REC Brian-Feldman-20260529 R02

[Brian Feldman] (0:00 - 0:24)

And so these are just absolutely colossal, you know, 10 billion plus dollar investments in a single data center campus. So to your comment earlier, like, the numbers really are, it's staggering how large these have become, but it's also, it's infrastructure. And that goes back to our comment of like, infrastructure or real estate, these are really long-term assets.

[Nancy Lashine] (0:26 - 3:07)

Hello, and thanks for tuning in to Real Estate Capital. I'm your host, Nancy Lachine of Park Madison Partners. Capital is a lifeblood of the real estate industry, but the decisions on where and how it's allocated are driven by people and personalities.

Who are they? What motivates them? What can we learn from their experiences?

On this show, we introduce you to some of the real estate industry's most influential thought leaders and decision makers. And we talk about what is important to them, how they make critical decisions, who has influenced them, and a lot more. Today's guest is Brian Feldman, Vice President of Development at Compass Data Centers, a leading hyperscale data center developer backed by Brookfield Infrastructure Partners and Ontario Teachers Pension Plan.

Compass is one of the industry's most active hyperscale campus developers with a track record stretching back to 2011. Data centers have become the defining story in institutional real estate. Last year, data center investment surpassed office investment in the US for the first time, a remarkable milestone, and the largest hyperscalers collectively spent over $350 billion on data center infrastructure in 2025 alone.

Individual campus projects now routinely represent 10 billion or more in capital, a scale that rivals the largest real estate developments ever undertaken. Brian is at the forefront of these projects. He spent the early part of his career at AWS and Meta procuring and planning data center capacity for two of the world's largest hyperscale operators before crossing to the developer side at Compass.

That experience on both sides of the table gives him a grounded perspective on how this market actually works and where it may be headed. In our conversation, Brian provides a peek behind the curtain from how a hyperscale developer underwrites a site, to why power has become the single most important variable in the development equation, to how Compass thinks about building assets designed to last 100 years in an industry where the technology changes every three to five. We also talk about lease structures, capital formation, and the obsolescence question that's on every investor's mind.

Brian makes a compelling case for why the underlying demand for computers about as durable a thesis as exists in real estate today. I think you will enjoy this conversation with Brian. Brian, it's such a pleasure to have you on.

You know, when I'm thinking about, if you were, when you were 12 or 13 years old and thinking about what do you want to do when you grow up, did you even know what a data center is?

[Brian Feldman] (3:09 - 3:20)

Maybe I'm not as old as I look. I think when I was 12 or 13, none of us knew what a data center was. I don't know if a data center even existed.

No, but it's funny to like, the path life takes you on.

[Nancy Lashine] (3:20 - 3:37)

You're in the hottest sector that anyone who chose either a tech career or a real estate career could possibly be in. I think lots of folks are really interested to hear your story and your career path is so interesting. So maybe just start, just start, tell us how you got started in the real estate business.

[Brian Feldman] (3:38 - 4:37)

Yeah, yeah, I just feel very fortunate and lucky that everything led to where I am now. And it was completely by accident. I, so I went to Columbia Business School in New York with the intention of focusing in real estate and finance.

And I had a couple of different opportunities in front of me of what to do next. And I ended up boomeranging right back to Amazon where I was before business school, but with AWS in a role that was strategic long range planning. And so at the time, I didn't love it because it wasn't real estate transactions, which is what I really wanted to do.

But what it was, was putting all of the pieces of digital infrastructure together and figuring out strategies for expansion, strategies to deal with power constraints, deal with fiber constraints, deal with data locality, data sovereignty, all of these things that at the time I was going like, man, I wanna be buying land. Like I wanna be doing real estate transactions. But now I think back to that and it was so formative of just figuring out.

[Nancy Lashine] (4:37 - 4:38)

What years was that?

[Brian Feldman] (4:38 - 4:39)

This was in 2017.

[Nancy Lashine] (4:40 - 4:40)

Oh, wow.

[Brian Feldman] (4:40 - 5:05)

So this was 2017 to 2018. So I've been in the data center space for- Eight years ago, nine years ago. Ancient history now.

Then I transitioned to more of an execution focused role where in addition to a lot of the strategy stuff, I'd actually go on the ground and procure the data center capacity. So get the kilowatts and get the megawatts for new product launches, new region launches, and some strategic initiatives for AWS across the Americas.

[Nancy Lashine] (5:06 - 5:12)

Can I stop you there? What does that mean? How do you get on the ground and procure the capacity?

What were you actually doing?

[Brian Feldman] (5:13 - 6:40)

So it can mean a few different things. So from my context, I was just speaking about is going out and buying it. So just like you would go lease an office building, say Amazon needs 200,000 square feet of office space or industrial space to achieve some end goal, same thing with data centers.

We need five megawatts to go launch this new product in X city. So it's going out, it's running the RFP process, it's meeting with the developers, meeting with the landlords, wrangling all of your internal stakeholders to figure out how to rank order each option. And I'm happy to, we'll get into that a little later, but it's, and then it's buying it.

So then it's negotiating the lease agreement. Data centers, it's a little different. You don't think about things in terms of square feet.

You think about things in terms of kilowatts and megawatts and now gigawatts, which is still insane, but it's a transition. And so that's one way you procure capacity. Another is you buy land to develop a data center.

So that's longer term. And so you're buying land. And then equally as important is you're procuring the power contract to energize that land.

And back in 2017, 2018, things were multiple factors of magnitude smaller and utilities and community were extremely happy to have the data center growth. So you'd call them up and say, hey, I have 50 megawatts. They'd go, amazing.

Let's start drafting an ESA. How can we help you? Very different story today, obviously.

[Nancy Lashine] (6:41 - 6:46)

When you would buy land though, what was important about where you located the land?

[Brian Feldman] (6:47 - 8:41)

Yeah, it's a complicated question. First is the hyperscalers sets companies like AWS, Meta, Google, Microsoft, Oracle, maybe some others like NVIDIA, CoreWeave, some of the other Neo clouds. They're not data center companies.

They're building data centers to solve a business requirement that requires that infrastructure. So your first piece is, what are we solving for? Is this, are we launching a data center in Canada to capture Canadian workloads?

Or are we doing another data center in Virginia for government workloads? Or is it, I just need 200 megawatts anywhere as fast as possible, completely location agnostic. So that's your first step is like, what does the business need?

