theCUBE + NYSE Wired: AI Factories - Data Centers of the Future | Horace Zona, DigitalBridge

Clips
More from theCUBE + NYSE Wired: AI Factories - Data Centers of the Future

Horace Zona

Managing Director

DigitalBridge

play_circle_outline AI as a new form of compute requiring distinct infrastructure

play_circle_outline End-to-end digital infrastructure: towers, fiber, data centers, subsea, HPC

play_circle_outline Inference compute identified as primary economic opportunity

play_circle_outline Creative Capital Structures for Chips and Data Centers: CoreWeave's Domain Driven GPU Financings

play_circle_outline Edge and Metro AI Factories: Networking Evolution with InfiniBand, Mellanox, KV Cache, Disaggregated Serving

Info

Horace Zona, DigitalBridge

Horace Zona

Managing Director DigitalBridge

search

John Furrier

>> Palo Alto Studio Connection, Silicon Valley and Wall Street. I'm John Furrier, co-host of theCUBE, here with Dave Vellante, my co-host. Hello, I'm John Furrier, host of theCUBE, here in our Palo Alto— New York City, NYSE studio. We have a Palo Alto studio. Connecting Silicon Valley to Wall Street. And it really is about the deep tech interfacing with Wall Street. That's the NYSE Wired program and Open Community. It's our AI Factory series. We got Horace Zona here, managing director, DigitalBridge. Seen many waves of innovation. Horace Zona, great to see you here in theCUBE and NYSE Wired program. We were talking before we came on camera about historical perspective and can't wait to get into it.

Horace Zona

>> I am too. Thanks so much for having me today. I'm excited to sit down and chat.

John Furrier

>> So you're involved in a lot of the key infrastructure deals, but also you have a historical perspective. You've seen the internet wave, you've seen the telecom wave, you've seen the SaaS cloud wave. Now you got the AI wave. there's been a lot of talk around AI taking over the world and killing people and all the Anthropic nonsense. This bubble talk. Are we in a bubble? Not in a bubble. All that aside, we are in the most massive demand curve I've ever seen in my career, where you have all these forces and the confluence of these forces. As someone who's been seeing the waves, what is your take on this? Because a lot of us are seeing this. Is that right? What— am I crazy? What are you seeing?

Horace Zona

>> Well, we're seeing tremendous opportunity to make fundamental, fundamentally driven, very thoughtful investments across the entire capital structure. So as a firm, we're unlike most GPs in that our entire focus is investing whether it be public equity, private equity, private credit. All we do is invest, invest in digital infrastructure. So it's that level of focus where our portfolio companies really give us insights about where the opportunity is. And so to draw historical perspective, I started in '93 as a banker within the media and communications space, really focusing on digital infrastructure. So that would be, broadband, wireless, fiber. And at times you've seen very significant pockets of growth, but not with a really detailed business model. And that ultimately can end up in a bad place. And I'd say then to really roll that forward to today, we see a very healthy ecosystem with genuine demand drivers and the supply is there to fulfill it. And so not to necessarily just ape what Jensen says, but, AI is a completely new form of compute. It cannot necessarily leverage off of prior infrastructures. And so as a value proposition, we see a real opportunity there for end users and the developers to bring to marketplace that intelligence that people are willing to pay for. And that's obviously what this is all about.

John Furrier

>> When you say new forms of compute, I totally agree with it because I wrote a post a couple of weeks ago, said it's math that needs to be fed more math. And if you look at the matrix multiplication, if you look at the density of the NVIDIA architecture, it's mathematics that's driving all that activity. Now, the systems around it are getting tweaked. You got KV cache, you got a lot of memory, and they're getting bigger and there's demand for it. But there's also It's not over. You got the edge, but there's also an infrastructure on the cloud side. We have the internet's built, it's steady state. There's no debate there. The hyperscalers, AWS, they pumped the last generation. You got Uber who built first generation app from columnar store to people, places, and things. So a lot of stuff's kind of been done. So it's like there's a shoulders of giants kind of mentality that's a lot different from the, hey, let's kill the local loop. Telecom deal, bring in broadband, lay down new fiber, web replaces this, so there was a displacement in the other generations. It doesn't feel like a displacement infrastructure because some stuff will go away and change and refactor, but what is your take on that? Because if you look at things like the internet really in the '90s, that was telecom. You had PBXs became digital. Eyeballs on the web being the valuation. Mary Meeker had a slide, the online population, it was like a chart she tracks at Morgan Stanley back in the day. It's like, hey, more people are coming on the information superhighway, the web. That was just a curve, that was just a ramp. So you have now people connected with SaaS and phones. How does that change your view on digital infrastructure? Because in a way, a lot of it's already out there.

