AI Capital Spending Impacts

Speaker A: Welcome to Next in Tech, an S and P Global podcast where the world of emerging tech lives. I'm your host, Eric Hanselman, chief analyst for industry research at S and P Global. And today we're going to be taking a look at AI capital investment from some angles that we haven't touched on before. And with me to discuss it is our resident optimist, Bren Daly. Bren, welcome back to the podcast.

Speaker B: Oh, it's great to be here, Eric. Always fun to talk with you and always fun to talk money.

Speaker A: Well, we've talked a lot about different aspects of the heavy duty capital spend that's going on supporting AI and talked about what uh, all of the various data center demands happen to be. We also delved into uh, the market's reactions to all this spend, which is one of those areas that I've been a little surprised about. The markets seem to be sort of wishy washy as to whether or not radically increasing the amount of capital that you're dedicating towards infrastructure is a good thing or a bad thing. Is this necessary to be a player in the AI world or not?

Speaker B: Markets at this point seem to be

Speaker A: okay with the levels which hyperscalers are spending. Some of the neo clouds are being lent to debt, uh, markets that are, that are kicking up. But you actually done some research recently that I thought was interesting about another aspect that seems to be coming into play, which is a lot of the AI capital volumes having an impact in areas of investment and M and A. Can you dig into a little of what you've been looking at and what you've been finding?

Speaker B: Yeah, I guess, you know, from my perspective again, and I'm looking mostly at the capital flows and when you think about capital allocation, a company is just a series of decisions, right? It's what do we invest in, in other words, where do we direct capital? Much in the same way that like our lives as personal finance, right? They're just a series of uh, or they reflect a series of decisions that we make. Do I buy a new house? If so, where, how much do I spend? Do I want this new car, Do I invest in a new wardrobe or whatever, right? So individual decisions collectively, that makes up corporate uh, strategy and the decisions historically for technology companies particularly I'm speaking about the large cap hyperscalers now but going back, uh, in the early days of the technology industry, the large cap companies have basically two paths that they can go down to develop products, right? Because growth is and absolute in the technology industry. It is the determinant of Value So how much a company it's worth and it is a long term strategy that you have to follow. Right. That's the North Star for all of technology has always been growth. And so companies achieve this by one of two means. Largely broadly speaking, you either buy or you build. And what we saw for example in the early, so you know, rewind the clock 20 to the 25 years, go back coming out of the dot com collapse. Right. A lot of that time was built on, a lot of the development there was built on, was built organically. It was built on acquisitions. And certainly when we came to the cloud era through call it after the great financial crisis, up until the pandemic, largely we look at that past previous decade and it was just by applications consolidation. Right. And when we look at the large technology, large cap technology companies, they were just looking at consolidation. Take a company like Oracle, you have the database, you start stacking apps on top of that, they're not having their engineers write code. They're going out and buying NetSuite, they're going out and buying Eloqua. Right. They're buying their way into all of these adjacent markets. And in contrast, and this is what's been striking mostly if we fast forward to state of play now, is that for the AI development and the build out has all been done off balance sheet and supported by capital markets obviously. But it is not a case where Microsoft is an investor in OpenAI, but it is not an owner the way it owns GitHub.

Speaker A: Anything that it happens to have acquired and built in and built through inorganic growth, as you were pointing out.

Speaker B: Yeah. And Microsoft over the past quarter century, you're talking north of 200 acquisitions, including the largest tech acquisition in history, which was them buying electronic cards. Right. Also they own LinkedIn, they've amassed all of these properties. And yet when we look at how they're approaching AI, it is very much writing checks to Nvidia. Instead of writing checks to a uh, startup and their investors, they're writing checks to Nvidia or wherever they're sending their money in the AI. But it is very much development as opposed to acquisition.

Speaker A: Well, but they've always had to build infrastructure. They were writing checks to intel for CPUs and things. And this is only what, a couple, three orders of magnitude larger than that. So what could go wrong?

Speaker B: Right, well and uh, that's the thing is that yes, they've always, but if you look, go back in the uh, collectively the capex for your hyperscalers again during the cloud era, they were building out data centers much like they are today. Right. They were writing checks to the providers, both on the facilities and the technology, but they were writing checks at a pace of collectively about 40, 50 billion a quarter. Now they're writing checks collectively of 250 a quarter. Right. And this is only increasing. So what we're seeing is that because all of the expenses of uh, of AI are recognized, uh, the expenses are upfront, the build out has to be paid for and there's a multi year lag or at least several quarter lag. All of that backlog is for 2027 and 2028 and so on. In the cloud era, what we saw was yes, you have to build out a facility, but it was usually one or two quarters later that the revenue would start to flow directly back to the vendor.