And then that forces you to draw a boundary on a map. Oh, then it's more simple. It really comes down to timeline.

So how quickly can I develop a data center? What are my longest tent poles? Is it entitlements?

Is it power, which is most typical today? Is it my supply chain? Is it some other factor unique to that site?

What does my power situation look like? So how many megawatts can I get on what schedule? Again, influencing timeline.

And then your third dimension is cost. Is it, what is the upfront cost of purchasing the land, of developing the data center? What is my power cost?

What is my total cost of ownership? And what is the NPV of doing all that to the business? Or maybe it's net present value or net present cost.

And how does that compare with option A versus option B versus option C? So you gather all of these data points to try to put together that strategy. And then the hyperscalers, you go to approval, which is very much like an IC memo, and you hopefully get approval to execute that.

Then you go on the ground and you do it.

[Nancy Lashine] (8:41 - 9:02)

How is your perspective though? So now you're working for AWS. So you're basically an in-house procurer of services needed.

So how did you think about data center development in those days? It sounds a little bit like you're thinking about it. You're describing how you're thinking about it today as a developer at Compass Data Centers.

[Brian Feldman] (9:03 - 11:02)

The biggest difference between the two at a hyperscaler, again, they're not data center companies. They are building this infrastructure to support web services, to do social media products, to host cloud services, whatever it may be. And so because of that, it's a fixed cost center.

And so really in the P&L of these broader business lines, you're on the L side. And that means you don't have as much license to be creative. There's a lot more downside risk than there is potential upside.

Because let's think about it this way. If I need to go get 400 megawatts of capacity, if I go get 400 megawatts of capacity, I've done my job. You know, like I'm meeting expectations.

We are providing the business with what they need. If I go get 300, massive problem. The business is constrained by the infrastructure.

So that means we're leaving dollars on the table. That is not good. But on the flip side, if I go get 500 megawatts, well now I'm paying for capacity that we don't need.

So our P&L is again misaligned where I've now done too much and we need to shed that capacity or any number of different possible scenarios. So you really are bound by what the capacity requirement is and how that's changing over time. On the developer side, you know, if I go and I procure a 600 acre site and I can get 800 megawatts of capacity there, Compass makes a tremendous amount of net enterprise value from that.

So I have a direct impact on the bottom line of the business. We are a data center company. That's what we do.

And so we're okay taking risks. Like we're gonna evaluate them. Obviously we're gonna look for the highest, best and risk adjusted return.

We're going to try to get the highest return on our invested capital that we can. We're gonna go take big swings and we're gonna do that because that's what we do versus a hyperscaler which is a social media company or a search company or whatever, however else you want to define them.

[Nancy Lashine] (11:03 - 11:22)

So implicit in that is the answer to this question but let me ask it explicitly. Why did you leave? You left AWS, you went to Meta and then from there you left, you did some other things and ultimately ended up at a data center development company.

Why all that transition? What was your thought process?

[Brian Feldman] (11:23 - 12:20)

I follow opportunity when it presents itself. I think when opportunity knocks at your door you're a fool not to at least answer it and see what is potentially there. That's how I got into data centers in the first place and what's led me to Compass which has been fantastic.

I think I ultimately always wanted to go to this side of the fence. It's again like the real estate, the strategy, the everything from the land acquisition to the entitlements, to the energization, to the development, to the leasing. All of that is core to our business and we have to excel at every single portion of that or we won't be successful in the industry.

And so for me and like a lot of others it's sort of a natural progression where you start at one of these hyperscalers for a long, you stay there for a few years, you learn a lot of things and then a lot of folks like myself flip to this side of the industry which is frankly it's a lot more fun, you get a lot more creative license, you get to do stuff that's really impactful and you get to wear a lot of different hats.

[Nancy Lashine] (12:20 - 12:25)

How many data center development, how competitive is the data center development business?

[Brian Feldman] (12:26 - 13:07)

Extremely. So there are hundreds of data center developers in the United States alone. You know and there's a very long tail that do anything from just one asset to powered land sales.

At the end of the day for a data center operator like Compass or my competitors, in one sense it's pretty commodified. Like we sell space, power, and cooling. And what I sell, kind of like office buildings, right?

Like a Tishman Spire development is gonna be very undiscernible from a related development, from a Heinz development, like it's all office space. Data center.

[Nancy Lashine] (13:07 - 13:12)

Ooh, some people are definitely gonna take issue with that but we don't have to do that today.

[Brian Feldman] (13:13 - 14:39)

Well, I'll get into the differences in a second but for the end user, it's office space. Same thing with the data center is it's data center space. It's energy capacity and cooling capacity.

And so within that, like for a hyperscaler, there's very little marginal difference as to whether you lease from company A, B, or C. But within that, and sorry I'm gonna contradict myself now, there's probably 10, maybe 20 really reputable data center developers that have done large amounts of capacity with multiple hyperscalers in multiple regions that really capture a lot of the business. Now the field is getting more and more crowded every day and I just mentioned Related.

Related launched Related Digital a couple years ago. They're a new entrant in the space. When there's dollars to be made, when there's spreads this high, it attracts a lot of capital and a lot of attention.

Fortunately, the demand has just been unsatiated and a rising tide lifts all boats. But to your original question, it's extremely competitive. And so there's 10 or so companies that I would say are blue chip data center operators that make up the majority of the leasing with hyperscalers.

And how did they get there? They entered the market earlier, so they made a lot of the mistakes early. And then they know how to operate so that you get repeat business from the same five customers really that we all lease from.

[Nancy Lashine] (14:39 - 14:58)

Let's talk a little bit about how you underwrite these deals. How do you think about, how do you underwrite them? How do you decide where to buy land, how to price it?

When do you start to negotiate with a potential tenant and how do you, and then we'll get into the whole risk aspect of it.

[Brian Feldman] (14:59 - 17:29)

Yeah. So the first question is, data centers like to be next to other data centers. So you've got a very sure thing.

And actually the reason behind that is, data centers are all about interrelational databases and interrelational compute. And so if I have a facility that can be right next to another facility, that means they can talk to each other almost instantaneously and really almost act like one greater facility. And then if I have 10 of them together, same thing.