Horace Zona

>> Sure, well, we look at the infrastructure from end to end, right? Anything that is moving a bit is really part of what we think of moving or storing a bit. Is what we think of as digital infrastructure. So that's towers, that's residential broadband, that's enterprise, enterprise fiber and broadband, satellite, subsea cable, of course, data centers, and then high-performance compute. I don't like to use the Neo Cloud phrase.

John Furrier

>> AI Cloud, they call it. Sure. That's what CoreWeave called themselves, high-performance compute.

Horace Zona

>> And so What's interesting is that, this is largely driven by the product and the need that you're trying to fill for the customer. So cloud computing is very simple, right? You're— if I'm working on a PowerPoint in New York and you're in Palo Alto and we want to work on it together, we need really very low latency for that, for us to coexist so we're not messing up with each other and frustrating each other. AI is very different, right? I am now querying whether it's Claude or ChatGPT or whomever with a task, and that task can be easy or complicated so that latency is not necessarily as important today for AI compute. So if I give it a very difficult task, if I've uploaded 10 documents, I want it to perform several different analyses. I want different presentation schemes that may take 30 seconds. It could take a minute.

John Furrier

>> And so you don't care. It's good.

Horace Zona

>> I'm not driven by that.

John Furrier

>> Yeah.

Horace Zona

>> And we haven't even gotten into what real agentic compute is all about. So I'm unleashing 2 or 3 bots to 24/7 discover, and analyze, find different opportunities. So I'd say that these are complementary. Networks, but they're different. And so you're really not getting into that Innovator's Dilemma where one system is dislocating another. Not, not in our perspective.

John Furrier

>> And the latency is a great point because if you're driving a car that's autonomous or you're not driving, you have a chance to— or your physical robot running in manufacturing, doing heavy lifting, there's a safety issue. They need real time. So you want to have a policy. Again, that's, I think, where you're getting that compute's different.

Horace Zona

>> Right, exactly. So physical AI, right? The robotics, that is a revolution that is starting very, very early. In fact, we haven't made an investment there yet, but we're looking forward to that space and keeping a really close eye.

John Furrier

>> Yeah, I'm really fascinated by your networking background because why I'm going there is because Scott McNealy had an expression, network is the computer, back in the Sun Microsystems days. You're smiling, you remember. I actually interviewed him in 2013, and I said, what do you think about cloud? He goes, it's— he said, I just should have called it the cloud, right? Because I feel like it is the computer. But if you look at NVIDIA, the whole premise of their entire architecture is networking-based. Mellanox was the beginning of it. No one wanted to buy that company. There was only like 3 buyers. NVIDIA picked it up. InfiniBand network that made the whole system happen. But now the KV cache is so critical and is growing. You got disaggregated serving. Hello, inference. So you're starting to see that new compute architecture emerge. How do you look at that? Because the next question would be, okay, if AI factory is a big fat node, Big Tech's data center, gigawatt, big metro node, there's going to be edge factories. A telecom tower has a building, they got power and they got networking and also they got RF on the other end. So you start to see, okay, if you inject intelligence That's it. That could be an investment thesis. I'd love to get your reaction on that because, I've been saying, look, if you inject intelligence, that might even change how licensed spectrum might operate. And that's just a haymaker. But that's kind of the thinking that people are starting to come to grips with. What's your thoughts on it? Because that's like, that's an arc. It's not maybe next year it might come. That'll power things.

Horace Zona

>> Inference is really the big tentpole.

John Furrier

>> Right.

Horace Zona

>> That's where there is tremendous economic opportunity. That's where at an enterprise level, we can see a tremendous uptake in the compute for those businesses. It's largely going to be productivity and efficiency enhancing. If you're not doing it, your competitors are doing it. They're going to eat you alive. And so we will see the implementation of that. Coming very fast here over the next several years. And so what we're really investing in are the data centers and the high-performance compute businesses, the chip factories that will make that happen, the picks and the shovels. And what's interesting is that post-COVID, when we saw such an uptake in cloud computing where a 20-megawatt or a 50-megawatt data center was a very significant investment, only 3 years ago. Now, 2 years ago, we see hundreds of megawatt data centers. And of course, now we see gigawatt data factories. Now, that is not all cloud computing, but what we've seen is that demand get driven and the supply has to come to where the power is. So now we're getting into this whole NIMBY thing where it is a significant concern as an industry. We're very focused on doing the right thing for our communities where we live. And so what we want to see, though, is the compute happen and bring it back. So whether it's through a tower, through a fiber to where the end user needs it.

John Furrier

>> On DigitalBridge, what's your parameters or I guess the line where you don't invest because you brought energy up? Energy is the bounding function. Does that take you off the reservation, so to speak, or is that in line with what you're thinking? Because we're seeing geothermal use cases coming out in the West, obviously a terawatt could come out of the super hot rocks out there. That's like the journey to the center of the earth for energy. We don't know yet. We don't know. But there's also energy futures. Is that off scope? Is that in line?

