

But once you get to FID or the beginning of construction, bank loans, institutional loans, and other private credit opportunities, all are available. And the market's been very supportive in a mutually beneficial way.

So what we see from our team is a typical senior-- for an IG hyperscaler long-term tenant lease contract that's not cancellable data center opportunity, at the inception of construction, senior secured debt can finance 75%, 85% of the cost. And that is based on a future stabilized value of 50% LTV.

It's also really usually for the tier 1 customers and markets. That's a 115 sculpted DSCR coverage ratio. That is sized through the end of the first contract and perhaps the first renewal after that, which maybe there's some pushback against that recently. But historically, it's been giving credit to renewals.

Then during construction, it's very traditional to put on either second-lien or HoldCo debt. That's junior, which is very constructive to developers because that second layer of debt financing can bring you up to 100% or over 100% loan-to-cost, where you're increasing the debt load as you derisk the project, and still, then, based on a consolidated-- stabilized LTV of, say, no more than 75%.

So from a lender's perspective, you're lending against a halo asset-- hard asset, low obsolescence, a great tenant, a great developer, an owner. Your advance to the value is 75%. You look at dark value to make sure that you're not too much tied to the tenant and not the asset value itself.

And then lenders in turn, because the supply of data center financing is so large, the credit spread opportunities at first-lien senior secured debt has been pretty sticky and pretty solid on a relative value basis. And the same thing with second-lien or junior HoldCo debt, there's been a pretty stable market for that.

And innovative opportunities exist beyond that-- so equipment financing, GPU financing, turbine financing, mezzanine slices, DevCo, borrowing base platforms. The market's there for that. And I think the credit investors have a great risk-adjusted return opportunity with good returns here.

And the developers building out this data center boom are able to exchange that good value proposition for debt investors with getting a large percentage of the CapEx funded through nonrecourse debt.

So that's the basics. And I do think that's the construction financing or preconstruction through construction term. The long-term question might be next, but I'll pause there, Rich.

Yeah, so we have a lot of Q&A coming in, which doesn't surprise me. So I want to start transitioning to that. But before I do that, I want to finish with what I think is a really important question.

We've spent the last almost 40 minutes talking about how data centers and infrastructure are highly correlated together. And you can't have a data center unless you're powering it. So, Jeff, I guess I'll ask you a question, and I want to come back to Matt and Seb. But are we seeing the emergence of a third asset class that's neither real estate or infrastructure, but is, in fact, both?

So my take on it is that infrastructure and real estate, as two asset classes, very much is a Venn diagram, always has been, and continues to be more and more an overlapping Venn diagram. And then certain subclasses are moving in different rates.

But data centers is the perfect example of the dot in the middle of that Venn diagram. I mean, infrastructure, equity investors, real estate equity investors, and infrastructure debt, real estate debt, everyone is looking at the same opportunity set and looking at it positively.

So there could be small differences in it. But I do see it as either it's an asset class, or at the very least, it is a marriage or a cooperation between two asset classes working together on the same theme with the same goal.

I also think that infrastructure capital, to the extent it's viewed as longer-term, can be viewed as a good long-term home for some of the real estate development capital in terms of exit strategy. And I think that all the markets-- infrastructure debt across banks, institutions, and ABS markets-- will continue to see value in these halo assets long-term.

Matt, I'll ask you just a quick question. How much of your time is spent speaking to traditional real estate investors or traditional infrastructure investors?

Yeah, so, obviously, growing up in the commercial real estate side of this, I have a bias to the real estate perspective. But I completely respect Jeff's opinion on this as well. I've seen industrial buildings converted into data centers. It is a hard, tangible asset where we're signing a lease with a tenant.

Yeah, I mean, we're speaking to all sorts of real estate investors about this. And then it's amazing that after we finish that meeting, they're like, this is really good. We should bring in our infrastructure investors to talk to you as well.

And it's the same thing on the financing team-- on the financing side. It's like, hey, we're going to speak to our traditional real estate lenders. And they're going to be like, actually, you should talk to the infrastructure lenders. So it really does, to Jeff's point, sit in the middle of that Venn diagram.

Yeah, and so I'll give you last word. Do you think there's a best practice that when you speak to investors, is it good that they have a merger of the two disciplines? How do you think through that?

Yeah, I think if you're looking at the actual risks that you're taking when you're investing in the sector and the risks that you need to manage through, they are a mixture of both real estate-style risks and infrastructure-style risks. You need to have teams who are managing these assets who actually have a background, either just purely in data centers, so they understand intuitively both sets, or that mixture of real estate and infrastructure knowledge coming through to merge.

I think if you try to push forwards investment in this asset class with missing half of that puzzle, you can end up moving into situations you're not quite eyes-open on. And it can be scary down the line.

So I'm going to transition to Q&A in the final 15 to 20 minutes. I'm going to preface my questions by warning you guys that there is not softball questions coming. These are hard questions, but they're very good questions.