Then it just becomes like expansion capacity to keep that site growing and growing, even though it's in physically different buildings. So the closer I can be to existing facilities, the better. However, for the last several years, power has not been readily available.

And the ability to do that is becoming more and more and more difficult. So you're really doing like your broad site selection is really now, where is power available? And so I'll talk about underwriting in just a second, but this is how you get there.

Where is power available? How far away are we from major long haul fiber lines? And what latency does that equate to relative to other major data centers?

And then does that state have, if we're talking about the United States, are there data center tax incentives? Because that really changes the dimensions we talked about at the beginning of total cost of ownership and just your net present value or your net present cost. So that's gonna drive you to certain markets.

So how do I underwrite something is I'm really looking at my return on invested capital. And I'm looking at it to say, okay, what's the cost of my land acquisition? Compass is a very disciplined company where we build basically the exact same facility.

And it's designed in such a way that it'll work in Arizona just as well as it'll work in Alberta, Canada. And so our costs for our facility are extremely controlled because our supply chain is developed well in advance. So there's very little variability between building a data center in Phoenix versus building a data center in Chicago.

So because of that, my building costs are pretty prescriptive. Of course, there's gonna be variabilities with labor, local tax regimes, climactic conditions, things like that, but pretty standard. So I'm underwriting the land.

I know what my building costs are. I'm gonna fudge those variables a bit and ultimately get to a return on invested capital. The last piece is market risk.

And your question of like, when do we start talking to customers?

[Nancy Lashine] (17:30 - 17:33)

But sorry, what return on capital are you working towards?

[Brian Feldman] (17:34 - 17:47)

I mean, you wanna be sure you're above at least 8%. And high single digits, low teens are attractive. You know, obviously the higher, the better, but- And these are unlevered.

[Nancy Lashine] (17:47 - 17:57)

So- But it's interesting, you wouldn't develop an office building at an eight or even a 10% yield on cost. You wouldn't develop much of anything. You wouldn't develop anything maybe residential.

[Brian Feldman] (17:58 - 19:12)

You have to remember this is a, at this scale, it's really more like an infrastructure play, where you're deploying hundreds of millions, if not billions of dollars of capital into a single asset or a single asset campus. And the players who are doing that are again, infrastructure. And this is a really fun topic we could get into, but are data centers real estate or are they infrastructure?

And the answer is very much both. And it depends on which type of asset you're looking at. If you're looking at a carrier hotel, so, you know, a legacy data center facility, probably in a downtown facility, it's still both.

But in my mind, that's more real estate because the location is really driving a lot of the value. If you're looking at a hyperscale data center campus that's in rural Texas or outside of Phoenix or in rural Virginia, that's very much more on the infrastructure side where you're deploying so much capital likely from your equity partners who are infrastructure funds, they're expecting an infrastructure level of return. Now, I'm sandbagging a bit.

Ideally, you'd like to have your returns substantially higher than that, but that's sort of your bottom threshold to make things work.

[Nancy Lashine] (19:12 - 19:27)

Can I just, a sidebar for a second? Yeah. If return on capital in the U.S. in most markets is somewhere between, call it eight and 12%, depending upon where and what, what might they be in less developed markets outside this country?

[Brian Feldman] (19:27 - 20:34)

Yeah, so it depends, obviously. You're gonna look for a risk-adjusted return. And so your geopolitical risk, your leasing risk, your development risk, your construction risk is all gonna be very different in Germany versus Vietnam versus Indonesia.

And different players have different strategies for how they approach their capital sources and how they approach that return on invested capital. Compass, we have some stuff in Europe. We've done international deals in the past.

For the past couple, we've really been focused on the United States because we feel it's the best risk-adjusted return far and away compared to any international market. And that's why you, and we're not alone in this thinking, that's why you've seen 60 to 75% of all data center growth happen in the U.S. But to answer your question, it's risk-adjusted return. And so you're gonna price in those risks, determine what your spread is, and then look for a return that's coming through it.

Ideally, you'd command a much higher leasing rate in one of those more risky markets than you would in the States from a hyperscaler to compensate you for that risk.

[Nancy Lashine] (20:34 - 20:41)

Well, you also, you have, I mean, the costs are gonna be a lot less. And you have more available, like in Asia, you have more available electricity.

[Brian Feldman] (20:43 - 22:05)

Yeah, exactly. And there's something that he said about everybody knows their own home market the best, like from a regulatory standpoint, from understanding of risks versus opportunities. And there's something to be very said about that in data centers as well, but there are a number of very good global players.

And then the best players are our customers themselves. And you have to remember, like, when your question is, is the data center space really competitive? Microsoft, Google, Meta, Amazon, Oracle, they all build a lot of their own data centers.

Their portfolios are skewed anywhere from 60 to 70 to 80% owned. And so they're only going to go lease more or less when they have to. And so because of that, your biggest competition is also your biggest customer.

And that's even more so when you're going into a new market where they're gonna decide, hey, should we buy land and partner with a local developer, pay them a 4% development fee, and then own this? Or should we go lease capacity from a provider? What's better on a risk return standpoint?

What gets us capacity faster? And for the hyperscalers, they're ultimately thinking, what allows me to achieve my end goal the soonest, the most cost-effectively, and with the least amount of risk? For a developer, we're saying, hey, where do I deploy my capital for the highest risk-adjusted return?

Is that in Vietnam or is that in Ireland? Or is that in the US?

[Nancy Lashine] (22:06 - 22:09)

So when do you start to talk to tenants in this process?

[Brian Feldman] (22:09 - 24:17)

Yeah, so there's a lot of different strategies with this. And I'll talk about compasses and I'll kind of give a brief overview of the options that are available in the market. So we won't begin actively marketing a site until we have certainty of timeline and certainty of power.

And so that means that we've likely closed on the land, or we're very close, we've signed an energy service agreement to bring power to the site, or we're very close and extremely confident that we will, and we know exactly what our timelines are. Because that's what our customers are ultimately buying from us, is they're saying, hey, we'll have 100 megawatts available on X day, and we'll have 200 megawatts available on Y day. And if we don't have extreme confidence in being able to provide that, we're not gonna market it yet.

