บทถอดความฉบับเต็มของ Infrastructure for Intelligence: A Conversation with Blue Owl

but a meaningful fraction. There is a large amount of the economy that is being spent on data centers. The stat that I like to cite that I think is very interesting to me is last year was the first year where the capex on data centers was greater than the capex spent on new office builds. And I found that very interesting because. I didn't open up the newspaper every day last year and hear about a new office building, but I did hear about a new data center. But I think that helps put it into perspective. It's a meaningful part of the new growth of let the real asset economy, but it's still relatively small compared to other asset classes. speaker-0: I think people who are out there hear about AI, but there was this need before kind of AI came in, which is cloud computing. Right. And then you layer on top of that AI. So so how much more are the needs once you have those two pieces together, cloud computing and AI? speaker-1: Yeah. The way we try to kind of frame the size of the market is how much CapEx is there going to be required in the next five years. And you're right. And we like to ⁓ think about this as the the wave on wave of digital infrastructure demand. That's our kind of ⁓ way of describing it internally. Cloud computing has been around for ten years. AI started maybe two to three years ago, really seriously impacting the data center sector. Over the next five years, there is expected to be roughly seven trillion dollars to be spent on the development of digital infrastructure. Seven trillion with a T. Correct. That comes from a McKinsey study. And that seven trillion represents the land, the buildings, importantly, the computer service inside, the cabling, everything that goes into it. speaker-0: Seven trillion. Yeah. speaker-1: That's obviously a big number. And the way we think about it is we're not addressing all seven trillion. Despite maybe our grand ambition we're not going to do all of it ourselves. But I think it underscores the importance of why we are involved at all. When you look at the companies that are working in data centers, these are some of the largest, most well capitalized companies in the world triple A, AA rated credit spreads. But even they cannot do it themselves. speaker-0: Yeah. speaker-1: There needs to be a partnership between public companies, private capital that allows us to build out this capacity to further support the development of data centers, which helps underpin the overall growth of the economy. speaker-0: up the the financing need. So if I I think of kind of the the mag seven, right? ⁓ the these are just massive earners. I mean they're they they generate so much free cash flow. Did do they need external financing? speaker-1: Yeah, I I think it's really a question of financing, but also a question of expertise. So if you think about what some of these companies are at their core, you think about Google, it's a search engine, it's an advertising machine. You think about Meta, it's really an advertising machine now as well. Microsoft, Oracle Enterprise Software. Amazon maybe stands out a little bit as, you know, it it's a company that is very, very good at sending packages around the world and then also great at storing data and sending data across the world. speaker-0: A W R. speaker-1: So when we think about that, when you talk about the expertise of these companies, none of them are natural experts in building and owning long life infrastructural real estate assets. That's not their cool business. I think if you were a shareholder of these companies, you might say Microsoft or Amazon or any of these companies, your business is to develop great software and to generate recurring revenues from that software, not necessarily Invest into large CapEx items. So from an allocation of capital, if these companies had a dollar to spend, they might want to spend a dollar investing in RD for software or buying back their shares rather than investing in a relatively conservative yield profile of data center ownership. And then you talk about the expertise side. The difficulty ⁓ That data center development poses can be quite material. In order to create a data center campus where one didn't exist beforehand, it may take years of planning, of development, years of construction before you get to a functioning operating data center. And that requires a relatively specialized set of skills. We have within our portfolio companies over a thousand people whose job it is to wake up every morning and think about data centers. And that's not the core competency necessarily of one of these software companies, nor I think should it be. And so those two things together really help you understand why number one, it's helpful for them to have a partner just to create it. But number two, the synergies of working together with an expert in data center development can be beneficial for both sides. speaker-0: So so maybe that gets into to how these are are structured. Are are these are they the tenants? Are they the are they the the owner? Like how how do these how do these