Well, good afternoon, everybody, and welcome to our Fourth Quarter Outlook Webinar. Before we get started, I do want to highlight the Resources tab, where you can find more of our insights specifically related to some of what we're going to talk about today, and a survey to share your feedback on the day's event. Additionally, please ask any questions you might have with the Q&A tab, and we'll address them at the end of our prepared remarks, provided the time allows. My name is Jake Symoniak, and I'm a Senior Global Equity Strategist here at Manning & Napier. Today, I'm joined by my colleague, James Slentz. James is the Managing Director of our Technology group on the Core Equities team. James, before we get going, I was hoping you could tell us a little bit about yourself, maybe your experience here at the firm, and what it is you do now. Yeah, thanks for the intro, Jake. I've been at Manning & Napier for over 15 years, and I've been helping to cover technology that whole time. As Managing Director of the tech group, my primary role is leading a team of three people who are basically scouring the tech coverage universe for companies that are best positioned to benefit from secular trends across the tech landscape, and to think about how those trends might be creating winners and losers both in and outside our coverage area. One of the ways that we identify those winners and losers is by breaking down the value chain into its constituent parts and thinking about all the different bottom-up activities that enable some of the top-down technology paradigm or architectural shifts that seem to happen at these sort of predictable intervals. And so the presentations today will hopefully give people a little insight into how we go about doing that. And suffice to say, the landscape has gotten a lot more interesting over the past 12-18 months. Great. Thanks, James. I appreciate that. So before we do get started, I'm going to do what I typically do at the outset of these and kind of touch base on what it is that we hope to talk about today and maybe provide a bit of a brief outline. So to start, we are going to talk briefly just about the state of the U.S. economy, kind of what it is we're seeing, and maybe get into, I guess, the decision tree for the Federal Reserve here and maybe what that could mean moving forward. I think what's going to be a little bit different than what we've done in the past is we're going to spend less time talking about the economy and the broad market today and really get to where we want to take the conversation, which is having a much deeper conversation about the AI side of things, and the goal is to really outline our approach to thinking about, I guess, what I would describe as the value chain or the AI stack and discuss not only the various layers, but talk through how we think about them from an investment perspective through the lens of our strategy and valuation disciplines here at Manning & Napier. I think, importantly, we're not just going to talk about, I think, the first-level beneficiaries, but we're also going to talk about some of the opportunities that we believe exist and have been created in other parts of the market that may not be benefiting as clearly from AI today. From there, we'll wrap up the prepared remarks, and I do think we're going to have some time to get into questions and answers today. From there, let's get right into it. And I think the conversation about the U.S. economy, in a lot of ways, is similar to what we've talked about, what we've been seeing for a while now, where we have a resilient economy, and it's really been supported by consumption and specifically consumption on the higher end of the income spectrum, and then also strength from fixed investment, and specifically non-residential fixed investment, which, again, we're going to get into on the AI side of things. So James, if I could, from your perspective, what are you seeing in terms of domestic demand from your coverage areas here in the U.S.? Yeah, so it's sort of bifurcated, right? I think just from the perspective of sort of the consumer-focused tech companies, whether it's e-commerce, advertising, media, I think the commentary from management teams in those sectors are suggesting the consumer is in pretty decent shape here. I can't think of anyone that's really commented that they're seeing signs of significant stress. I think the enterprise is a very different story. Aside from this group, this sort of narrow group of companies that are embarking on this massive AI investment cycle, we're really seeing pretty tepid growth in IT budgets because things are changing so quickly. Companies are saying, "Look, things are changing really quickly. Let me evaluate all this new technology in front of me before I firm up my roadmap and make some spending decisions," and so I think we're seeing elongated sales cycles and very limited net new budget growth in those areas, so again, even within enterprise, it's quite bifurcated. I think that's interesting, too, because even in kind of a sector of the market that you cover where we've seen particular strength over the last, I guess, several years at this point, it reflects a lot of what we're seeing in the broader economy, right? I mean, anybody who's been on these for any period of time now has heard us talk about how we've almost been in these rolling recessionary periods in different, I guess, recessionary pockets in different parts of the economy where headline strength really hides some of the weakness we've seen in industrial parts of the economy, maybe on the small business side, housing-related parts of the economy. I think when we think about where we are today, that bifurcation, that specific weakness, specifically in rate-sensitive parts of the market where maybe you're more exposed to bank lending, mortgages, it helps to explain why, despite the fact that headline growth has remained resilient, we are to a point now where, for the first time in nine months, we're back to the Fed cutting interest rates. Yeah, I know we've had a lot of conversations on that recently. So question for you, I was just curious, what sorts of trade-offs do you think the Fed is facing when they're thinking about