We're not gonna be actively seeking lease negotiations. Some people do a earlier kind of look, where they'll say, you know, and they basically do it at risk and say, hey, we've got this site, we think we can get this much power, let's start preliminarily negotiating a lease agreement, and as that materializes, we can always adjust. There's also an entirely different model in the industry called a powered shell model, which is where, let's say I have acreage, I can get energy to that acreage, I'm then going to court the hyperscalers to basically have them give me their basis of design, give me their specifications, and then build them the exact data center that they want, which I will then lease back to them.

The primary difference there is, you know, that takes longer, like they're gonna come in much, much earlier in the development cycle, the returns are typically a little lower, and then often the mechanical, electrical, and plumbing gear, which is the majority of the capital outflow of the data center, is done by the hyperscaler themselves. So you're just offering them a shell with power built to their design, and then they're gonna come in and outfit it. So you have lower construction risk, lower operational risk, low capital requirements, but again, you're generally, because it's lower risk, you're yielding a slightly lower return.

[Nancy Lashine] (24:17 - 24:31)

You were talking, Brian, about tax incentives. Can you tell us a little bit more about what kind of tax incentives different, is it states or municipalities offer, and how much of a difference that can make in the rate of return of a data center?

[Brian Feldman] (24:33 - 27:21)

So a great question, and this is a very hot-button issue currently, is data centers are on the forefront of a lot of people's minds all across the country. So sales and use tax exemptions are really the big driver. And what that means is, if I'm building a 100-megawatt data center, it probably costs around, it historically is about 10 million a megawatt, today it's likely closer to 12 to 14 million a megawatt, and that's just for the structure itself.

If you layer in the actual cost of the data center, IT equipment, so the actual chips and servers and racks inside the facility, that cost can double or even triple. And so if you're paying anywhere from a fixed to 10% sales tax on all that equipment, that quickly adds up to hundreds of millions of dollars. And so historically, certain states around the country, around the US, have passed sales and use tax exemptions, and this drove a lot of the data center growth in Virginia, in Arizona, a lot of data center growth happened in Oregon because there is no state-level sales tax, same thing in Georgia, and in various other markets.

And so back over the last, call it decade, those sales and use tax exemptions drove a tremendous amount of growth to those specific markets. Because if I'm a hyperscaler, and my net present cost is $300 million lower going to Arizona instead of California, I'm gonna go to Arizona 10 times out of 10. Unless there's a very specific reason due to latency or a customer or something that I need to be in California, I'm gonna take all of that fungible data center capacity, put it in an area with, let's put it in a state with a sales and use tax exemption.

Fifth piece is property tax. So in order to incentivize investment, some local municipalities will offer a property tax abatement. That has significantly diminished.

Sales and use tax exemptions are now a very hot-button issue politically where a lot of constituents are upset, going, hey, why are we giving these big tech companies millions of dollars of breaks? Why are we offering this? And of course, there's a lot of different attitudes around this across the spectrum, but it is changing.

Some states are still passing and contemplating sales and use tax exemptions to incentivize more data center growth. What I would say is if you look at the communities where large amounts of data centers have located, they very likely have very nice parks, very well-funded police and fire departments because data centers throw off an incredible amount of revenue. And even with state-level tax exemptions, you're still making millions upon millions of dollars of investment in the local community and still driving a much, much higher tax base.

[Nancy Lashine] (27:22 - 27:26)

When you say they throw off lots of revenue, in what respect?

[Brian Feldman] (27:26 - 28:23)

So if I'm building a data center campus, I'm employing thousands of construction workers. All those construction workers are making money, they're paying income tax, they're contributing to the local economy by either living nearby, eating nearby, doing other activities nearby. After the data center is stabilized, you're still employing a pretty decent number of people to work in the facility.

Now, it's not as much as a fact. So it's job creation. Job creation is a big part of it.

Even with tax incentives, there's still likely a lot of money flowing into the local community through a tax base, be it property tax, be it you'll have a property tax abasement up to a certain level, and then above that, you'll be paying taxes. And these are multi-billion dollar facilities. So even with a very low tax base, for a lot of these smaller municipalities, the data center becomes the largest tax-paying entity by a factor of magnitude.

[Nancy Lashine] (28:23 - 28:26)

Is there a time when robots build these data centers?

[Brian Feldman] (28:27 - 29:20)

David, that's fun, or I'll say, I don't know. You're hitting on a real thread though, which is there is a labor shortage. And labor is one of the factors you have to consider with a new development.

Like labor, it's always solvable, but it's how long does it take and how expensive is it going to get? Is this a 20% premium? Is it a 50% premium?

Does it add six months to my development timeline? Does it add? And so you're balancing those.

And so I think there's more than enough people who are in the space, who are jumping into the space, and you can make a very, very, very good living, and a lot of people are, and it's in very high demand, which is attracting more workers. It's kind of like the data center in space question. Is it going to happen?

Yeah, it could. There's a lot of stuff that has to happen first.

[Nancy Lashine] (29:20 - 29:26)

To train though, to be a skilled data center construction worker, what's the skill sets?

[Brian Feldman] (29:28 - 30:21)

A lot of it's the same as other types of real estate construction. Concrete workers, you've got steel workers. For data centers, there's a lot more emphasis on mechanical components and electrical components.

There are a number of technical colleges all across the country, of which some Compass is sponsoring, that have data center career tracks. So you get exposure to equipment from our largest suppliers and our largest OEMs, which are really predominant in the industry, so that you can go out and immediately start servicing gear from Vertiv, or servicing gear from Schneider, or Siemens, or any of the other large players. That track is really accelerating because it's a great opportunity.

It's vocational programs around electricians, around plumbers, around mechanics that can do a lot of the servicing and ongoing maintenance.

[Nancy Lashine] (30:22 - 30:38)

You talked about having your supply chain already lined up when you actually start it by the land and start construction. Where does most of the hardware come from? Is it made here in America?

Is it made in Asia? Yeah. How does that operate?

[Brian Feldman] (30:39 - 32:15)

So interesting question. That's another piece of the data center boom that people often don't talk about, is as we're building more and more gigawatts of data center capacity, that means I need gigawatts of transformers, of cooling units, of power units, of battery packs, of generators. All of that also has to be built.

And so you're seeing a boom in manufacturing as well for data center equipment in the US, in Europe, in APAC. And so to your question, Compass is a little unique where we partner with a couple different suppliers and we sole source our equipment from them. So we partner directly with Vertiv, with Schneider, with Siemens, and with Caterpillar.