work? speaker-1: Yeah, so a typical structure is actually very simple. ⁓ the data center operator ⁓ is the landlord. So ⁓ it's a company that we all own or property that we all own. We'll buy the land. We will enter into a long-term lease with a hyperscaler. Those long-term leases can be between 15 and 20 years in length, so really a material amount of time. We'll build this data center, and then at that point, we are the landlord. They are the tenant, like any other commercial building or infrastructure asset, and will receive a predetermined cash flow, like rent, a monthly rent check, effectively, every month for the next fifteen years and then hopefully renewed there over time. So it sounds like it might be complicated, but it really is very similar to other traditional real estate or infrastructure asset classes. speaker-0: Think of AI landscape now. ⁓ you're hearing about all these

deals, and people probably write about them, kind of the circular finance where you know Nvidia is is investing here and there, turning around and and buying chips. Is is that sounds pretty simple. Is is there any wrinkles to any any of this? Is there SPVs that ⁓ Meta or Google are setting up? Are you are you actually facing the parent, or is there some sort of structure in between? speaker-1: Yeah. It's slightly a complicated answer because there are also different things at play. To take the first part of your question, let's separate out maybe the AI ecosystem from the infrastructure ecosystem. The AI ecosystem, I think it's fair to say, is something that chipmakers like NVIDIA are trying to encourage. Ultimately, NVIDIA wants many more customers to buy their chips into perpetuity. Sure. And so speaker-0: It's bit of like vendor financing structure. speaker-1: A lot of these chip buyers are interlinked because they're also creating what their businesses will do. ⁓ OpenAI, anthropic companies like this didn't exist five years ago. So their business plans are still evolving. And it's in NVIDIA's best interest to help foster these businesses to be long-term customers. That's one side of the business. When we talk about our side of the business, We are really involved in the long life infrastructure, the long-life owning of the land, the building. And so while that AI ecosystem is a helpful tailwind and drives demand, we are still ultimately facing these investment grade-rated entities. ⁓ Our business is underpinned, again, as we said at the beginning, really by the fundamental credits of the off-takers. So we're not facing a company that didn't exist 10 years ago, that unclear exactly how its business plan will play out. We're facing investment grade rated entities. The second question you mentioned about SPVs and you mentioned Meta and Google. We did do a joint venture recently with Meta, a deal that I worked on called Project Hyperion, where there was a joint venture set up between Meta and Blue Owl owned entities to develop a data center in Louisiana. While that sounds maybe slightly complicated at the beginning, the fundamental underwriting was still the same. We own a property, that property is leased to Meta. And so that linkage, the property being leased to Meta and Meta paying rent, or in all the other cases we have we own a property and it's leased to a hyperscaler, and that hyperscaler pays our rent, that remains the same. And that's the fundamental business that we're entering into. We're not betting. Does AI grow at thirty or fifty percent, or does this company commercialize quicker than that company? We're betting does an investment great company pay its rent? speaker-0: What we're talking about really i i is private credit. This is more niche private credit, but it is private credit. Why wouldn't ⁓ some of the these big hyperscalers just tap the public markets for these? speaker-1: Yeah, and I think the answer is they do, and they will, and they they will continue to do so. ⁓ obviously their interests are in maintaining their high credit ratings because that helps go into their cost of capital, which helps support their equity. So it's all an interrelated story. But the reality is all sources of financing are going to be needed. We talked about that seven trillion dollar number. In order to hit that target, I think companies and private capital are going to do the following things. Kyperscalers are going to use their free cash flow that they generate to pay for things. They're going to issue debt to pay for things. They may even issue equity to pay for things. And they will enter into partnerships with private capital like us, where we will develop and they will lease over a long term period. These four tools all need to be used to address the total opportunity. And that's where we're seeing where we can play in as part of a wider ecosystem. speaker-0: And and I guess this ⁓ maybe talks to to the the why private credit is just the flexibility in that that you can structure things, you can own things, you can you can do a lot of different things where maybe the public markets can't. speaker-1: Yeah, the public markets are fantastic in the fact that they are wide, they are deep, they