Horace Zona

>> Our co-founder, Marc Ganzi, has been speaking about this now for quite some time, to the extent that even with SoftBank making an offer to acquire DigitalBridge, we went out and we've made a subsequent agreement to buy ArcLight Capital, an energy infrastructure investment business. So yes, we do see and value that area and we look forward to partnering with those people in the near future. So yes. Energy is at the core of many of our discussions and there's an opportunity

John Furrier

>> to—

>> of course, what's the mindset? I'm learning a lot from you in this session. So I like to take a few liberties, if you don't mind. When Jensen came into New York a couple of weeks ago with KKR, Goldman, all the players, Brookfield, I think was in there, a few others.

Horace Zona

>> Goldman.

John Furrier

>> Goldman was in there. So, okay. What I saw was Jensen basically trying to get the monkey off his back on this whole circular financing thing, which he was taking a lot of heat for. Bloomberg had a story. They made up a story or they tried to hang a story out there. I don't think NVIDIA is nefarious in the circular finance. I think they're just making the market. That's my opinion. But I think what I heard him saying is, look, I want to bring pros in. We need financial capital, financial players, partners to finance the physical plants. I'll use that word loosely. And then it's okay. I get that. That's a good thing. And then I say, okay, that's the physical plant. So how do you see the financial world coming behind it? Because a lot of these old school data centers have old school contracts, 10, 20 year leases. The contracts are coming in around 3 to 4 years. Back end years are not loaded at all. That may be 2 years visibility. So there's a lot of risk. And yeah, there are finance people looking at this because The numbers aren't rounding errors. They're significant Bs and maybe a T in the future.

Horace Zona

>> That's true. So here's maybe a more simplified way to think about it. Let's talk about the infrastructure itself. So data centers historically have had a useful life in the 10 to 15 year timeframe. And so banks and institutional investors and private credit have all invested whether you have an investment-grade offtake or non-investment-grade offtake, and brought capital to the table to actually finance that growth, to go from that 20-megawatt data center to that 1- and 2-gigawatt data center complex. So finance has been very constructive there. At DigitalBridge, our credit fund worked with Magnetar to stand up CoreWeave's first two GPU financing vehicles, which at the time CoreWeave was not a very significant company.

John Furrier

>> And the numbers were big too.

Horace Zona

>> And right.

John Furrier

>> A lot of people were throwing shade on that deal.

Horace Zona

>> Well, it was $2.3 billion in the first deal, $7.3 in the second. We had a— they had a fantastic— they had negotiated a fantastic contract with Microsoft. And so what we brought to the table was some creativity and some belief that this infrastructure, these chips had significant value because of the offtake, right? Microsoft had a 5-year compute deal with CoreWeave and we helped finance that.

John Furrier

>> And so that was bankable in your mind?

Horace Zona

>> Completely, absolutely bankable. Now, the banks wouldn't have done it and institutional investors weren't ready to do it, but private credit was ready to do it. So I would say now take a step back from those infrastructure deals. 5-year compute deals for chips, 15-year leases with data centers. there's a duration imbalance there, right? So what we have found is that there are opportunities to finance both stacks at the same time. So finance the chips and the data center all at once or finance them independently. And so really, every situation is somewhat different. But we can work with those funds.

John Furrier

>> So the key was creativity and not being hung up on not needing a rating, right?

Horace Zona

>> Using our own fundamental insight and knowledge of the space and trying to find the right resolution for the company. CoreWeave, which is at the time looking for sources of capital and hadn't found many, and then working for our LPs as well.

John Furrier

>> So this is not an obvious thing. So you have to have domain experience. This is where, do you see this as a big part of the market? Because I think this is going to be more of the trend. The domain risk management is going to come from domain expertise, not just numbers. So you made a bet. You made a bet on CoreWeave is what you did.

Horace Zona

>> And a great contract with Microsoft. Yeah. And NVIDIA as well, right? We, part of our fundamental analysis was the belief that that ecosystem, that group of companies wanted OpenAI to succeed. Right. Because that's where ultimately the compute—

John Furrier

>> and that was a pretty obvious bet too. They were on a trajectory. Correct. We'll be at the CoreWeave conference in 2 weeks too. We'll get more data from them.

Horace Zona

>> Of course. But back to your point, when Jensen came in and announced this $500 billion MOU, I'm not sure at the end of the day what problem they're trying to solve. We do believe chips are financeable. Now, does the institutional market feel that way? Does the bank market feel that way? There's evidence that they're coming to grips with that, but not on the scale order of how data centers have been financed. And let's remember that if you have— if it takes $100 to build a data center, it probably costs $250 for the chips that go into the data center. So there's a scale order there that needs to get financed, and that's why you break it out. Correct. So I think that what Jensen and NVIDIA are trying to accomplish is to provide a valuation floor of some sort where more financing sources can look at a 5-year average life or a 6-year average life and find comfort in that like we have.