And we de-risk our supply chain by basically pre-ordering a certain level of megawatts continuously year after year. So there's assembly lines fully dedicated to Compass that are just churning out the product that we need. And then we're deciding, okay, 50 megawatts goes here in Virginia, 50 megawatts goes here in Chicago, 50 megawatts goes here in Mississippi.

And that way we have continuity of supply so we don't have any variability from our supply chain and our delivery dates. And so again, once we have confidence of power and confidence of development, we know that the supply chain is taken care of on the back end and it's de-risked. It's risky in a different way in that we're sole sourcing from certain suppliers and something could happen to them.

But we think we've sole sourced with the best suppliers in the industry and it hasn't worked out well so far as most others use these brands that are very familiar with them as well.

[Nancy Lashine] (32:16 - 32:19)

Does that say there's inventory on your balance sheet until you use it then?

[Brian Feldman] (32:19 - 32:33)

The idea is the moment that we receive, that we take receipt of it, it's got a home to go to in a data center. So if we ever got to the point where we have a stockpile, yes, but that hasn't happened yet. We thought there may have, it's not anymore.

[Nancy Lashine] (32:33 - 32:33)

Right, right.

[Brian Feldman] (32:34 - 32:34)

Yeah.

[Nancy Lashine] (32:34 - 32:39)

And are they mostly manufacturing in the US or all across the world?

[Brian Feldman] (32:40 - 33:14)

There are components that are sourced globally. So Vertiv is based in the EU, Siemens is based in the EU, Schneider is based in the EU. They do a lot of their manufacturing domestically and it's kind of cool at our DFW campus outside of Dallas in Red Oak, there actually is a Schneider production facility.

So the units that we're using in our data center are brought there as pieces assembled on site and then delivered right next door. And once campus is complete, they'll continue to use that assembly line to export to other data center campuses in the DFW region.

[Nancy Lashine] (33:15 - 33:23)

Got it, interesting. So tell us, talk to us a little bit about lease term and how you think about lease term and duration risk.

[Brian Feldman] (33:24 - 34:20)

Yeah, so this has also evolved, you know, with my time in the industry, I imagine it will continue to evolve. For the scale that we're talking about with multi-hundred megawatt data center campuses, multi-billion dollar capital outflows, lease term now market is probably 15 years. It's probably 15 years with one or two five-year extensions.

Maybe we'll see 20-year lease terms in the near future, but market today is 15. The reason that's so important is it's effectively a stabilized cashflow from a very credit-worthy tenant over that entire 15-year period. And so your ABS spreads are very attractive to a whole suit of potential buyers because it's, again, it's long-term duration, it's stabilized cashflow from a very credit-worthy tenant, and it's, once you're through the construction and development phase, incredibly de-risked.

[Nancy Lashine] (34:21 - 34:26)

And typically what sort of yield are you generating over those 15 years as an investor?

[Brian Feldman] (34:27 - 35:00)

You know, it's a good question. As an investor, I'm probably not qualified to comment on it because it can run the gamut. Again, it's likely going to be some spread over treasuries because, again, it's effectively a 15-year bond where if Microsoft is the tenant of the data center or if Meta is the tenant of the data center, they're signed up to a 15-year lease obligation.

You're buying a 15-year cashflow with Meta as your tenant. So whatever that spread commands above base rate is what the market dictates at the time based on the credit-worthiness of that tenant.

[Nancy Lashine] (35:01 - 35:06)

Right. And so maybe you get your capital back through the 15-year cashflow and maybe not.

[Brian Feldman] (35:07 - 35:13)

I think they're probably getting high single-digit returns is, again, what the target is.

[Nancy Lashine] (35:14 - 35:17)

Tell us, how are you capitalizing your data centers?

[Brian Feldman] (35:18 - 35:37)

Yeah, so Compass, we basically have two equity partners, Ontario Teachers Pension Plan and Brookfield Infrastructure Partners. We use more platform-level finance, though, and not project-specific. And we have a $6 billion credit facility.

So we purchase land. I know the numbers are just staggering, right?

[Nancy Lashine] (35:37 - 35:38)

Staggering, just staggering.

[Brian Feldman] (35:39 - 38:05)

And so we would purchase land either with cash on our balance sheet or through that credit facility up to a certain point of leverage, whatever makes the most sense from a cashflow optimization and corporate finance standpoint. And then we go out and we get construction debt on it. And that construction debt could be from our equity partners or it could be from the open market or it could, and very likely now, it's just through our credit facility.

And then that finance to the construction. And we only build multi-building campuses. So very likely you're gonna be close to stabilization on your first asset or your first few assets while you're still constructing the subsequent assets.

And then you have a decision if you wanna go with the CMBS route or an ABS route to then sell the equity of that stabilized asset off. Historically, we've been really fortunate where we've had the lowest spreads in the industry because of our risk profile of building the exact same data center with very controlled costs and very controlled ongoing operational costs, only leasing to very high credit-worthy tenants. We've led the industry a few times in our ABS spreads over base rate just because we're so de-risked and because we're such a well-deserved company.

That led to late last year, we were able to publicly announce a deal with KKR where they're effectively, they have the option to buy our stabilized assets, pre-negotiated rates, to put them effectively on a yield co or a stable co. And that's a little unique to Compass because again, we have such predictability of cash flows, predictability of operating margins because our buildings are the same, because we build these long-term assets that are intended to last 100 plus years. There's very little long-term duration risk.

And because we lease to the same very high credit-worthy tenants, that relationship made sense. Now, once that happens and we have stabilized assets going into that vehicle with KKR, effectively like a forward equity sale, we then have a tremendous amount of cash. We are now completely self-funded and no longer need to go to the markets or I should say we go to the markets less or no longer need to do any form of equity raise and can rely less on our $6 billion credit facility because we have these stabilized assets that are throwing off so much cash as soon as they reach lease commencement and are put into that KKR vehicle.

[Nancy Lashine] (38:06 - 38:48)

My eyes are popping. Who is KKR offering that stabilized cashflow to? What types of investors and what kind of vehicles?

I have no idea. Okay, fair enough. It's just the numbers are so staggering and I did a podcast with Roy March about a year ago and he was just talking about the amount of capital being sucked into the data center industry and maybe coming out of core opportunities for office, retail, industrial, typical assets and you hear these numbers and it does feel astronomical relative to anything else going on in the real estate business.