I'm going to start with what I think is a really important one. As you're thinking about risks, call it, over the next 10 years, what is the biggest risk? Is it that there will be a shift in demand during the long term pipeline of a data center? Or do you think it's going to be a decrease in the credit ratings of the hyperscalers that lead them to spend less?

I can ultimately ask you guys all these questions. But, Matt, I'm going to put you on the spot. How do you think about that risk? Is it that demand's going to go away from data centers or the hyperscalers are going to spend less? And maybe those are interrelated to some degree.

Yeah, I think they are a bit related there. I mean, I think the demand from the hyperscalers means that they are going to continue to spend on it. The nuanced part of that is, as publicly traded companies, they are still beholden to their underlying investors and their stockholders.

And so ultimately, those stockholders and bondholders are going to have some input on how much CapEx is going to be spent by them or risk a potential decrease in credit ratings associated with that. From a demand perspective, though, that still feels as strong as ever.

And really, the biggest reason we see that is when we're constantly talking to these tenants about, hey, is there any capacity available in '26? We're like, no. Is there any capacity available in '27? Yeah, very, very little, and it's being gobbled up very quickly.

And from that demand side of it, it is such a supply-constrained market as well. So that demand can never really be met in the short-term with just the power limitations, the long lead equipment limitations, frankly, the land and labor limitations, the specialized labor that goes into a lot of these as well.

So again, it is a little bit of a tied-together approach. But it does feel like the demand for these use cases is very strong, still.

Sure. Seb, as an American that lives in New York City, I hear and see headlines all the time about "not in my backyard" for data centers and community pushback. Do you see the same things in Europe? And then, Matt, I'd love to hear from how you manage that pushback. But, Seb, I'll start with you.

Sure. Yeah, without a doubt. You see that creeping up in pretty much every single major market across Europe. In general, these assets tend to sit where you see normal industrial. So the impact on residential has to be managed. But it's not the biggest challenge.

Where you have seen more recently more pushback come through is, for example, water usage. And where you get droughts hitting certain areas and there's data centers in those areas, that tends to get bounced back there.

In reality, you can build these facilities. So the water usage and water wastage is pretty de minimis. But it still is one of those points that does tend to come up. I think from our perspective, though, actually, these challenges in getting permitting and getting entitlements to move forwards on new data center projects actually does good things for projects that are already in progress.

The more difficult it is to create these assets and the more this NIMBYism continues to grow, actually, the better value preservation you have for facilities that are already in existence. And it just helps limit that potential oversupply risk in certain locations.

So, Matt, what would you add to that, and how do you manage that pushback?

Yeah, you're exactly right. I mean, these are added supply limitations. I think when we're looking to develop in a new market or a new area, one of the most important people you need to talk to is the local newspaper editor. I mean, it is public perception that goes a long ways here.

That NIMBYism is very real. Community engagement matters a lot. So with almost all of these new developments we're looking at, there's some level of community engagement. Are we doing scholarships with the local community college? Are we doing educational sessions? Are we providing artwork from the local high school to be put up into our data centers?

There's just so many different ways to educate the community about the tax revenue that can be available from this, the added benefits that we're putting into the infrastructure in the community. I mean, community engagement is becoming such a paramount point in actually getting data centers approved in these jurisdictions.

Hey, Jeff, I'm not going to let you off the hard questions. And we have a couple coming in for you, 2 of them in particular that I don't think are necessarily related, but I want to address both of them.

You're starting to hear a lot about off-balance-sheet financing for data centers and SPVs. I don't think that's a particularly new concept for commercial real estate. But can you walk through how you think about that? Are there risks? Is it misunderstood? How would you respond to that question?

Absolutely. So just on the last one, to me, answering from a capital markets macro perspective, the big risk, with all this construction financing happening in the buildout, there's going to be a multitrillion dollar refinancing wall in 5-- bracket 5 years away.

So I do think that everyone typically assumes the ABS market will be able to digest all of the debt that is used to finance all this buildout. I think, reality, it'll be a tapestry of different markets. But that is the one risk we focus on a lot and what the refinancings look like.

In terms of the SPV off-balance-sheet financing, from my perspective, which I don't see everything, that's the vast majority of how these works. It's nothing new. It's simply project finance.

Traditional project finance, you have, during the construction period, a construction firm that's providing a wrap of cost overruns and delay. Then you have, in this case, a tenant, otherwise known as an offtaker. And then you have a bankruptcy remote ring fence entity that's an SPV that is the borrower which owns the asset.

And everything hangs together through contractual frameworks. That's project finance 101. That's been done in energy and power, transportation, and now data centers for forever-- 20, 30, 40 years. Just because it's off-balance-sheet on the Mag Seven or the big tech hyperscale offtakers, will have to disclose that lease obligation as potential debt exposure to rating agencies over time.

So it's not like it's being ignored or swept under the rug. It's just the most efficient way to do asset-level financing from both the hyperscalers' perspective and, I think ours, because we want to be exposed to the asset, not necessarily to the corporate as infrastructure investors.

And I think if people hear headlines that say everything's off-balance-sheet and hidden, that is not the case. I think the case is that this is just being done in a project finance style, which is very normal and one-on-one.