Speaker A: It was still build it and they will come, but yet they were coming rapidly enough and that cash flow that backend infrastructure was generating was hitting the balance sheet that much more rapidly.

Speaker B: Yeah, everything comes down to liquidity, right? How liquid is a market? And right now what we're seeing is liquidity is very hard to find because most of the flow, the capital flow is outgoing at this point. And it's gotten to the point where many of the providers, and these are again, Microsoft will do 200 billion of revenue this year. Right. Amazon does 200 billion in a quarter. Right. These are hundreds of billions of dollars of revenue. Right, Good. Well established companies that know how to run at that kind of scale. Hundreds of billions of dollars of revenue growing by the way, at uh, high teens, almost 20% in some cases. These companies know what to do with large chunks of cash. And in the past they've been able to just generate that cash and go out and do whatever they want with it, but they've now exhausted it. Right. The demand for upfront expense in the AI era is so high that they've exhausted all of the cash. Google for the first time in its history reported negative free cash flow. This is a company that used to do $70 billion of free cash flow. And that, that's money that just after all of your expenses, after all your salaries, after all of the rent, after all every single expense that a, uh, company faces, they still have play money left over and now they don't. And it's only going to get worse for the next two years. That's eight quarters they're projected to be cash flow negative. This is a stunning reversal.

Speaker A: And so they've taken that big step backwards. And so you say they don't have as much play money. I think one of the things that was interesting about that was at a point in some of your research, that means they are not playing as much when we take a look at those markets. And it seems like suddenly what had been the engine of growth for these businesses starts to then point back to if you can't grow inorganically, you gotta grow organically. And there seem to be challenges on that side as well. But uh, I guess to begin with one, the inorganic growth thing is clearly a bit of a challenge because we've seen slowdowns, right?

Speaker B: It is, but their core business right now. And back to your earlier, earlier point about the market, kind of looking through the capital allocation decisions that many of the hyperscalers are making. Right. And the best sort of piece of evidence I would put forward on that is Google was able to sell $85 billion worth of equity back in March. They sold. And by far that's for people who don't know. $85 billion, it was the largest, by far the largest equity offering in history.

Speaker A: Okay.

Speaker B: And the market absorbed it and they could have sold another probably 15 or 20. Uh, so that's on the equity side. And again, think about, they're raising money because they don't have the money that they used to at hand, right? Their corporate treasury, which has been stuffed because they have a very slick algorithm. And historically they have just used that algorithm to place ads, sell ads, maximize ads, optimize all of this. And it is humming along and it is generating tons of cash and it will, because it's a fairly mature and very well established and it has 90% of the online advertising market. So that's going to generate cash just as the Windows franchise is going to continue to generate cash for Microsoft and so they can choose to reinvest it. Amazon took the uh, cash that was generated by their online retailing, which they're now the single largest employer in the United States by the way, Amazon is. And then they funded 15 years ago, 20 years ago now, they funded the development of AWS, right. So they didn't buy anything really.

Speaker A: Uh, not in that same scale as others have gone out, but yeah, exactly.

Speaker B: They didn't buy anything to really launch aws. They just reinvested the cash into developing this business. And it's somewhat analogous to AI. It's just they're adding zeros to the price, right? So instead of putting 40 million, they're going to put 400 million. That kind of scale. It truly is an order of magnitude. And again, the problem Is that the return on that investment from a corporate investment side, the return on the investment is much longer off, uh, than it is for an acquisition. Right. Because if you think about an acquisition done well, mind you, not a frivolous dilutive acquisition, but an acquisition immediately contributes, once it's closed, immediately contributes to the top line. And you can recognize, ideally you can recognize deficiencies in the business, therefore become more profitable. So you're growing, which is the key check mark that Wall street always wants to see. So you're growing and you're making more money. Keep doing that. Like Oracle, it was a database company that almost went out of business in the 90s.

Speaker A: Rinse, repeat. And all you got to do is keep building out infrastructure to support it

Speaker B: more and more, Spend, more spend. And so you just keep expanding and expanding. And that worked during the previous transitions from mainframes to client server, then from client server to the cloud. That worked swimmingly. Right. We had huge software platforms emerge. And by and large software was incredibly profitable. I mean you're talking about 80% gross margins and 30, 40% operating margins for most software companies. Software was like fat and happy. It was great to be a software company. Right. And so you could just take all that extra money and go buy another fat and happy company, bring them together and you suddenly have a larger, more profitable company. And Wall street says, oh, that's worth more next year than it is this year. So we're going to bid up your stock.