[Brian Feldman] (38:49 - 40:02)

It really is and it's only inflating further as people now, they routinely throw around these gigawatt size data center campuses and you have to remember like a gigawatt of power is the equivalent of the entire, maybe it's a gigawatt and a half of the entire city of San Francisco and so these are just absolutely colossal, 10 billion plus dollar investments in a single data center campus. So to your comment earlier, like the numbers really are, it's staggering how large these have become but it's also its infrastructure and that goes back to our comment of like infrastructure or real estate. These are really long-term assets where the intent is not in 15 years, the data center is obsolete and you're gonna scrape it and rebuild it.

It's really, if you're doing it right, they're intended to be infinitely upgradable and like at Compass, that's how we build all of our gen sets, all of our power modules, all of our cooling units are all external. So in 15 years, if we completely changed form factors and we're now using quantum computing and need like cryogenic cooling chambers, no problem, we'll just unplug our vertib units and plug in the new units right in the same spot.

[Nancy Lashine] (40:03 - 40:42)

But don't you incur huge incremental costs for doing that? And so the economics, this obviously gets to the question of the takeout and that's the question on everybody's mind is, this is great, we understand why you'd want a long-term lease with these great credits and makes lots of sense but what happens at the end of 15 years or at the end of this takeout? What if the technology has changed so much that the capital cost to retrofit or to make this a current piece of technology just changes the economics substantially?

How do you get comfortable with that decision today?

[Brian Feldman] (40:44 - 43:50)

It's a great question, the residual value question, the obsolescence question. If you're designing your data centers correctly, it's infinitely upgradable because again, it's space, power and cooling. Maybe your cooling form factor changes.

Again, if you're designing it correctly, you can upgrade it. Maybe your power density changes, which we've seen over the industry tremendously over the last number of years, you can upgrade it. If your data center is designed well to really be modular and to be future-proof, again, it's space, power and cooling.

So how you bring those to bear in the data center environment can really change and it's okay. And at Compass, we pride ourselves on building 100-year assets that are infinitely upgradable where in 15 years, the form factor is totally different, that's okay. That's your question of how do you underwrite that?

Who pays for that? When we are leasing a data center to a customer, we are leasing them the shell, the pad, the shell and all of the mechanical, electrical and plumbing equipment. We then expect to make a return on that lease agreement over the 10 to 15-year life of that lease.

Thinking about year 16, if that lease is renewed, great. We may need to refresh or change some of that equipment. We may not need to.

And then we wanna sign another 15-year lease on that same unit. If in 15 years, the form factor has drastically changed where power densities are now wildly different, cooling topologies are wildly different, we'll work out something with our tenant they're not going to expect us to bear all of the capital for that because the data center that they leased from us, they're now asking for something drastically different. At the same time, with another 15-year lease agreement at a new negotiated renewal rate, you'd expect to receive the same earning higher return after you basically rebuild that.

And let me say one other thing on this topic because this is super interesting. People often really flag data centers as not creating a lot of jobs. And what I don't think people necessarily understand is if I'm building a 10-building data center campus, let's say I'm building one building or even two buildings a year for the next five or 10 years.

10 years go by, building one is now 10 years old. It likely is gonna be in need of a hardware refresh where the tenant's gonna take all the chips out, put all new chips in, and when they do that, they probably want different cooling. They probably want different power architecture.

So now we are then retrofitting that building, working with the tenant to make it work for the next 10-year cycle. And all those construction workers then go from the electricians, the plumbers, the pipe fitters, the concrete, the masonry workers, they all go from building 10 back to building one, and then the construction cycle just continues. So they're temporary jobs, but what people don't often realize is the level of permanence because these buildings do require a constant upgrade and retrofit.

And that's one baked into the economics and two based in Middle East terms.

[Nancy Lashine] (43:51 - 44:04)

I suppose whether these buildings are leased from a data center developer like yourselves or they own them themselves, they have to do that work in any case after some period of time.

[Brian Feldman] (44:04 - 44:41)

And this is part of what I used to work on back at the hyperscalers is how often are you doing that hardware refresh? And how different of a form factor is it? Is it every five years, you're just replacing the same types of racks in kind and so there's really no major changes needed, or are you doing a massive migration to something much higher efficiency where you wanna go from air cooled to liquid cooled, and it really changes the building design.

And again, that's where the differentiator of really good data center operator from a less sophisticated operator becomes apparent because we can very easily do that in our buildings. And it doesn't require us to scrape the site and rebuild it.

[Nancy Lashine] (44:42 - 45:11)

So let me ask the bubble question just head on. Because of course that's a question that's on everybody's mind is, is this the ultimate bubble? How durable is this demand from the hyperscalers for data centers?

And are we just feeding into something that will clearly, like regional malls, life sciences, studios that were developed about five, seven years ago, is this demand going to fizzle?

[Brian Feldman] (45:13 - 46:45)

Yeah, that is the multi trillion dollar question, right? I think if you think about it pragmatically, there has to be something, there has to be a disruptive event at some point in the future. This can't go on forever, would be naive to think can.

Sooner or later, there has to be some sort of shock that reorients the industry and that changes things. Right now, I don't think anyone has any line of sight to what that may be. Could be the private credit markets, how some of these leases are now off balance sheet.

And maybe not priced appropriately by the rating agencies or there's some accounting arbitrage happening. It could be that AI doesn't monetize as fast as people are thinking. And all of these investments, that the hyperscalers effectively stop making these investments because they're not getting the return on their invested capital that they expect.

It could be some combination thereof. I don't know. What I will say though is, if you pull back a bit, and two things can be true at the same time.

One, the industry is getting pretty frothy, where there is just a lot of just fantastical ideas of every day it's a new five gigawatt data center campus plan somewhere. And someone is putting nuclear reactors underground or data centers in space. It's really getting, I don't know how to say, like fantastical.

[Nancy Lashine] (46:46 - 46:47)

And so one thing- The world just died by, yeah.

[Brian Feldman] (46:47 - 48:06)

Yeah, exactly. And that grabs the headlines of people's attention and it is scary. On the other hand, this is the other thing that is true, is if you pull back, what's fundamentally driving the industry is it's a societal demand for compute.