can move quickly. But to do something in the public sphere, often it is here is the deal, and the public sphere, the job is to say, okay, well, here is the price. Yep. The difference in private capital is we can have a long term partnership with one of these clients, one of these customers, and they will say, We want to do a deal, we're not sure exactly how it should look. And we can work together and structure something where we can provide not only here is the price, but here is how the structure works. And that additional flexibility is critical when you're investing in something that hasn't really happened before in this way. Right? Ten years ago, these structures didn't exist, the hyperscalers business were not as large, data centers were not growing as quickly. And now they are, and so the flexibility in conjunction with public markets activities is the way to make sure that you have all the tools in your toolbox to solve this capital need. speaker-0: So we talked about a little bit how or where you know private credit fits in. certainly probably are maybe the most well known in this space and and on the tech side, ⁓ but others do this too. So so why blue owl? Why not ⁓ somebody else? speaker-1: Yeah. It it's a good question. One I ask ⁓ ask myself every morning, you know, how do I make sure that I'm going to win the day? ⁓ the way I like to think about it is when you look at what Blue Owl can provide and what our peers can provide, I think about it as two axes. One the first one is availability of capital. Blue Owl is a large publicly traded asset manager with hundreds of billions of AUM. But there are peers of us that also fit that bill. But that's a necessary requirement. You need to

make sure that you have the capital to stand behind some of these large constructions because they may be billions or tens of billions in size. The other access is sector expertise. If we think about the differentiation that Blue Owl has at our team, at the Blue Owl level, we have about 75 people who are specifically dedicated to digital infrastructure. We wake up, we think about digital infrastructure. Go to bed, we dream about digital infrastructure. How do these two things overlap? When we are negotiating with a counterparty who wants the expertise of someone who is invested in digital infrastructure, if they go to an asset manager, that asset manager may not be an expert themselves. They may own a portfolio company that invests in digital infrastructure, which their portfolio company handles, and every quarter, a board meeting happens and they have a decision. But it's not the job of the investment manager in that case. For Blue Owl, it is the job of the investment manager. Our job is to know everything about data centers, how they are developed, how they are financed, how they are monetized. The second layer is the expertise. We have on our team people who are experts in development, experts in power, experts in leasing type of scalars, experts in operating them. And so this combination puts us in a bit of an unusual space. We are able to be experts in digital infrastructure and have the capital to speak to it. And that's how we think, or at least I hope, that we can differentiate ourselves to our peers. And so far it seems to be a good combination. We've had a long track record in developing these data centers. And I think it really ties back to that combination of we have the expertise. speaker-0: Well that's a good commercial. speaker-1: The capital and therefore the ability to really access this market in scale. speaker-0: You you mentioned one deal example. Can you give maybe give another one? You s you mentioned Hyperion. Is there is there ⁓ is beignet or like other I love all the different code names you have for speaker-1: Right, that's right. so Hy Hyperion or or Beignet, that this was the deal in Louisiana, very proud of it. ⁓ it was a deal that was relatively publicly in the press. So while this project Hyperion is obviously a large one, there are many others and you know, they're like they're all my favorite children, right? it's because they all share the same fundamental. We are a landlord where we're leasing out on a long term contract to invest from great counterparties. And so whether that counterparty is a triple A rated counterparty, double A rated counterparty, a investment grade rated counterparty, they all have similar characteristics and I'm very happy to do any of them and I hope that we'll do many more over the next few years. speaker-0: Take Hyperion as an example. I mean, it and I think people understand the concept of of a data center. You know, maybe if you if you're in the corridor in V Northern Virginia where they're they're everywhere, or you know, we have some in in Ohio and Columb in the Columbus area, but probably listeners who haven't seen a data center, they they may think it looks like some sort of industrial, you know, logistics center. Like maybe give it give an example of like how big this is, like the scale of these projects. They're they're massive. speaker-1: Everything that you said is correct. One data center by itself really does look