cutting into what has really been quite a resilient economy so far? Yeah, it's a good question. And I think it's probably the most important non-AI-related macro question that's out there. We're in a period where inflationary pressures are certainly rising, right? You look at goods inflation, and that's pushing higher. You look at leading indicators for services inflation, those are pushing higher. And it's happening in an environment where inflation is already running kind of well above the Fed's 2% target. You take a look at headline growth, and consumption remains pretty strong. Non-residential fixed investment remains incredibly robust. Again, that's something we're going to get to. But really coming out of Jackson Hole, the tone from the Fed shifted. We've certainly pivoted to the second of the two mandates, which is labor market as opposed to just inflation. You look there, you are seeing some softness. And again, you look at those rate-sensitive parts of the economy. You look at small businesses. You look at housing. And I don't think there's any doubt that you sort of need cuts. You need rates on the long end to move lower so that you can price lending at a lower rate. And you can actually see a pickup in activity there. I think what's concerning at this point is, I mean, you look at the long end of the curve now, and look at the 10-year. You're at a higher rate now than you were when the Fed cut. And I think it's fair to wonder, if you're cutting into this environment where growth has remained resilient and inflationary pressures are rising, you could end up, especially with the fiscal stimulus that's coming down the pike over the next couple of quarters, you could end up with the long end of the yield curve actually moving higher and be in a position where you're in a worse starting point for some of those parts of the economy than you are right now, and certainly where you were before the Fed cut. So I know for us, that's going to be one of the key things that we're keeping an eye on. I do want to move on from this, though, right? I don't want to spend a ton of time on the econ. I know we have a lot of ground to cover, and I want to get into the AI component of this. We talked at length, talked a little bit about how one of the key pillars of strength we've seen has been non-residential fixed investment. There's no question that that's been driven by just the massive spend that we've seen kind of across the AI value chain. Now, I do want to dig into this, and I genuinely mean this when I say it. I don't think we could have a better person with us today to talk about some of these dynamics, talk about what we're seeing, talk about how we're thinking about it from an investment perspective. But before we get into the specifics, James, I'm going to turn it over to you to maybe set the stage and just talk a bit about what we've seen in terms of, I guess, performance in the equity market kind of leading into this to really frame the discussion. Yeah, sure. So I mean, there's obviously been a tremendous amount of enthusiasm, right, for pretty much anything directly or indirectly exposed to the AI theme. I think we've said the word bifurcation several times already. It has driven a lot of this sort of dichotomy, not even just within tech, but across sectors because AI touches a lot of different areas. And that dichotomy has been between this group of perceived AI winners and perceived AI losers. So when I was thinking about how to answer this question, I pulled a few different statistics. So let me give you a few of those. So the first one is the tech momentum factor recently hit levels not seen since 2002. So a little bit of a warning sign there. Here's another one. The last time the 100 biggest stocks in the S&P outperformed the broader market to such an extreme degree was 1999, and then you can look at a variety of other sort of leading indicators or measures of risk or valuation that indicate to us that the market has moved pretty far out on the risk curve. If you look at zero-day to expiration option activity, which is basically just gambling on whether a stock is going to go up or down, that activity is at record levels. Retail flows are at record levels, so I think it's a pretty unique market in a lot of different ways when you consider valuations, concentration, retail participation, and a lot of other metrics you might look at as sort of evidence for the market's risk appetite. Yeah, I don't think there's any question that there has been exuberance both in terms of the physical investment, and that's certainly been reflected in equity market returns for AI-linked stocks and really anything that's been perceived to be linked to AI as well. So we've talked a little bit about the equity market, and I think that's going to come up kind of recurrently through the course of the conversation that we're going to have. But as it pertains to the actual value chain itself, right, the things that are being built in the real world, we've broken the value chain down into four broad categories. We're going to go into pretty great depth on each of them. But if you could maybe lay those out before we do, I think that'd be helpful. Yeah. So hopefully, people can stick with us because there is a lot to get to. So at the very top of the value chain for any industry are the companies that have a relationship with the end user, so either a consumer or somebody working in a company, an enterprise buyer. So these are the companies that are monetizing end user activity. That's the very top of the value chain. So for AI, you can think of applications like ChatGPT or Microsoft's Copilot as examples of companies at the top layer, and all the value flows downward from those companies to the supply chain below. So one layer down from the application layer, you have the AI models that some of these products are built on. And there are really only five big Western companies that are still investing in the frontier there. Then one layer down from those, you have the data center operators that are actually building the infrastructure that the models run on. And then a layer below them, you have the components and infrastructure of the data center itself. So that includes not