And whether the form factor is AI or cloud or whatever else, we are all using our phone more and more. We are all using compute more and more. Society's demand for higher and higher, the data creation, the usage of that data and the compute needed to understand and use that data even further is still just increasing exponentially.

And so on that layer, we're still pretty early innings. You look at what some of the enterprise AI tools can do now and it's remarkable and it's still very early days. So I wouldn't be surprised, Nancy, if we have some kind of self-correcting mechanism in the next few years, what we saw in 2021 and 22, where post-COVID projections were way too high and the markets corrected, I wouldn't be surprised if we see something like that.

But I also think long-term, like the prospects of the industry, because it's just a reliance on compute, are up and to the right.

[Nancy Lashine] (48:07 - 48:24)

Do you think if we see some kind of correction, who knows where and how and why it comes, it impacts the assets that are fully leased to the hyperscalers or it really just impacts the economics of new prospective development?

[Brian Feldman] (48:26 - 50:44)

Really good question. I would guess it just more impacts new prospective development for a couple of reasons. Hyperscalers would have to start walking away from leases on existing assets for the valuations and the risk profile of those assets to change.

And even if that were to happen, there are pretty strict early termination clauses in those leases where you're effectively like buying out the remainder of the cash flows. So the likelihood of someone walking away from an existing asset is pretty high. The likelihood of someone leaving an existing asset is extremely low.

Also, like I can list on one hand, maybe it's two, but whatever, the number of times that a hyperscaler has walked away from a leased asset. It's incredibly expensive for them to do so. The opportunity cost is incredibly high and they're not good at it.

And the way the cloud is architected, it doesn't really warrant it, or AI workloads are warranted, it doesn't really work. So the notion that they would vacate existing buildings, not renew leases, it's always possible, but I think that risk is really low. I think what would be more likely is we get into an overbuild scenario where demand suddenly softens, now there's a lot of new development prospects that can't find a tenant.

Even that, I think it is somewhat challenging to imagine because the industry is so constrained by power, either grid power or onsite power generation, which is so complex, takes multiple years, is so expensive. You have this demand that's really untethered and then supply is really rooted, data center supply is really rooted in steel, concrete, engines, like real world physical items, and power and energy that is perpetually constrained. So for the past 10 years, we have been constantly playing catch up to this data center demand, and that only got much, much worse with AI.

And again, that demand is completely untethered, but here in the real world, when we actually go build this stuff, labor, we need people to build it, concrete to build it, steel to build it. And right now the real gating factor is power, and maybe that'll change in five or 10 years, but who knows?

[Nancy Lashine] (50:47 - 51:28)

Let me ask you a question a little out of left field. I don't know if you've spent any time thinking about this. After 9-11 and people started thinking about the risks we'd never thought about for obvious reasons, and the electricity grid became very much a topic of conversation.

You don't really hear people talking about that much these days, but as we obviously watch the incredible importance of the electricity grid on these hyperscale facilities, how do you think about the risk of just the security from foreign powers, from terrorism, from other factors on our energy grid?

[Brian Feldman] (51:30 - 54:03)

Yeah, I keep saying these are infrastructure, and really they're critical infrastructure. And we're taking this on April 8th, and look at what's happened in the last week where some data centers in the Middle East have been taken out. That just shows the importance of these assets and really how you have to think about them as critical infrastructure.

So going way back in our conversation, when hyperscalers are deciding to put their data center in Germany, in Israel, or in Saudi Arabia, this is something they're considering, the geopolitical risks. Of course, security is a massive consideration at the site level where locations of data centers typically aren't public. You sometimes have signage outside saying who's there.

Sometimes you don't, particularly surrounded by what's called K-rated or government fencing or even retaining walls to prevent any sort of access. Security is one of the biggest concerns at the physical level, but security in a broader sense is also extremely important in data center siting and whether you're going to country X or country Y or state X or state Y. So with that in mind, we're going to see like a heightened focus on security going forward just because of the geopolitical tensions.

Part of the reason why you're seeing so much investment in the U.S. for data centers is that region, the hyperscalers are all U.S.-based. It's the lowest geopolitical risk for them. It's the lowest data sovereignty risk for them. And of all the exogenous risks, that's lowest in the U.S. Now to your question about the energy grid, I actually think that's at the forefront of where we are in the industry because we're asking these grid providers for 500 megawatts, for a gigawatt, for...

No one has that type of capacity lying around. And so what the utility providers are doing is they're putting those costs back on the data center as they should so it doesn't impact rate payers and everybody's data. We want data centers to help bring power bills down, not the opposite.

And to do that, we need to pay our fair share, we need to pay for the capacity that we're contracting for, and we need to pay for the system upgrades that allow us to operate within those grids. Because of that, we're contending to work with grid providers all across the country. That is both stressing the grid and providing a lot of necessary focus for it to be upgraded.

It's the same thing globally.

[Nancy Lashine] (54:04 - 54:17)

You talked about data sovereignty, or you just mentioned the term, and I realize we haven't touched on that at all. It's a pretty hot issue in data center world. Just explain to our audience what that means and what the issues are.

[Brian Feldman] (54:19 - 56:25)

Yeah, so data sovereignty is effectively what is governing your data. So if you're in the United States, it's subject to the US laws. If you're in the state of California, it's subject to California laws.

As an example, Illinois has an interesting law in the book called the Biometric Information Privacy, BIPA. That is a layer on top of data governance for anything within the state of Illinois. And so data doesn't necessarily free flow across borders around the world in a way a lot of people think it does.

There are laws governing how you store the data, what you can do with the data, and who ultimately owns the data. Again, very different in the EU from here in the US, even very different in Canada from in the US. And so because you have this hodgepodge of different data standards for different countries all around the globe, that really influences where you make your data center investments.

If I am making a data center investment in Indonesia, as an example, it's very likely to serve domestic demand in Indonesia, because I'm gonna adhere to their laws, I'm gonna adhere to their policies around data flow, data locality, and data sovereignty. Some stuff, completely agnostic to that, like AI training, a lot of that is happening in the US because the companies are US-based and because that's allowed here. There may be an entirely different law passed regarding AI training in California versus in Texas in the next couple of years.