John Furrier

>> Yeah. And of course, they have the backstop options. They have all kinds of trajectory trend data, lots of tools in the bag. Yeah. And it's— I think that's market making. Do you think that that meeting was really more about confidence too? Or am I just conspiracy theorizing to think it was just shaking off that noise around backstopping and circular financing because they got the cash? Why wouldn't you make the market if I'm motivated to build a physical plant that people use my chips in? And by the way, they're having durability numbers too on them.

Horace Zona

>> I, from what I can see and for transactions that we have evaluated, I don't think that there is a fundamental circularity to these financings. I believe that NVIDIA is building a business and executing a plan where they're getting compute out into the marketplace. But I can tell you that they are hands-off after a certain point. And they want the marketplace to thrive, develop and thrive without necessarily their micromanaging. And I feel confident saying that.

John Furrier

>> So it's a nice invisible hand strategy for them. All right. So what's the focus for you guys now? Because, you get a lot of activity. I think we're at the early innings of this build out. The edge is going to be massive. I think it might look a lot like the early days of the internet buildout when you saw UUNETs of the world and Exoduses coming out. metro POPs were popular. I was talking to NVIDIA like, if you're going to have an AI factory in Texas, why not have one in New York? Like, why can't I just put a Metro factory in New York and leverage all the— well, access to power, a lot of dormant power in a lot of these buildings. there's an argument for regional topology.

Horace Zona

>> We've seen, some— I've seen a couple of different business models as of late that are really trying to just do that. So take advantage of opportunities where you have physical infrastructure with available power and connectivity and putting those together to present a valuable business plan for inference compute. And I think that there are a lot of merits to that. But I would also say that this is not an easy business.

John Furrier

>> No, it's not. Where's the data? Where's the market data?

Horace Zona

>> Right. So bringing these three things together at a time and place with the confidence of your off-taker to sign that 5-year contract or that 15-year contract, that's why we have a number of portfolio companies that have been quite successful.

John Furrier

>> Have you seen deals where the offtake contracts were significantly large and the company has yet to stand up and turn on clusters? And is that a symptom of a flawed execution or a lack of market? Because you got— I could have $1 billion in orders, put it in escrow, but I got to actually develop, deploy, de-risk the deployment?

Horace Zona

>> I think that there's a lot of speculation in the marketplace about the announcements of deals. And now let's see how they get executed. I'm sure, as in any other hypergrowth phase, we will see some projects that don't move along on the advertised timeline because of a number of challenges. That are honestly too many to list. But I don't think that that's a fundamental flaw in the business proposition.

John Furrier

>> Yeah, right.

Horace Zona

>> That gets back to the expertise, the focus of those operations and their ability to deliver.

John Furrier

>> All right. Final question. What are you excited about now from a business plan standpoint that you're looking at? What kind of category is it? The data centers, the chips? What kind of transactions are you excited to do deals on and will invest and lean into?

Horace Zona

>> So we remain excited about the entire ecosystem. We see predominance of activity in non-investment-grade data center development. The high-performance compute businesses are largely all non-investment-grade. And then your inference cloud companies are, of course, all nascent there. 2, 3 years old, they're not investment grade as well. That's an awesome place for us.

John Furrier

>> What does non-investment grade mean to the average person?

Horace Zona

>> Yeah. So that means, you don't have a rating from Moody's or S&P that says, you're BBB and above. So it doesn't—

John Furrier

>> it requires due diligence.

Horace Zona

>> Its size and age of operations, it doesn't necessarily mean a company is any less worthy. And that's why we find them to be terrific places.

John Furrier

>> That's where opportunity lies.

Horace Zona

>> That's where our opportunity lies because you can arbitrage the non-rating value extraction. Well, we like to think of it as helping these markets develop.

John Furrier

>> Yeah.

Horace Zona

>> And just to harken back to the initial CoreWeave GPU financings. Just 3 years later, many thought that they were more near investment grade than or Yeah, squarely investment grade, than non-investment grade.

John Furrier

>> Having NVIDIA involved kind of makes it a quasi-investment grade. That's a AAA in my mind. So it's great to have you on. I could go on forever learning a lot because I think the bounding function is not just energy, it's finance. And I think what you're doing and the kind of creativity, bringing the domain expertise and looking at this thing holistically is where I think a lot of the growth will come from. So, well, it's a real pleasure to sit down with you. Yeah, I appreciate it. I'm John Furrier, host of theCUBE. AI infrastructure is going to continue to boom, is only going to get bigger and bigger and global. The edge and the energy and the financial functions are going to be very important to make sure things can hedge the risk, understand risk, but also the upside potential. AI infrastructure is certainly not a bubble. The demand's there, the systems are in place. It's a matter of the entrepreneurs and the capital finding those constraints and making it happen. We're doing our part here on theCUBE. Thanks for watching.