Speaker A: It was a great restaurant, simple formula, easy to do. What could go wrong?

Speaker B: Exactly. And then these agents, these pesky agents just threw it all away. They just cratered the industry. And we thought somehow we were going to vibe code millions of lines of software that was going to run our businesses. It's not going to happen, mind you. But so somewhere between the two ideas that agents are going to displace all applications and we'll never touch workday again, we'll never open service and we'll never file a uh, help trouble ticket on servicenow again because agents will take care of it.

Speaker A: Yeah.

Speaker B: And I, I think probably not in my professional life.

Speaker A: Yeah, There, there's still that sort of systems of record thing that, you know, the data has to be somewhere and you uh, know there bits and pieces are there, but. Okay. May erosion in that business. Okay. Yeah. But hey, if you get fat margins to begin with, okay, maybe you give up a few points on that so that business seems to be okay.

Speaker B: Yeah. And what we saw when the immediate threat of extinction for SaaS vendors with SAS apocalypse. When that immediate threat lifted, we saw Salesforce get back to doing M and A, right? They bought contentful, they bought, you know, Those were billion dollar plus deals, both of them, right. ServiceNow got into Autodesk slot maintenance. Right. So these were all multibillion dollar deals, SaaS deals by design. Clearly refuting the idea that we're never going to sign another subscription in our lives except to OpenAI. So for me, that was a signal of, uh, the market is regaining some, um, moderation. Right. Because I think when we look at the development of AI, it was a technology led develop market, right? It was technology led. Look at all this cool stuff that OpenAI can do. ChatGPT, you just ask it a question and it answers everything. When it came out almost four years ago now, when it came out, there was incredible mania kind of took over and it was led by technology and we just started ascribing all of this power to AI, of what it could do, how it could improve our businesses, or even in the early days, how the changes it can make in our personal lives. All of this was just blue sky and valuations were only going up. And it was that point that the capital markets got caught up in that mania. And the animal spirits weren't just in the technology, it was also now in the finance side.

Speaker A: Interesting to see that identification of the animus that starts to grow and all that. And hey, we've got that little tulip craze thing to talk about. Who wouldn't want these amazing new varieties of tulips and all these other things that are gonna, you know, just be more valuable forever and ever? It only goes up, right? So with the animal spirits unleashed in the capital markets, it sounds though that they're all going out and looking to go devour infrastructure and building out more and more to be able to go feed the AI beast. To extend that metaphor, what happens now at this point, because we've had strategic acquirers, the, um, hyperscalers have been the folks who have been doing, who have, uh, been the linchpins of a lot of that inorganic growth piece. So what seems like it's happening now, suddenly they are not the principal acquirers. Yeah, you got SaaS providers out there who maybe pick up some of that slack. They've still got some cash in their coffers to be able to do that. Where does this start to go? Does that mean that they stop? That any level of acquisition activity starts to become, um, an issue? Do we have to follow the money trail to other places with acquisitions like to the, the people they're writing checks to. We've just seen this really interesting Nvidia and Mediatek transaction. Wow, nice. Somebody's clearly got some money out there. It just happens to not be what have been our traditional strategic acquirers.