And we'll have to contemplate that and re-architect workloads because of it, and that will really shift and change where a data center investment goes in the same way you're already seeing it, like there's more regulatory risk of it happening in Canada than there is in the US. It's even higher in the EU. So I'm gonna make my data center investments in an area where I have the most control of my data and I can manipulate it in ways that I want to produce the products that I think are gonna be of my business.

And so it's just another element to when you're doing your data center siting that you have to think through.

[Nancy Lashine] (56:26 - 56:31)

Are data laws, data sovereignty laws, that different state to state?

[Brian Feldman] (56:32 - 57:00)

You know, data sovereignty might not be the right term, but there are different data governance laws. And so I brought up the one in Illinois. That's a great example that has a lot of unintended consequences that we're still sifting through.

There are other examples like that globally. I'm blanking on the acts that the EU passed recently, but same fact, where it limits how you can, what you can do with the data and that ultimately has a ripple effect on data center investment.

[Nancy Lashine] (57:01 - 57:20)

Right. We could keep going and going and going. This is such an interesting topic.

I wanna ask you, Brian, as you are talking about things that most of us are just holding on by our fingernails, trying to understand what you're doing every day, if we can at all, what do you think you'll be doing in five years?

[Brian Feldman] (57:22 - 58:27)

I hope more or less the same thing I am now, doing large scale data center development, working closely to provide solutions for our customers and providing worthwhile returns for our investors. Like I completely backed into this space by accident and I absolutely love it. So I hope in five years, I'm more or less doing the same thing.

Maybe the form factor changes, maybe the scale continues to change, but it is so fun to have an idea of, hey, I think this market is right for investment for X, Y, Z reasons. To articulate that thesis, to then get to go to that market, prove it, work on the ground to find a site that fits into your model, acquire the site, build it, create an operational data center and then exit hopefully with a sizable return. Like that to me is just super fun.

Five years from now, Nancy, honestly, like the power timelines on this stuff is so long. I'll probably be working on the same projects. But you know, I don't see a huge thing.

[Nancy Lashine] (58:28 - 58:31)

You mentioned form factor a few times. Is that a term of art?

[Brian Feldman] (58:32 - 59:38)

No, I just think, I think, again, data centers are space, power and cooling. Three years ago, liquid cooling was around. Today it's ubiquitous.

Maybe there will be something else in another three to five years. Another great example is power densities. So six or seven years ago, a 20 kVA rack, so a 20 kilowatt rated rack, power dense, that was intense.

Now with liquid cooling, you've got racks routinely at 150 kilowatts of power. So almost 10 X out of 200, 250. If you look at what Nvidia is broadcasting, the next generation will be, or you look at what Google is broadcasting, the next generation will be.

It's a single rack can consume a megawatt of power. And so it's again, just another upside. That means we'll need a different electrical topology, likely a different cooling topology that will take a slightly different form factor.

But I'm confident that our designs and our engineers will be able to handle it. We'll continue to advance in the industry.

[Nancy Lashine] (59:40 - 59:42)

Is there anything that keeps you up at night?

[Brian Feldman] (59:44 - 59:45)

My kids.

[Nancy Lashine] (59:45 - 59:46)

Yeah.

[Brian Feldman] (59:46 - 1:00:18)

I have young kids. In the industry, it's the AI bubble question, I think. Something has to give sooner or later.

Like this level of investment at this pace cannot continue forever. What's that gonna be? How will it materialize?

Is it a very soft landing or is it an abrupt stop? Who knows? Again, it would be naive to think that this runway will continue forever.

We know from history that it will not, but what does the end look like? It might be a pause. It might be a pause.

[Nancy Lashine] (1:00:18 - 1:00:20)

It might just be a yellow light.

[Brian Feldman] (1:00:21 - 1:00:40)

I mean, and if you look at the fundamentals, like our desire for compute, to me, that's what makes the most sense. But then it's still, how long is that pause? What happens?

Inversely though, when there are pauses like that, it's great opportunities to acquire things that others have walked away from where you can have a little bit more long-term focus. So we'll see.

[Nancy Lashine] (1:00:40 - 1:00:45)

A lot of people have been saying that about the office market the last two years. So there you go.

[Brian Feldman] (1:00:45 - 1:00:46)

Yeah, exactly, exactly.

[Nancy Lashine] (1:00:46 - 1:00:53)

Although I do feel a little silly comparing anything to the data center boom, given the quantum of capital that we've been talking about.

[Brian Feldman] (1:00:55 - 1:01:33)

I'm sure you saw the statistic from about six or so months where the level of data center investment surpassed the level of office investment in the US. That's pretty telling, but it's a little scary in its own way. But it's just, these buildings have gotten so large and so complicated, and a single data center campus is $10 billion.

You would need the largest, that's like all of Hudson Yards. It's like the largest office development. So we really are just, it's an interesting inflection point, but again, it's critical infrastructure.

It's like, this is what's powering this video call we're on and everything else we're gonna do for the rest of the day.

[Nancy Lashine] (1:01:34 - 1:01:51)

It's so true. I'd love to talk to you for the rest of the day, but I think we've definitely hit our limit, our podcast limit. So Brian, thank you so much for joining us.

And any advice you want to give to anybody listening who's interested in getting into the data center development business?

[Brian Feldman] (1:01:52 - 1:02:20)

Well, a fun question. I would say the opportunities are abundant in the industry at present. And understand the technical side, understand the electrical side, understand the cooling side, understand what actually happens in a data center, not just the investment or the market side.

And that will behoove you really well towards making good decisions at the macro level.

[Nancy Lashine] (1:02:22 - 1:02:29)

Great advice. Thank you so much for joining us. This has been a phenomenal conversation.

Really appreciate your time, Brian.

[Brian Feldman] (1:02:29 - 1:02:36)

Nancy, thank you so much for having me. Like I said, I've been a fan of the show for years and I'm just really humbled and excited to get to be a part of it. So thank you.

[Nancy Lashine] (1:02:37 - 1:03:09)

I hope you enjoyed this episode of Real Estate Capital. Before you go, I have a quick favor to ask. We put a lot of thought and effort into this show and making sure we bring you insights from real estate leaders that you don't normally find in the mainstream media.

So if you're enjoying the show, please remember to follow it on your favorite podcasting app so you never miss an episode. We'd also love for you to share it with others or give us a review on Apple Podcasts so others can find us. Thanks again for tuning in.

For more information about our firm, please visit our website at parkmadisonpartners.com.