just the chips, but also the buildings, the power, the cooling necessary to get all this stuff working. And then you could even go one layer deeper and talk about the semiconductor capital equipment suppliers, the chip manufacturing companies. And so those are sort of the very, very base of the pyramid or the foundation on which a lot of this other stuff rides. And that foundation, I think, is where we want to start today. So when I think about this, this is what actually powers the data center. This is kind of the base upon which everything else is built out. Can you talk to us a bit about what this layer of the pyramid entails? Sure. So at its simplest level, this is all the stuff that the hyperscalers, Amazon, Microsoft, Oracle, Google are spending their CapEx dollars on. So it's everything from land, power, cooling, chips, cabling, rack space, and the electricity to light those data centers up. And just as sort of an interesting tidbit, I thought I'd comment on just the size and scope of today's data centers because I don't know if anyone has seen or driven by any of Amazon or Walmart's big retail fulfillment centers. They're these massive buildings, right? Well, today's data centers, the ones that are powering AI, in some cases, they're multiples the size of those. So Microsoft just brought online the biggest data center in the world. It's a 315-acre site with 1.2 million sq ft. So the scale of these things is just absolutely massive. So when I think about the data center and I think about the spending that goes into building one of these, one of the first questions I have is how that actually breaks down, right? What percent of the spending goes to maybe the physical infrastructure, the building itself, the power equipment versus what it is we're putting inside of these? Sure. Yeah, so it's roughly a 40/60 split, 40% being what you might call infrastructure or long-lived assets, so that's the building, the HVAC equipment, the power supplies. The other 60% are the components that are inside the data center, which tend to be shorter-lived assets, and the biggest chunk of that are the chips that AI models run on. Okay. And I guess maybe taking another step deeper then, from our perspective here at Manning & Napier, how do we approach investing in this layer of the value chain? Yeah, it's a good question. So NVIDIA has sort of been the poster child here. I do think they remain sort of the most direct way to play broad-based adoption of AI. And to be fair, it's a phenomenal business. They've got something like 80% market share of the chips that are used for AI. They benefit from a massive installed base of developers. There's a huge software moat around the business because developers who build AI applications have standardized, in a lot of cases, on NVIDIA's frameworks for developing those applications. But NVIDIA's rapid growth has also benefited the entire semiconductor supply chain. So they aren't the only winner. And so one of the things we consider when we think about this holistically is how much direct exposure do we want to a business that has historically exhibited a fair amount of cyclicality in an industry that is certainly subject to periodic boom-bust conditions. And the question we ask ourselves is, at some point, this investment boom is going to end. And when that happens, are we going to be left with a supply glut? And how do we mitigate some of the uncertainty or the risk around the uncertainty as to when that might happen? And so we've tried to do that a couple of different ways. One way is we own a company like TSM, which manufactures a very wide range of semiconductors, not just the chips that are powering this AI revolution. Another company that we really like is called Cadence Design, and they make software that's used in the chip design process. And again, that's all different sorts of chips. And their spend is primarily levered to semiconductor R&D activity, not necessarily revenue growth. And if you look through cycle, semiconductor R&D tends to be a lot more stable. Another company that I'd mention is Amphenol. They make the copper interconnects, which is a component in NVIDIA's latest chips. But they're also a pretty important supplier to almost every other category of electronics. And so while we appreciate NVIDIA and the strength of the business there, and we own it, we've also tried to think about ways we can invest behind the AI theme while taking slightly less risk. So that's on the component side. I think when it comes to sort of the infrastructure side of things, we've found fewer opportunities. There's a lot of companies that are seeing what I'd characterize as profitless growth, where there's a lot of competing suppliers, or it's an enormously capital-intensive business, or they've got really stretched balance sheets. And so we're less enthused about those sorts of opportunities, even the ones that are growing quite quickly. And so I'm purposely sort of not naming names here, but just going back to the point I made earlier about some of the rampant enthusiasm for some of this stuff, I think in the aggregate, there's a ton of market cap that has performed extremely well this year that we think when you step back and take a longer-term view of things, they're unlikely to see sustainable profits from AI over the longer term. I guess from our perspective here at Manning & Napier, we have always invested with three very distinct investment strategies. Do you think it's fair to say, given the way that you steered the conversation more towards kind of the compute hardware side of things, it's easier for us to find strategy fits on that kind of 60% side of the spending versus the 40% on the infrastructure itself? Oh, absolutely. I mean, there's more profiles in market categories that are consolidated, that have pricing power, where there's very few alternatives for their customers, and that's a characteristic of semiconductors and much of the semiconductor supply chain, which is why when we think about what are the right ponds to fish in here, we're fishing in those ponds, and I think it's sort of the exact inverse in a lot of the other areas that we've explored. Great. I appreciate that. So with that, let's move on to the