Horace Zona

>> theCUBE.

Horace Zona, DigitalBridge

search

John Furrier

>> Palo Alto Studio Connection, Silicon Valley and Wall Street. I'm John Furrier, co-host of theCUBE, here with Dave Vellante, my co-host. Hello, I'm John Furrier, host of theCUBE, here in our Palo Alto— New York City, NYSE studio. We have a Palo Alto studio. Connecting Silicon Valley to Wall Street. And it really is about the deep tech interfacing with Wall Street. That's the NYSE Wired program and Open Community. It's our AI Factory series. We got Horace Zona here, managing director, DigitalBridge. Seen many waves of innovation. Horace Zona, great to see you here in theCUBE and NYSE Wired program. We were talking before we came on camera about historical perspective and can't wait to get into it.

Horace Zona

>> I am too. Thanks so much for having me today. I'm excited to sit down and chat.

John Furrier

>> So you're involved in a lot of the key infrastructure deals, but also you have a historical perspective. You've seen the internet wave, you've seen the telecom wave, you've seen the SaaS cloud wave. Now you got the AI wave. there's been a lot of talk around AI taking over the world and killing people and all the Anthropic nonsense. This bubble talk. Are we in a bubble? Not in a bubble. All that aside, we are in the most massive demand curve I've ever seen in my career, where you have all these forces and the confluence of these forces. As someone who's been seeing the waves, what is your take on this? Because a lot of us are seeing this. Is that right? What— am I crazy? What are you seeing?

Horace Zona

>> Well, we're seeing tremendous opportunity to make fundamental, fundamentally driven, very thoughtful investments across the entire capital structure. So as a firm, we're unlike most GPs in that our entire focus is investing whether it be public equity, private equity, private credit. All we do is invest, invest in digital infrastructure. So it's that level of focus where our portfolio companies really give us insights about where the opportunity is. And so to draw historical perspective, I started in '93 as a banker within the media and communications space, really focusing on digital infrastructure. So that would be, broadband, wireless, fiber. And at times you've seen very significant pockets of growth, but not with a really detailed business model. And that ultimately can end up in a bad place. And I'd say then to really roll that forward to today, we see a very healthy ecosystem with genuine demand drivers and the supply is there to fulfill it. And so not to necessarily just ape what Jensen says, but, AI is a completely new form of compute. It cannot necessarily leverage off of prior infrastructures. And so as a value proposition, we see a real opportunity there for end users and the developers to bring to marketplace that intelligence that people are willing to pay for. And that's obviously what this is all about.

John Furrier

>> When you say new forms of compute, I totally agree with it because I wrote a post a couple of weeks ago, said it's math that needs to be fed more math. And if you look at the matrix multiplication, if you look at the density of the NVIDIA architecture, it's mathematics that's driving all that activity. Now, the systems around it are getting tweaked. You got KV cache, you got a lot of memory, and they're getting bigger and there's demand for it. But there's also It's not over. You got the edge, but there's also an infrastructure on the cloud side. We have the internet's built, it's steady state. There's no debate there. The hyperscalers, AWS, they pumped the last generation. You got Uber who built first generation app from columnar store to people, places, and things. So a lot of stuff's kind of been done. So it's like there's a shoulders of giants kind of mentality that's a lot different from the, hey, let's kill the local loop. Telecom deal, bring in broadband, lay down new fiber, web replaces this, so there was a displacement in the other generations. It doesn't feel like a displacement infrastructure because some stuff will go away and change and refactor, but what is your take on that? Because if you look at things like the internet really in the '90s, that was telecom. You had PBXs became digital. Eyeballs on the web being the valuation. Mary Meeker had a slide, the online population, it was like a chart she tracks at Morgan Stanley back in the day. It's like, hey, more people are coming on the information superhighway, the web. That was just a curve, that was just a ramp. So you have now people connected with SaaS and phones. How does that change your view on digital infrastructure? Because in a way, a lot of it's already out there.

Horace Zona

>> Sure, well, we look at the infrastructure from end to end, right? Anything that is moving a bit is really part of what we think of moving or storing a bit. Is what we think of as digital infrastructure. So that's towers, that's residential broadband, that's enterprise, enterprise fiber and broadband, satellite, subsea cable, of course, data centers, and then high-performance compute. I don't like to use the Neo Cloud phrase.

John Furrier

>> AI Cloud, they call it. Sure. That's what CoreWeave called themselves, high-performance compute.