like a large in light industrial logistics center. It's meant to be a big building without walls out of the way. One of the benefits, however, is it doesn't have trucks moving in and out every hour. It's just a computer that, you know, is on and helping provide the internet. a large data center building right now is around five hundred thousand square feet. And so when you talk about campuses of these data centers, we have some investments where the amount of data centers on any one campus is maybe ten or fifteen of these five hundred thousand square feet buildings. So it starts becoming, you know, really meaningfully large in in various areas. speaker-0: And how many acres is this? Is this speaker-1: So the largest transaction that we've done, ⁓ the total campus spans about five thousand acres. Five thousand acres. So I think that's roughly the size of Manhattan. So it it is quite large. speaker-0: Okay, so I I I paused on seven trillion. I'm gonna pause here. The size of Manhattan I mean that's that's incredible. speaker-1: Right. We so we did a site visit and it took about three hours to drive around. Not three hours to walk around, but three hours to to drive around the site. speaker-0: A few blocks in Manhattan's probably three hours of traffic. speaker-1: Exactly, exactly. but I think it talks to the important part of it, which is data centers really underpin everything that we do now that's online. And so if you just think about, you know, how you interact with technology, if you think about today and what it was like five years ago, what it was like ten years ago, I think many people would say that the use of technology has just been exponentially increasing. That technology has to sit somewhere. Your ability to access data on your phone or your computer has to sit somewhere. That somewhere is a data center. And so that helps to explain why these data centers are big and getting bigger and why they're located in in various regions. I don't think that on an average day many people necessarily go past a data center. ⁓ but it is sometimes interesting, you know, if you really want to understand where the heart of the internet lies. It lies in a big windowless box. speaker-0: I think that's that's helpful. And I want to talk about some of the the project risks and and and you think about Manhattan ⁓ is power and just think of powering Manhattan, right? And so like how how do you think about getting the power to these massive campuses? speaker-1: It's a challenge. Yeah. ⁓ and maybe ⁓ that's actually why

⁓ a firm like Blue Wild Digital Infrastructure exists. If it were easy, you would Yeah Yeah, you would need a specialist to do it. ultimately what this power challenge is is how do we modernize the grid? How do we put enough generation? How do we put enough transmission into areas where it's needed, and then how do we transport that information from speaker-0: Everybody would do it. speaker-1: The data center to the people that are ultimately using it. Part of it absolutely is building new generation, building new ⁓ power plants that are able to ⁓ to set up and make sure the computers are always on. But in addition, a lot of the time these data centers need to be connected to the grid because the grid is ultimately the most reliable source of power that we can have. So the answer to your question is it's difficult, it's a multi-year process. But importantly, that power risk is passed on to the tenant. So our job, our responsibility for our investments is making sure that there can be power available and making sure that the tenant is comfortable that there will be power available. But many, many times the tenant will just sign the power agreement directly with the power provider and will be there as a good developer, actually creating that opportunity. But we won't be taking that power risk ourselves. It's not our job is not let's buy a piece of land and hope that there is power, or let's try to take a speculative bet on whether power is available. That's the job of a power developer. And while I think that that can be a successful and profitable endeav endeavor, that's not where our area of expertise lies. speaker-0: I've heard ⁓ from others the the challenges of power. speaker-1: And I think you may have seen there was something that the White House recently published called the Rate Payers Protection Pledge, which was signed by all of Microsoft, Amazon, Meta, Google and Oracle, unsurprisingly, similar tenants that we target, plus a couple of the other large AI names. And what they committed to do was to build or pay for new infrastructure for the grid, whether they use it or not, also invest in local jobs, also partner with communities. And what that is able, or what they are saying, and what we have seen and observed is that it's actually able to pull down energy prices in a region when done properly, because the new investment that the hyperscalers are putting in essentially takes some of the burden off the rate players in that region. So it's not intuitive when you say it when you think about it initially. Okay, more demand for power means higher prices. But in