data center operators themselves. Who is it that owns and operates these, and how do the business models and economics kind of vary across ownership or operator type? Yeah. So there's three categories of spenders. The biggest ones are the hyperscale cloud service providers. So you think of your Amazons, your Googles, your Microsofts. Those are companies that operate big, diversified cloud infrastructure businesses with hundreds of thousands of customers. And those are businesses that have historically been insulated from competition because they benefit from massive barriers to scale, huge barriers to entry. It's also very hard for customers to switch away from them. And they tend to bundle lots of different software with the hardware that they're renting out. So they actually make a really good margin despite these businesses sort of resembling utility-style providers of compute. The second and newest category of operator are the neoclouds. And when I say neoclouds, I'm referring to companies like CoreWeave, Nebius. I might even bucket Oracle in there. And those businesses are just architecturally and operationally very different businesses than the ones that are being operated by the hyperscalers. Those companies are effectively reselling access to GPUs, often just for a handful of customers. And because GPUs are in such scarce supply, NVIDIA has actually played a pretty interesting role in facilitating their growth by allocating them, by favorably allocating chips to them, which is, in some cases, forced even some of the hyperscalers to sign capacity commitments with the neoclouds. And so there's a lot of interesting sort of odd dynamics playing out in that space. The last category are the data center REITs, who I would consider more of your legacy providers of colocation and interconnection services. And those companies are renting racks based on their data centers to a wide range of customers, whether it's enterprises or telecommunication providers or some of the large ISPs. But this is a pretty commoditized business, in my view. And in general, they've been losing share to the other two categories. So I don't really consider them meaningful players when we're thinking about who benefits from AI. Okay. So I guess taking a step back, summing up what you said, we have the hyperscalers, we have the neoclouds, we have kind of the traditional data center REITs. Who is it that are the bigger spenders here? And maybe importantly, I guess increasingly importantly, as we move further and further along and these numbers get bigger and bigger and bigger, how is the spending being funded? How is that maybe changing? Yeah, good question, so the biggest spenders are still the hyperscalers, so if you include Meta along with Amazon, Google, Microsoft, in aggregate, they're going to spend something like $350 billion on CapEx this year, something over $400 billion on CapEx next year, and they can afford it because just those four companies generate almost $500 billion in operating cash flow, roughly $500 billion, so a big chunk of the investment we've seen to date in data centers has come from internally generated cash flows from their core businesses, whether it's e-commerce or advertising or enterprise software, so the big change that we're seeing more recently, and this is within the last six months or so, is that the neoclouds are starting to sign deals with some of the large AI companies like OpenAI. We've seen NVIDIA take stakes in a lot of these companies. And in some cases, they're actually helping to finance the growth of these companies. And so the neoclouds are spending really aggressively to gain scale and compete more head-on with the hyperscalers for AI workloads. And in the aggregate, all of those companies are burning cash. And so they've had to depend on new issuance, whether it's debt or equity, because they don't have core businesses that generate sustainable streams of free cash flow. And so that's quite notable to us when we think about the sustainability of the current investment cycle and sort of where the risks lie. It's interesting that it seems like what has been predominantly cash flow funded investment is certainly transitioning to more debt-fueled investment than we have seen in the past. I think when I think about that within the context of an article I saw recently talking about Meta taking on, I think it was $29 billion in debt, which I think for the first time is one of the first times one of the hyperscalers has taken on debt for investment. Is it fair to say that companies are increasingly looking to fund investment in this space with debt? And if so, maybe what's kind of the bull case and the bear case moving forward? Yeah, to answer the first part of your question, they absolutely are. And that's because lenders are sort of frothing at the mouth to give them money. So they're able to raise money on pretty attractive terms today. The bull case is that the world is compute constrained. In other words, demand for AI is going to continue to grow faster than supply as the models get better and better. And the bulls would say, "Look, we're going to get innovations in areas like materials science, like robotics, like self-driving cars, like pharmaceuticals." And this is ultimately going to have an impact on broader standards of living and quality of life and GDP growth. And in that case, you're likely to see direct AI monetization take off and help to finance some of this buildup that we're seeing happening, which would obviously alleviate some of the need for external financing. The bear case is that these companies raise too much debt in the near to intermediate term, and the supply eventually catches up to demand, or there's a credit crunch, and we end up having to deal with some of the fallout from that, and I think in that latter case, I think we'll probably look back and assess that a lot of the capital that's flowing into this space and being invested in data centers will perhaps not have earned the ROI that investors thought it would. So I think that partially goes to kind of the next question I had. But where do we see the risks as being concentrated in this part of the value chain? Well, I think most of the risk is concentrated