Speaker B: Yeah. And um, the key is the market as an uncertain entity. It looks for signals, like whole markets look for signals. And if you think about Hollywood, a studio in Hollywood develops some superhero movie. Every other studio has to develop a superhero movie, right. And we saw it in Cloud. Google Bought Wiz for $32 billion, largest venture capital exit in history. 30 times trailing sales at the time. Unheard of, multiple. Now in the AI era, that's a down round, honestly. But acquirers look around at their comparables and they see movement and they see dollars flowing and they say, hey, you know, it's kind of fomo, right? We can't afford not to do what Microsoft is doing or what Amazon is doing, or what Oracle is doing, or what Nvidia is doing. Right. There, there is. It's natural comparison is a human state. Right. We live in comparisons in many ways. And going back to this idea of strategy is nothing more than a series of decisions. And right now the key to understanding how this plays out is these decisions that they're making now around AI, particularly around the AI buildout. These are decisions that have ramifications for the next two, three, four, sometimes even five years. When you look at the actual obligations, the contractual obligations in these backlog orders, they are for 2028 in many cases. Right. If you think about a data center being developed today wouldn't even come online, you couldn't get it permitted, you couldn't get the foundation dug until 2027 at the absolute earliest. More likely is probably 2028, honestly. Right. So you're making decisions again back to that. It's like decisions reflect values and the value that tech always has is growth. And the problem is they are locking themselves into multi year contracts. And along the way all of those contracts are going to take money to support the operations, the infrastructure. Right. You got to pay the lighting bill, the heating and cooling, you got to pay electricity, you got to buy new servers. Suddenly Nvidia chips depreciate quicker than we thought. So we've got to re up that. A lot of these decisions that are being made today are going to continue to weigh on the company. So even if they tonight said forget these AI, we want to go back to doing the kind of deals Microsoft just for is in its heyday, last decade, it was doing at least a deal a month, 10, 12, 15 deals a year, year in and year out. It was that kind of machine. Similar sort of cadence for Oracle back in its prime when it was mopping up all of these. You know, same with SAP when it was expanding into travel and expense, and then it would expand into marketing and then it was expanding into E commerce. Right. All of those are billion dollar acquisitions. Plus for SAP, built on their core platform of erp. So they just had this business and they used the cash flow that this generated. Spit out this cash, go buy this next part.

Speaker A: Flywheel keeps spinning and life is good.

Speaker B: And even if they wanted to dust off that playbook, even when Microsoft said okay, or Google said, hey, we want to go buy Wiz again, they couldn't do it. They would have trouble finding, you know, financing for that transaction. So even if they wanted to, they couldn't. Which suggests that this drought, at least among the AI hyperscalers or however you think about those infrastructure providers, M and A, is not going to come back. And all markets need leaders. All markets need something to set the tone in the overall market. Because what Microsoft did and what Oracle did, that had ramifications for everybody beneath them.

Speaker A: So this winds up starting to look like pick your model. Is this sort of the El Nino effect that's coming to. To M and A? This sounds like we've winnowed down the set of potential acquirers. Certainly pulled a lot of money off the table in terms of what's available for investments. Sounds like maybe it's a handful of folks who are still making some level of money. Unless, uh, it seems like the other piece that starts to come to this is the expectation that suddenly there will be a lot of revenue. And that seems like that's that that other piece of this, which is somebody's. If they're going to get back to their winning ways, they got to start making some more money in ways that they have not demonstrated an ability to do so far. Uh, so are we thinking that the cash is going to start come rolling in off all this stuff? We're starting to see some of these models change for AI providers in terms of what they're charging. And we're not really seeing massive amounts of revenue being generated out of this, or at least not on the scale that's comparable to the capital investments that it has to support.

Speaker B: Yeah, um, on the model side, it is clearly inverted, right? For every dollar they take revenue, they're burning three, or at least OpenAI as anthropic, is more restrained, more Fiscally responsible, and they're talking about perhaps getting to break even next year. But just looking at OpenAI, when you think about so much of their economics overall, so much of it is yet to be resolved. And so it's very hard to know like how the model of the models, what does it actually work? And we look at right now, it is all consuming. It is just consuming. It's consuming capital, it's consuming information, it's consuming markets, it's consuming. It's all consuming. But there has to be a period where there is some payback, where there is some return. And right now we don't know exactly what that's going to be. Even open AI, right. They want to bring in enterprise sales. Okay. And so they hire what's her name from Slack, and she lasted a year there. Then they say, no, we're going to have advertising. So we hire a Facebook person to run our advertising business. And it's like, you know, just keep trying. But meanwhile, as you're experimenting with business models, the cost of that service is only going higher in I think also what I see when I look at projecting out. You've been in technology longer than I have and you know the general direction on pricing is deflationary. Right. Uh, it disappears into at some level, commodity. And whether you're looking at mobile, whether you're looking back to CPU power, look at your first iPhone versus what you're walking around now with now, it's like it doesn't even. It's not even the same species.

Speaker A: We've been doing smaller, faster, cheaper forever.

Speaker B: Exactly.

Speaker A: And that last one has got a little bit of a zinger in this

Speaker B: model because it's all predicated on this idea of AI intelligence or artificial intelligence being a premium offering. And I think what we're ultimately going to find is I could have said the same thing about mobile. You can take your data anywhere and trying to extract a premium out of that.

Speaker A: And people will Pay more for 5G. Right?