reality, what we've been able to see where data center development has been done right, communities have actually been able to benefit from speaker-0: Gotcha. I did want to ask ⁓ about this other risk, and it's it's sort of obsolescence. I mean, ⁓ we've been studying AI since probably the fifties, but like in a commercial form the last two or three years. Can these data centers, these giant campuses, can can they be obsolete in a few years? What what does that do? Is that a risk? speaker-1: Yeah. It the way we think about it is actually relatively easy. So when you invest in a data center, as a landlord, you invest in the land and the building. Yeah. You do not invest in the chips or the computers inside. That's the responsibility of the tenants. Why is it the responsibility of the tenants? Because these chips truly do change and evolve very quickly. Yeah. I think we hear from NVIDIA every three to five years there's a new chip coming out. Maybe it can become more powerful, it can do more things. There are competitor chips. All these items are changing. So, how do we think about this technological innovation? The way I like to look at it is a data center is effectively a very sophisticated power socket. If you're in your own home, you may buy a new TV, you may buy a new computer, a new iPad, whatever it is. And every evolution of this is the chip getting more powerful, the technology getting more powerful. But nevertheless, you're always plugging that into the power socket. The data center is that sophisticated power socket. We have data centers that we own that I'll give you an example. One of the data centers we own is the original AOL You've Got Mail data center. Nice. It's wantful. It is not an AOL you've got mail data center right now. It is a hyperscale data center. It's been repurposed, it's been reused, but it is a data center that is a very efficient way to convert a unit of power into a unit of data. And so regardless of how we think about these chips, every single chip wants the best way to convert a unit of power into a unit of data. And that's what the data center provides. speaker-0: And and these and these are guaranteed, right? So so if if there are more CapEx requirements, they may either reneg renegotiate with you and you might be open to it or or or not. ⁓ but they can't just stop paying you. speaker-1: No, no, of course. The payment is a lease contract, right? Every month they have to pay that amount. Often the leases are triple net. So things like CapEx and taxes and all these things are passed through the tenants. And then if there is a big CapEx refresh cycle, what has happened in the recent few years is that amount often gets effectively rentalized into the price of the new contract to allow the tenant to continue using the space efficiently. So we are sitting on what we think is a relatively boring stream of cash flows. Every month you get the check. ⁓ and this new technology side of it, it just keeps on reinforcing the amount of demand there is for that ultimate power socket that we are providing. speaker-0: Going forward is there I throw this out here. Elon Musk said he's gonna put these out in space. Yes. So what does that do? speaker-1: Space is the final frontier. No, I, you know, I think it's an interesting question because ⁓ with a problem like the bottleneck in land and power, everyone is

trying to work on a solution. And Elon Musk owns a satellite company, and perhaps data centers in satellites in space will be a solution. There are still some material challenges thinking about this just from a technical side. I'll give you a very easy one. Occasionally you have to go into the data center and change out the servers. Yeah. It is much easier to fly to Northern Virginia than it is to get to space. To space, yes. ⁓ so there are some things that that need to be looked at. I think at one point people were looking at putting data centers underwater. Yeah. Because water under under the water, it's very cold. That hasn't necessarily materialized. Not all the data centers are underwater now. ⁓ you know, who knows what it will be like in the next decades to come. ⁓ but I think that the Easiest way to solve the problem is still let's build a data center. ⁓ let's try to work as a developer with the communities to get the land, to get the power. ⁓ and then, you know, maybe the the simplest one is the hyperscalers themselves are paying for this, right? They're paying rent based on the cost of the data center. I think if I went to them and said you need to pay for the data center and the cost of the rocket up to space, they would say, you know, let's try to find more data centers on the earth. speaker-0: You know, there is construction risk, project risk. I would think you then have the launch risk on top of that makes it a bit more ⁓ speaker-1: Exactly. Nothing is easier in space, I think is probably a the way to say. speaker-0: Cooling, the cooling might be cheaper, but ⁓ the the whole execution might not be. speaker-1: So not to go into the the rabbit hole of space, but ⁓ space is very cold, you're totally correct. Space is also a vacuum. Yeah. And because it's