with the neoclouds, so the CoreWeaves, the Oracles, the Nebius, relative to the hyperscalers, they're burning cash, they tend to exhibit significant customer concentration. As an example, CoreWeave basically has three customers: Microsoft, and OpenAI, NVIDIA, and Meta, and their biggest customer, Microsoft, is also a direct competitor, and so when I think about the combination of customer concentration, the amount of debt that they already have, the amount of debt they're likely to need going forward to finance this buildup, I think that puts them in a very risky position. We've actually evaluated the unit economics for these businesses, and CoreWeave will tell you, they're not assuming a payback on their capital investments until the fourth year that they're renting out GPUs or the chips that they're buying from NVIDIA to run these AI workloads. And that's in contrast to what we're hearing from practitioners, which is that in some cases, NVIDIA's chips will only last two-to-three years under heavy utilization for things like model training. And so I think the math just doesn't work for some of these businesses. And it's pretty far from certain that they ever make money, even in a best-case scenario. So I think the answer to that question and maybe your comments on the data center REITs earlier probably speak to this last question I'm going to ask. But from an investment perspective, how do we think about this layer of the value chain? Yeah, so you can probably guess we don't have a very optimistic view of the neocloud providers, including Oracle. I think operationally, these businesses look much more like financing vehicles to us than they do like real businesses, and oftentimes, I think they aren't adding much value other than letting their customers keep debt off their own balance sheets, so on the other hand, I do think the hyperscaler incumbents are pretty well positioned to benefit from growing adoption of AI in the long run. They're effectively toll road-like businesses. They're highly diversified. Amazon, Microsoft, Google, those are very well-capitalized companies, and we remain pretty heavily invested in those businesses for clients. So at this point in the conversation, we've talked in detail about the physical infrastructure itself that's supporting the AI buildout. And I want to move on to the models themselves. Who are the main players in this space? Yeah. So there's basically five Western providers who are still viable competitors in terms of their ability to invest at the scale required to create AI models at the very edge of the technology frontier. And those companies are xAI, which is one of Elon Musk's companies, Anthropic, OpenAI, Google, and Meta. And then you have a couple of big Chinese players who are also in the space, like DeepSeek and Alibaba. Okay. And the amount of money being spent here, it's just staggering. Are these companies funding the buildout and the work they're doing from revenues and cash flows? And if not, again, where is this funding coming from? So it varies. These companies, just to provide some context, just the very narrow set of companies that we just mentioned, I think there's five or six of them. They're going to spend something like $150 billion just training next-generation AI models next year. So that is explicitly not a revenue-generating activity. That's investment just aimed at making the current models better. So there's a huge cost here. Google and Meta are funding those costs with their existing core online advertising businesses, which are enormous and enormously profitable. The other players are in a very different boat. So OpenAI has been raising money like crazy from a very wide range of sources, including NVIDIA, which recently committed to investing a significant sum of money. Anthropic has gotten funding from Amazon and Google, but similarly is highly reliant on ongoing debt issuance. And the same goes for xAI, which just raised $20 billion yesterday in its latest fundraising. So you brought this up or referenced it a couple of times in that answer, but it feels like, especially over the course of the last couple of weeks, the pace of these deals that we've seen being announced in the space just seems to be picking up. Can you talk about some of the deals that we've seen? And again, maybe give us kind of a bull argument and a bear argument as it pertains to what I think increasingly looks like vendor financing style deals. Yeah. So it's been really hard to keep up with, frankly. The two big ones were NVIDIA recently announced that they would invest up to $100 billion in OpenAI, contingent on OpenAI deploying a certain amount of data center capacity using NVIDIA's chips. And that came shortly after OpenAI announced a $300 billion deal to lease infrastructure from Oracle, which the market greeted with the enthusiasm you might expect in the current environment. I think a bull case here is the same as it is for the neoclouds. I think if the world is structurally short compute capacity, and it's a matter of time before the models get good enough that they can not only augment, but actually automate huge swaths of knowledge work, then you could think about NVIDIA giving OpenAI a $100 billion equity infusion as just accelerating the timeline until we can all go to the beach and live on our universal basic income checks, but the bear case is that we're sort of rapidly inflating an investment bubble, and we actually don't need five different companies all investing hundreds of billions of dollars to make models that are more or less the same. So some of the stuff we're seeing on the vendor financing front is sort of eerily similar to what happened in the lead-up to the tech bubble popping in 2000, and that definitely gives us some pause. Right. So in staying consistent with the way that this conversation has kind of flowed, I am going to ask you, from an investment perspective through our lens here at Manning & Napier, how do we think about investing in this layer of the value chain? Yeah. So we're kind of in a cyclicality in debt. My sense is that we're almost certainly going to see some level of overinvestment if we haven't seen it already. There's this saying that you never see a shortage that doesn't