Horace Zona

>> And so What's interesting is that, this is largely driven by the product and the need that you're trying to fill for the customer. So cloud computing is very simple, right? You're— if I'm working on a PowerPoint in New York and you're in Palo Alto and we want to work on it together, we need really very low latency for that, for us to coexist so we're not messing up with each other and frustrating each other. AI is very different, right? I am now querying whether it's Claude or ChatGPT or whomever with a task, and that task can be easy or complicated so that latency is not necessarily as important today for AI compute. So if I give it a very difficult task, if I've uploaded 10 documents, I want it to perform several different analyses. I want different presentation schemes that may take 30 seconds. It could take a minute.

John Furrier

>> And so you don't care. It's good.

Horace Zona

>> I'm not driven by that.

John Furrier

>> Yeah.

Horace Zona

>> And we haven't even gotten into what real agentic compute is all about. So I'm unleashing 2 or 3 bots to 24/7 discover, and analyze, find different opportunities. So I'd say that these are complementary. Networks, but they're different. And so you're really not getting into that Innovator's Dilemma where one system is dislocating another. Not, not in our perspective.

John Furrier

>> And the latency is a great point because if you're driving a car that's autonomous or you're not driving, you have a chance to— or your physical robot running in manufacturing, doing heavy lifting, there's a safety issue. They need real time. So you want to have a policy. Again, that's, I think, where you're getting that compute's different.

Horace Zona

>> Right, exactly. So physical AI, right? The robotics, that is a revolution that is starting very, very early. In fact, we haven't made an investment there yet, but we're looking forward to that space and keeping a really close eye.

John Furrier

>> Yeah, I'm really fascinated by your networking background because why I'm going there is because Scott McNealy had an expression, network is the computer, back in the Sun Microsystems days. You're smiling, you remember. I actually interviewed him in 2013, and I said, what do you think about cloud? He goes, it's— he said, I just should have called it the cloud, right? Because I feel like it is the computer. But if you look at NVIDIA, the whole premise of their entire architecture is networking-based. Mellanox was the beginning of it. No one wanted to buy that company. There was only like 3 buyers. NVIDIA picked it up. InfiniBand network that made the whole system happen. But now the KV cache is so critical and is growing. You got disaggregated serving. Hello, inference. So you're starting to see that new compute architecture emerge. How do you look at that? Because the next question would be, okay, if AI factory is a big fat node, Big Tech's data center, gigawatt, big metro node, there's going to be edge factories. A telecom tower has a building, they got power and they got networking and also they got RF on the other end. So you start to see, okay, if you inject intelligence That's it. That could be an investment thesis. I'd love to get your reaction on that because, I've been saying, look, if you inject intelligence, that might even change how licensed spectrum might operate. And that's just a haymaker. But that's kind of the thinking that people are starting to come to grips with. What's your thoughts on it? Because that's like, that's an arc. It's not maybe next year it might come. That'll power things.

Horace Zona

>> Inference is really the big tentpole.

John Furrier

>> Right.

Horace Zona

>> That's where there is tremendous economic opportunity. That's where at an enterprise level, we can see a tremendous uptake in the compute for those businesses. It's largely going to be productivity and efficiency enhancing. If you're not doing it, your competitors are doing it. They're going to eat you alive. And so we will see the implementation of that. Coming very fast here over the next several years. And so what we're really investing in are the data centers and the high-performance compute businesses, the chip factories that will make that happen, the picks and the shovels. And what's interesting is that post-COVID, when we saw such an uptake in cloud computing where a 20-megawatt or a 50-megawatt data center was a very significant investment, only 3 years ago. Now, 2 years ago, we see hundreds of megawatt data centers. And of course, now we see gigawatt data factories. Now, that is not all cloud computing, but what we've seen is that demand get driven and the supply has to come to where the power is. So now we're getting into this whole NIMBY thing where it is a significant concern as an industry. We're very focused on doing the right thing for our communities where we live. And so what we want to see, though, is the compute happen and bring it back. So whether it's through a tower, through a fiber to where the end user needs it.

John Furrier

>> On DigitalBridge, what's your parameters or I guess the line where you don't invest because you brought energy up? Energy is the bounding function. Does that take you off the reservation, so to speak, or is that in line with what you're thinking? Because we're seeing geothermal use cases coming out in the West, obviously a terawatt could come out of the super hot rocks out there. That's like the journey to the center of the earth for energy. We don't know yet. We don't know. But there's also energy futures. Is that off scope? Is that in line?