a vacuum, it's actually quite hard to cool something. So the way cooling works is you basically run something cold over something warm. And the particles interact and you take them off. But because space is empty and truly empty, vacuum empty, it's actually not as trivial as you would think to cool something down. Anyway, that goes into like I'm not a physicist, it goes into another realm, but Yeah, there are definitely some challenges to think about before space is a viable option. speaker-0: We hit on a bunch of th these for this. I kind of went through the the checklist. Did I miss any? speaker-1: It's kind of a question of, you know, what am I worried about? Yeah. ⁓ I think that when you look at the business, it can be very simple. It is there are years of work of development, which is truly just hard work and sweat. It is making sure that you get the right zoning and permits and entitlements. And that is complicated and takes a long time, but we're not building before anything is done. Yeah. So it's not a monetary risk, that's an effort risk. We may spend years. Trying to developing a site and get nowhere. Yeah. But once the site is constructed, and we've constructed maybe a hundred data centers now, plus or minus, ⁓ so we know how to do it. Yeah. And this really ticks those kinds of boxes. So once the data center is up and running, once we put money into it, I c I personally feel pretty comfortable. speaker-0: You can sleep at night. That's right. All right. Well, kind of wrapping this up a little bit, a a couple final questions. ⁓ one, listeners, viewers, how how do they get more understanding of of this build out? I mean, it's fascinating, right? And you're and you and you compare it to to railroads. And I've heard the the US highway system where I mean this this is the Manhattan Project. These are like ⁓ big dollars. Like how do people get get smarter about this? speaker-1: It's a good question. I think for many years data centers have been perhaps a little bit under the radar. ⁓ people have just and maybe that's actually a good part of them, which was people like that the internet just works. People like that you switch on their phone and it works. You press the app, it works. But it works because of this interlocking system of cell towels and fiber and is at heart data centers. ⁓ now there is much more narrative explaining what a data center is. The fundamental thing I would say is Data centers are far less complicated and less scary than they appear. It is truly a light industrial warehouse where instead of selling packages, they're sending data. Yeah. And from an owner perspective, we are viewing it as this is a simple building and we get paid rent from a great counterparty. It is as simple as that. And you couldn't talk about the technology and the interlocking spheres of digital infrastructure, but that's what you should always orientate yourself to. This is a building with a long-term contract. speaker-0: It it's not that complicated. All right. Last question. Sure. What is one fun fact about Alexa that nobody knows? speaker-1: Not that complicated. ⁓ that's a tough question. I don't know whether I'm a very fun person. speaker-0: That maybe that's fun. Fun to know that you're not fun. That's speaker-1: you know, I really enjoy reading, which I know sounds a little silly because you know, hopefully everyone enjoys reading, but I particularly enjoy reading science fiction and fantasy. So maybe a fun fact about me, I'm a nerd, I very much enjoy reading fantasy books. ⁓ speaker-0: What is what is your favorite science fiction book? speaker-1: My favorite science fiction book, funnily enough, is actually called Hyperion. Yeah. ⁓ it's a book by an author I think his name is Dan Simmons, and I very much enjoy it. I think that's probably my speaker-0: I yes, that and that's where the project gets its name. I'm pretty sure that's where it's got but yes, I s there's a lot of the sci fi names that come up and if if you read these books or or you're a fan, you're like, huh, I wonder if that's that's from it. That's very good. Well, very good. Well Lexi, thanks so much for coming on the FEG and Site Bridge and we really enjoyed this and this was very beneficial. I learned something, I enjoyed it. So speaker-1: I'm not entirely sure whether it's a hundred percent. Thank you. Thank you so much for having me here. Really appreciate it. speaker-0: If you are interested in more information on FEG, check out our website at www.feg.com. And don't forget to subscribe to our communications so you don't miss the next episode. Please keep in mind that this information is intended to be general education that needs to be framed within the unique risk and return objectives of each client. Therefore, nobody should consider these to be FEG recommendations. This podcast was prepared by FEG. Neither the information nor any opinion expressed in this podcast constitutes an offer or an invitation to make an offer to buy or sell any securities. The views and opinions expressed by guest speakers are solely their own and do not necessarily represent the views or opinions of their firm or of FEG.