end in a glut, right? And so what I'm most wary of are companies with stretched balance sheets that are who aren't in control of their own destiny because they're dependent on ongoing debt and equity issuance. The other category of companies that I'm a bit wary of today are the companies that are signing these big deals with some of the AI model companies like OpenAI, where OpenAI is committed to spending tens or even hundreds of billions of dollars that they don't yet have. And by the way, OpenAI has publicly stated that they plan to burn over $100 billion in cash between now and the end of the decade. And so when you think about the lead time associated with these investments and bringing on the capacity and some of the future committed capital, there's a lot that can happen between now and the end of the decade. And so I think as an investor in an Oracle or a CoreWeave, for instance, you're really hoping that we don't get some sort of credit scare that wipes out liquidity or that investors don't get disillusioned with AI because I think the funding that these companies are currently relying upon could dry up pretty rapidly. And just, I guess, for the interest of clarity, as we've talked across the layers of the value chain thus far, the first two kind of physical infrastructure layers have presented, I think, pretty ample opportunity to invest. Who are the publicly listed players in this part of the value chain that we, I guess, investors are actually able to get exposure to through the equity market? Yeah. I mean, most of the model providers are private, obviously. But I would say Oracle has sort of become a public markets proxy for OpenAI given the size of the commercial agreement between those two companies. Microsoft is also another way to play growth in OpenAI, one that we think is a lot safer because Microsoft is basically monetizing engagement with ChatGPT, which we'll talk about in a second. The other way to get exposure to the model companies is via Google and Meta, who are both investing very heavily in their own leading models. And so Google is one name where we have a particularly favorable viewpoint on because they're sort of vertically integrated in a way that a lot of the other companies aren't. And we think that there's a good chance that they're sort of a low-cost provider of AI over the long term. And so those would be sort of the handful of ways that you can play this layer. Okay. So let's move on to the last part of the value chain now, the application providers. This is admittedly an incredibly broad set of companies. So could you talk to us a little bit about just kind of what this layer entails? Yeah, so all the big model providers that we just talked about are playing at the application layer with various chatbots. So you think about applications like Grok, Claude, ChatGPT, Gemini, or Meta AI. And my guess is if consumers played around with a bunch of different chatbots, they probably wouldn't notice meaningful significant differences between them. But the models are also powering a very wide range of applications and tools. So things like Microsoft's Copilots, those are powered by ChatGPT today. There are some coding assistants that developers use, like Cursor and Replit, that are powered by Claude. But for all intents and purposes, ChatGPT is sort of the thousand-pound gorilla. They have the most usage and the most revenue. I think they've scaled up to 800 million users already. And they've got a revenue run rate that exceeds $10 billion. And so when you think about the application layer, I would think primarily ChatGPT today. Okay. And I think the important question here, really for everything we've talked about so far, is what sort of revenue opportunities do we think ultimately exist at this level of the value chain? And maybe more importantly, as it pertains to some of the numbers that we're seeing in terms of investment projected into the out years, what sort of revenue assumptions do we have to underwrite to believe that we're going to see the type of investment that is being talked about? Yeah. That's one of the key questions, right, when you think about the sustainability of all this. So in terms of your first question, where does the revenue generation come from? I'll sort of split it into consumer and enterprise. So on the consumer side, most of the revenue today is coming from paid subscriptions to people who are paying for the more advanced versions of ChatGPT and Gemini. And so it's mostly paid subscriptions today, but I think it's a matter of time before we see the AI chatbots are all going to monetize via advertising eventually. And I wouldn't be surprised at all if five years down the road, these AI chatbots have meaningfully expanded the market for digital advertising. I expect that to ultimately be the primary way that the consumer activity on these AI products is monetized. On the enterprise side, we've seen pretty rapid growth in adoption of AI coding assistants. That's the dominant use case today, and the biggest. Cursor was the fastest enterprise software company ever to scale to $500 million in annualized run rate revenue. But if you take everything I just talked about in the aggregate, and while it's growing really quickly, the total amount of revenue generated from all those activities is something on the order of $15-$20 billion today, and you might notice that that pales in comparison to the hundreds of billions of dollars currently being spent on infrastructure and chips already and the trillions, and that's trillions with a T, in future committed investment. And so when you think about that scale of investment, you're presuming not just continued exponential growth in all of the revenue-generating applications I just mentioned, but you're also betting on AI having a much more meaningful impact in areas that thus far have been a lot slower to take off, particularly in the enterprise. And you go back to some of the things I mentioned about in terms of robotics and self-driving cars and things like that. And so I think we'll get there eventually, but I'm not a believer in the idea that AI is going to automate all the jobs anytime soon. And so our view is that there's likely to be a large and growing