Speaker B: Exactly, exactly. So it's just deflationary pressure all the way around. And I love seeing what open weight the models now that are coming along. Right. And I think about it just in our own work in our own shop here, that I think is far more likely to be adopted. An on prem open weight model, cheap, lightweight, but sufficient for the task at hand. We don't need Mach 1 when we just want to go to the store.

Speaker A: And that whole the pursuit of tokenomics now as a concept model, uh, routing so that you can do cost optimization to not what is the best model. But the primary focus is what is the least expensive model. All speaks to those decreasing margins, lowering costs, lowering revenue. Yeah, exactly.

Speaker B: As you're identifying it speaks to also I would say a fundamental disconnect between maybe your world and my world. So in your world we'll just have you be the stand in for the proxy for technology. Okay. So that's the hat you're wearing. And you listen to AI and uh, um, OpenAI or even Cloud for that matter. And they're so focused on the improvements from rev to rev. Right. Feeds and speeds and how we're going to get to AGI before that next guy. And we're. No, we are, no, we're. This model is tuned now to be 10 times more. It's PhD level times 10 or whatever. And they're all like going back and forth and like the user me doesn't care. And the finance guy, I'll be the stand in. Since you are the technology, I'll be the finance. The finance guy for me says but to get to this performance level, which is not needed by the way, because demand, which is drives all markets, the demand is for just something.

Speaker A: Okay, good enough.

Speaker B: Yeah, good enough is good enough. And they're convinced that like it has to be, it has to be autonomously constructed thought. And it's like no, if you want to just like bang on. If you wanted to digest a prospectus or a uh, debt offering or whatever you want it like, like the earliest model. Just fine, the earliest.

Speaker A: But I'm looking really cool in my 10 gallon AI hat here. And I guess this is the point where you come in and say big hat, no cattle.

Speaker B: But also keep in mind like every rev and every iteration, every new model improvement, every release cost exponentially more. Right. Whether both on the training and on the inference. So this cost, you're increasing costs at a nonlinear rate to the, to what actually people need. Right? Like we just need. Okay, fine. Like out on the model frontier. We want those frontier models.

Speaker A: Yeehaw. Come on.

Speaker B: Yeah. And if you're doing cancer research, sure, right. Or if you're patterning weather, you're trying to forecast weather patterns for like that have huge multi billion dollar implications for crops and people and lives and all that. Absolutely. Use a new model. Go head, but treat yourself. But 80% of it is uh, 80, 90, I don't know, but is probably if we went back we would never even know. And that's what the model routing that you mentioned. And again think About Stripe paying $7 billion for open router Right. Like that was suddenly now that they not only know, like, what you're spending, but they know who you're spending to, and they know at the purchase levels, that's a very powerful seat to be in.

Speaker A: It is indeed. As always, your positivity is shining through on this. And I guess this is one of these things where there are at least so many different concerning factors to this. It does seem like we're getting into a world in which there's a lot more that we've got to start keeping an eye on in or agus at a point in which there are a set of indicators that are maybe starting to flash yellow a little bit just in terms of where we're looking. Buckle up. Maybe getting a little bumpy.

Speaker B: Yeah. I mean, imbalances can only exist in markets for so long. And right now the imbalance between the expense and the income is at an unsustainable rate. And so markets and all markets are great at leveling. They, you know, markets like the human body seeks homeostasis, like, they want balance. And so the market, we got a little imbalance, like, oh. And caught up in our excitement, the animal spirits again tipped the balance way too much into speculation. And, um, now we might, we might be leveling out a little bit of actually, what is this technology? How does it really improve our business or our lives or profitability, productivity. Whatever metric we look at, there's something there and we will enjoy those benefits. Right now, we're growing into that. And unfortunately, the provider side, again, which supply important part of the market, but not the determinant of the market. The supply is way out in front of the demand.

Speaker A: And one of those things, you got to wait for it to catch up. This has been great, Brennan. I, uh, appreciate all the insights, but we are at time for this episode. Thank you for being back and we'll keep an eye on things.

Speaker B: Awesome. Thank you. Eric. Good to see you as always.

Speaker A: That is it for this episode of, uh, Next in Tech. Thanks to our audience for staying with us and thanks to our production team, including Dylan Schiebold, Sophie Carr, Ran Meadesh, and, um, on the marketing and events teams. If you enjoyed this session, please like or subscribe on your favorite podcast medium. I, uh, hope you'll join us for our next episode because there is always something next in Tech.