Horace Zona

>> Our co-founder, Marc Ganzi, has been speaking about this now for quite some time, to the extent that even with SoftBank making an offer to acquire DigitalBridge, we went out and we've made a subsequent agreement to buy ArcLight Capital, an energy infrastructure investment business. So yes, we do see and value that area and we look forward to partnering with those people in the near future. So yes. Energy is at the core of many of our discussions and there's an opportunity

John Furrier

>> to—

>> of course, what's the mindset? I'm learning a lot from you in this session. So I like to take a few liberties, if you don't mind. When Jensen came into New York a couple of weeks ago with KKR, Goldman, all the players, Brookfield, I think was in there, a few others.

Horace Zona

>> Goldman.

John Furrier

>> Goldman was in there. So, okay. What I saw was Jensen basically trying to get the monkey off his back on this whole circular financing thing, which he was taking a lot of heat for. Bloomberg had a story. They made up a story or they tried to hang a story out there. I don't think NVIDIA is nefarious in the circular finance. I think they're just making the market. That's my opinion. But I think what I heard him saying is, look, I want to bring pros in. We need financial capital, financial players, partners to finance the physical plants. I'll use that word loosely. And then it's okay. I get that. That's a good thing. And then I say, okay, that's the physical plant. So how do you see the financial world coming behind it? Because a lot of these old school data centers have old school contracts, 10, 20 year leases. The contracts are coming in around 3 to 4 years. Back end years are not loaded at all. That may be 2 years visibility. So there's a lot of risk. And yeah, there are finance people looking at this because The numbers aren't rounding errors. They're significant Bs and maybe a T in the future.

Horace Zona

>> That's true. So here's maybe a more simplified way to think about it. Let's talk about the infrastructure itself. So data centers historically have had a useful life in the 10 to 15 year timeframe. And so banks and institutional investors and private credit have all invested whether you have an investment-grade offtake or non-investment-grade offtake, and brought capital to the table to actually finance that growth, to go from that 20-megawatt data center to that 1- and 2-gigawatt data center complex. So finance has been very constructive there. At DigitalBridge, our credit fund worked with Magnetar to stand up CoreWeave's first two GPU financing vehicles, which at the time CoreWeave was not a very significant company.

John Furrier

>> And the numbers were big too.

Horace Zona

>> And right.

John Furrier

>> A lot of people were throwing shade on that deal.

Horace Zona

>> Well, it was $2.3 billion in the first deal, $7.3 in the second. We had a— they had a fantastic— they had negotiated a fantastic contract with Microsoft. And so what we brought to the table was some creativity and some belief that this infrastructure, these chips had significant value because of the offtake, right? Microsoft had a 5-year compute deal with CoreWeave and we helped finance that.

John Furrier

>> And so that was bankable in your mind?

Horace Zona

>> Completely, absolutely bankable. Now, the banks wouldn't have done it and institutional investors weren't ready to do it, but private credit was ready to do it. So I would say now take a step back from those infrastructure deals. 5-year compute deals for chips, 15-year leases with data centers. there's a duration imbalance there, right? So what we have found is that there are opportunities to finance both stacks at the same time. So finance the chips and the data center all at once or finance them independently. And so really, every situation is somewhat different. But we can work with those funds.

John Furrier

>> So the key was creativity and not being hung up on not needing a rating, right?

Horace Zona

>> Using our own fundamental insight and knowledge of the space and trying to find the right resolution for the company. CoreWeave, which is at the time looking for sources of capital and hadn't found many, and then working for our LPs as well.

John Furrier

>> So this is not an obvious thing. So you have to have domain experience. This is where, do you see this as a big part of the market? Because I think this is going to be more of the trend. The domain risk management is going to come from domain expertise, not just numbers. So you made a bet. You made a bet on CoreWeave is what you did.

Horace Zona

>> And a great contract with Microsoft. Yeah. And NVIDIA as well, right? We, part of our fundamental analysis was the belief that that ecosystem, that group of companies wanted OpenAI to succeed. Right. Because that's where ultimately the compute—

John Furrier

>> and that was a pretty obvious bet too. They were on a trajectory. Correct. We'll be at the CoreWeave conference in 2 weeks too. We'll get more data from them.

Horace Zona

>> Of course. But back to your point, when Jensen came in and announced this $500 billion MOU, I'm not sure at the end of the day what problem they're trying to solve. We do believe chips are financeable. Now, does the institutional market feel that way? Does the bank market feel that way? There's evidence that they're coming to grips with that, but not on the scale order of how data centers have been financed. And let's remember that if you have— if it takes $100 to build a data center, it probably costs $250 for the chips that go into the data center. So there's a scale order there that needs to get financed, and that's why you break it out. Correct. So I think that what Jensen and NVIDIA are trying to accomplish is to provide a valuation floor of some sort where more financing sources can look at a 5-year average life or a 6-year average life and find comfort in that like we have.