mismatch in terms of the scale of monetization at the app layer and the amount of capital being committed in sort of the lower layers of the stack. So maybe building on some of the comments that you had in discussing that, where are the investment opportunities that are being created at this level of the value chain today? So most of the fastest growing application layer companies are still VC-funded. They're private. So it's hard to get direct exposure to them today. Google is one of the exceptions with Gemini. And as I mentioned, we are pretty favorable on their competitive positioning here. So I'm optimistic in the long-term potential for new company creation because of AI. But I think in sort of the near-intermediate term, the primary way to play growth in the application layer is by buying the hyperscalers because that's where a lot of these applications actually run. And so just to give an example, ChatGPT, when you're creating an image or having a conversation with ChatGPT, that's a revenue-generating activity for Microsoft's Cloud. And so that's sort of the primary way that we get exposure to this activity. I'm going to ask you a bit of a different question that I think is directionally similar to that as well. When we look at the market today, we talked a little bit about this kind of before we got into the discussion of the value chain itself. But anything AI-related has been something of a vacuum for capital and financial markets in the real world over the last several years, really. But I guess in most cases, when you get a dynamic like that, you get capital being pulled away from other parts of the market, right? There's a cost to putting money. There's an opportunity cost to putting money in one place versus another. Can you maybe talk about any opportunities that you have seen or are seeing being created in the marketplace or maybe that rush to pile into AI has kind of created attractive investment opportunities in other companies that people may be incorrectly diagnosing as, I guess, long-term losers of AI? Yeah. So if I were to frame the debate a little bit differently, I think the market is sort of trying to figure out to what extent AI is cannibalistic versus sort of additive in a lot of these different areas. Or maybe to say it differently, to what extent is AI a disruptive innovation or a sustaining one? So if it's a disruptive innovation, I think you're going to see a lot of new company creation that ends up, and those companies take share from some of the existing players we see today. If it's a sustaining innovation, you would expect a lot of the incumbent companies to actually benefit from AI. And so it's sort of a nuanced question. I think it's probably going to differ sector by sector, category by category. But I think in general, we're biased towards it being more of a sustaining innovation, which I think is probably a contrarian opinion today. Two areas that have been sort of left for dead and look pretty cheap to me are enterprise software and IT services. I think there is a consensus view today that AI poses a disruptive threat to these companies that we think will ultimately be proven incorrect. And maybe just to dial in on software for a moment, which we've been adding to, by the way, as these companies have gotten cheaper, the bear case is that AI is going to enable a rapid influx of alternatives to the incumbent providers of application software. And that increase in competitive intensity was ultimately going to erode returns and pricing while putting a sort of a cap on the total addressable market that's available to these companies. And I think that ignores a lot of how moats in software get built. The moats in software today are often built on distribution and switching costs. And customers, in many cases, don't have any interest in building their own software. And so we think that's a pretty attractive pond of fishing today. So I think before we wrap here, I am going to ask you the question that we probably get as much as any other one. Is this a bubble? Yeah. Well, that's a tough question to answer, right? I think one of the defining characteristics of bubbles is that you really can't see their clear outline until after the fact. And I think there's a very wide range of future outcomes from here and some plausible scenarios where advances in AI do end up justifying a lot of the spend that we're currently seeing. But I think a lot of the stuff at the same time, I think some of the things that we're currently witnessing with respect to circular investment, vendor financing, the amount of debt out there, I think AI-related companies already account for the majority of investment-grade debt issuance. That introduces a sort of fragility and path dependency into the ecosystem, which changes the risk-reward equation, right? I'm hesitant to say that we're in bubble territory today, but I do feel comfortable saying that I think at some point there will be some degree of overinvestment. I think we're veering into riskier territory quite quickly. I think all of that warrants caution in the near term. Great. Now, I guess I'm thinking back about the conversation we've had. We've covered a lot of ground. We've kind of gone in a bunch of different directions. We do have some time to get the question and answer. But maybe before we do, I'll pass it back to you one last time. Do you have any quick 30-second thoughts, 30-60-second thoughts to kind of sum up the conversation we have or maybe some key points you want to leave people with? Yeah. So I don't want to leave anybody with the impression that we're huge bears when it comes to AI. In fact, I think it's hard to be a good investor in technology if you're not sort of an inherent optimist. And so I am very optimistic long-term in the opportunities and possibilities of AI, particularly with respect to some of these areas that I mentioned, whether it's self-driving cars or pharmaceutical R&D, the prospect of novel life-saving drugs. I think AI could make a really meaningful impact there. But at the same time, I think you have to look at what's happening objectively with a clear-eyed view of things and look at some of the parallels with what's happened in the past. History tends to rhyme, right? And so I do think there's a level of systemic risk that's come into the equation here, which when you're thinking about risk-reward, that changes things, right? So that's sort of the balance or the tension that we're sort of thinking about today. With that, we do have some time for question and answer. We spent a lot of time talking about kind of the domestic American AI ecosystem. Can you talk to us a little bit about China's AI ecosystem and what we're seeing there? Yeah, sure, so China's AI ecosystem is developing extremely quickly. I would say they've been a fast follower, but in some cases, they have some advantages in terms of their ability to quickly build out infrastructure, particularly on the power side, which competitively advantages them, and so you look at companies like Tencent, Baidu, Alibaba. They're benefiting from AI in many of the same ways that their Western counterparts are, and DeepSeek is sort of their domestic version of OpenAI, so it's a very dynamic space. Even though they don't currently have access to sort of the state-of-the-art NVIDIA chips, they do have advantages on the power side and their ability to build really quickly. And so we've definitely been looking at and evaluating some of the Chinese names as they rapidly incorporate AI into their businesses in a variety of ways that in some cases are similar to the companies that we own, which provides us with sort of an interesting comparison set. And in some cases, it's a different market. So we're looking at it pretty closely. One thing that came up a couple of times there in that discussion was China's ability to kind of build out their power capacity. And I think there's been no shortage of headlines about the just demand on the U.S. grid that the rise in AI is going to create. It's been interesting that we've seen some of the hyperscalers contract out captive capacity, whether it's nuclear, natural gas, certainly something that stood out. Can you talk a little bit about the impact of AI on power demand? Sure. And this is a topic that we could probably speak about for quite some time. I think there's a lot of different sort of nuances to unpack. I think one of what's interesting is that I think the most pressing issue with respect to AI energy demand is today, anyway, is less about the absolute level of energy demand and more about how concentrated that energy demand is in a few locations like Northern Virginia, Texas, and Ohio. Our estimate is that all of the data center CapEx commitments that will come online over the next three years equate to something like an incremental 3% of total grid capacity. So it's not at the scale yet where I think there's a tremendous amount of strain on the overall grid. But the issue is that within certain geographic corridors, that implies a pretty significant surge in peak power demand. And in many cases, that's actually driving up the cost of electricity for residents in those areas. And so I think the market is responding to the demand signals that are out there. But in the sort of intermediate term, there is a risk of political fallout, which might be even greater than the risk of sort of the physical bottlenecks in some of those locations. The other thing that we've noticed or heard about is that utility providers are starting to ask some of the hyperscalers and other companies that are making these big data center spending commitments for more money upfront. And that's in an effort to prevent companies from just making a land grab for grid capacity that they don't end up using. And so when you have to commit to several hundred million dollars or $1 billion upfront, that's sort of an incentive not to just go out chasing deals. And so that is something that I think could possibly slow the pace of build-out sort of at the margin. So we've touched on China. We've touched on power. And I think maybe the last question for me, what impacts are we seeing from AI on the labor market? Yeah, so there's a lot of conjecture out there, a lot of theorizing about how this is all going to play out. I think there's maybe a narrative out there that AI is already leading to widespread productivity gains, and it's a matter of time before autonomous AI agents are going to replace huge swaths of knowledge workers in a variety of fields. I think if anything, if I were being charitable, there's some anecdotal evidence of that happening in very small pockets, so if you think about automation of call center operations or enabling developers to write more code and become more productive, I wouldn't say that in the aggregate, there's any sign that AI is having an impact on the labor force today. The best measures I've seen suggest that while the occupational mix is changing more quickly than it has in the past, it isn't a large difference, and it actually predates the widespread introduction of AI into the workplace. And I think even if you classify jobs by how likely they are to be automated, so you can look at things like manufacturing versus web development or something, we're actually seeing job openings decline faster in aggregate in the areas that should theoretically be more insulated from AI disruption risk. So it's definitely not apparent to me that AI is a huge factor in the labor market today. Well, James, I've really appreciated the conversation we've been able to have, right? We've covered a lot of ground. I think you've been able to take a very complex topic and kind of simplify it layer by layer. We've been able to talk about who the players are, kind of how we think about it from an investment perspective. And I've certainly appreciated the conversation that we've been able to have. With that said, we have come to the end of the time that we have for today. If anybody had a question that we didn't get to, we'll be sure to follow up with you after the presentation. And please continue to reach out with any questions, any follow-ups you might have. As a reminder for your convenience, recording of this webinar will be emailed to you. And again, please feel free to explore the resources tab for insights specifically on this topic. James, thanks for being here with us today. And to our listeners, thanks for attending the webinar. Thanks.