John Furrier

>> Yeah. And of course, they have the backstop options. They have all kinds of trajectory trend data, lots of tools in the bag. Yeah. And it's— I think that's market making. Do you think that that meeting was really more about confidence too? Or am I just conspiracy theorizing to think it was just shaking off that noise around backstopping and circular financing because they got the cash? Why wouldn't you make the market if I'm motivated to build a physical plant that people use my chips in? And by the way, they're having durability numbers too on them.

Horace Zona

>> I, from what I can see and for transactions that we have evaluated, I don't think that there is a fundamental circularity to these financings. I believe that NVIDIA is building a business and executing a plan where they're getting compute out into the marketplace. But I can tell you that they are hands-off after a certain point. And they want the marketplace to thrive, develop and thrive without necessarily their micromanaging. And I feel confident saying that.

John Furrier

>> So it's a nice invisible hand strategy for them. All right. So what's the focus for you guys now? Because, you get a lot of activity. I think we're at the early innings of this build out. The edge is going to be massive. I think it might look a lot like the early days of the internet buildout when you saw UUNETs of the world and Exoduses coming out. metro POPs were popular. I was talking to NVIDIA like, if you're going to have an AI factory in Texas, why not have one in New York? Like, why can't I just put a Metro factory in New York and leverage all the— well, access to power, a lot of dormant power in a lot of these buildings. there's an argument for regional topology.

Horace Zona

>> We've seen, some— I've seen a couple of different business models as of late that are really trying to just do that. So take advantage of opportunities where you have physical infrastructure with available power and connectivity and putting those together to present a valuable business plan for inference compute. And I think that there are a lot of merits to that. But I would also say that this is not an easy business.

John Furrier

>> No, it's not. Where's the data? Where's the market data?

Horace Zona

>> Right. So bringing these three things together at a time and place with the confidence of your off-taker to sign that 5-year contract or that 15-year contract, that's why we have a number of portfolio companies that have been quite successful.

John Furrier

>> Have you seen deals where the offtake contracts were significantly large and the company has yet to stand up and turn on clusters? And is that a symptom of a flawed execution or a lack of market? Because you got— I could have $1 billion in orders, put it in escrow, but I got to actually develop, deploy, de-risk the deployment?

Horace Zona

>> I think that there's a lot of speculation in the marketplace about the announcements of deals. And now let's see how they get executed. I'm sure, as in any other hypergrowth phase, we will see some projects that don't move along on the advertised timeline because of a number of challenges. That are honestly too many to list. But I don't think that that's a fundamental flaw in the business proposition.

John Furrier

>> Yeah, right.

Horace Zona

>> That gets back to the expertise, the focus of those operations and their ability to deliver.

John Furrier

>> All right. Final question. What are you excited about now from a business plan standpoint that you're looking at? What kind of category is it? The data centers, the chips? What kind of transactions are you excited to do deals on and will invest and lean into?

Horace Zona

>> So we remain excited about the entire ecosystem. We see predominance of activity in non-investment-grade data center development. The high-performance compute businesses are largely all non-investment-grade. And then your inference cloud companies are, of course, all nascent there. 2, 3 years old, they're not investment grade as well. That's an awesome place for us.

John Furrier

>> What does non-investment grade mean to the average person?

Horace Zona

>> Yeah. So that means, you don't have a rating from Moody's or S&P that says, you're BBB and above. So it doesn't—

John Furrier

>> it requires due diligence.

Horace Zona

>> Its size and age of operations, it doesn't necessarily mean a company is any less worthy. And that's why we find them to be terrific places.

John Furrier

>> That's where opportunity lies.

Horace Zona

>> That's where our opportunity lies because you can arbitrage the non-rating value extraction. Well, we like to think of it as helping these markets develop.

John Furrier

>> Yeah.

Horace Zona

>> And just to harken back to the initial CoreWeave GPU financings. Just 3 years later, many thought that they were more near investment grade than or Yeah, squarely investment grade, than non-investment grade.

John Furrier

>> Having NVIDIA involved kind of makes it a quasi-investment grade. That's a AAA in my mind. So it's great to have you on. I could go on forever learning a lot because I think the bounding function is not just energy, it's finance. And I think what you're doing and the kind of creativity, bringing the domain expertise and looking at this thing holistically is where I think a lot of the growth will come from. So, well, it's a real pleasure to sit down with you. Yeah, I appreciate it. I'm John Furrier, host of theCUBE. AI infrastructure is going to continue to boom, is only going to get bigger and bigger and global. The edge and the energy and the financial functions are going to be very important to make sure things can hedge the risk, understand risk, but also the upside potential. AI infrastructure is certainly not a bubble. The demand's there, the systems are in place. It's a matter of the entrepreneurs and the capital finding those constraints and making it happen. We're doing our part here on theCUBE. Thanks for watching.

Horace Zona

>> theCUBE.