The Pipeline: M&A and IPO Insights | S.4 Ep. 9 - The Big Tech Players Are Spending On AI, Not M&A

View Full Transcript

Joe Mantone   00:00:00

Welcome to another episode of the pipeline, an S&P Global Market Intelligence podcast that dives deep into the world of M&A and IPOs. I'm your host, Joe Mantone. And today, I will speak with Brenon Daly, the Financials Research Director at S&P Global's 451. Brenon oversees 451's Tech Dealmaker, which is an M&A analysis service, and he also oversees Knowledge Base, which is a proprietary M&A database provided by 451. With these 2 products, Brenon is constantly reviewing tech deal flow. But what I like about Brenon's research is he's not just focused on the deals that are happening. He also makes interesting observations about deals that aren't happening. For instance, Brenon recently published a report about AI hyperscalers choosing to build and not buy. This, of course, is a product of the massive CapEx spending that's been going on. S&P Global Ratings expects $1.7 trillion in CapEx spending from the hyperscalers in 2027, and that is up from $870 billion in 2026 and $470 billion in 2025 for these hyperscalers. And the hyperscalers includes Microsoft, Alphabet, Google, Amazon Ss and Oracle. Most of the conversation today will be around AI in the build-out that they're doing. And to get things started here, Brenon, where is all this money coming from?

Brenon Daly   00:01:41

It's good to be back on the podcast, Joe, and good to see you again. And the short answer is the money is coming from anywhere they can grab it because it is everywhere and everything is funding this great AI build-out. The key to understanding this is you cited some figures around $450 billion last year, $450 billion last year, maybe doubling $900 billion, rough and tough. But if you go back, we're approaching $1 trillion CapEx spending, okay, for the AI hyperscalers. And if you just go back before the pandemic, what we're spending this year is 10x of what we spent in 2019. So this just underscores, again, that's 10x more on data centers. without -- mind you, without much discernible revenue because all the expenses upfront and the revenue is still yet to come. These are -- we're spending now for services that are going to be delivered in '28, '29, '30. OpenAI's agreement with Oracle is a 5-year agreement. And this puts the -- you have to spend today and then harvest in 3, 4, 5 years. And so the spending today, the companies are under tremendous pressure. And so back to your question, where is the money coming from? It's everywhere. They've exhausted, they burned through all of their cash at hand drain the treasuries, free cash flow went from -- if you're Google, it went from $70 billion a year to you're now for the first time ever in the case of Google, negative free cash flow. So everybody went through all the money they had at hand that businesses generated and keep in mind, technology companies are very profitable, right? They just meant money and Microsoft is 40% operating margin. So they generate a lot of cash, but they're spending it even quicker. So after the internal resources are burned through and they're burned through very quickly, when you're talking about hundreds of billions of dollars in CapEx. So internal resource is gone. So then you have to go out into the landscape and to begin to get resourceful on where is the next dollar coming from. So you've seen it in the equity market. rewind the tape about 6 months ago, Google did an $85 billion equity offering, right? It just $85 billion largest equity offering in history. That's 20 IPOs or 30 IPOs depending on the size, right? So this was a massive and the market just took it. And so that's Google raising $85 billion in one fell swoop. And then you look at the debt, right? We're on track for probably $400 billion to $500 billion in debt. And it's not just the U.S. market. Amazon and Google have done sterling offerings, euro-denominated offerings. They're looking everywhere for cash to stand up these data centers in anticipation of making it all back with some flavor of AI offering.

Joe Mantone   00:05:03

Has there been any investor pushback, whether it's debt investors or equity investors when it comes to funding some of these projects?

Brenon Daly   00:05:12

Certainly, as you would expect, early days, everybody wanted it from the investor side. There is virtually no premium associated with any offering, whether in selling debt was easy. And by the way, we are tracking now, thanks to the hyperscalers. This year will be an all-time high investment-grade issuance. largely driven by investor appetite to participate in fixed income side, to participate in this incredible opportunity, once-in-a-lifetime opportunity that all the companies are telling you about AI is going to be. Early days, there was tremendous excitement in the investor community, and that's true for both on the equity side and on the debt side because this was the first real innovation, at least in technology sphere. This is the first real innovation we've seen since cloud. So this was the next leg of growth. And it wasn't just an iteration. This was a radical new offering that was going to jumpstart technology. And technology has always been a fast-growing industry. That's why it's always been premium category in terms of investor sentiment. Tech companies trade at 4 to 5x any other sector because they deliver outsized growth. And when AI came along, it not only raised the possibility of continuing that growth, but actually accelerating it and investors wanted in. And no matter what, like they wanted in if you were -- they signed off on SpaceX to come public at a $1 trillion-plus valuation next month, we're going to get a really good indication of investor appetite when Anthropic comes public.

Joe Mantone   00:07:12

Yes. And so that's why in the early days, like every company was trying to get AI, work AI into their earnings conference calls or every time they were in front of investors.

Brenon Daly   00:07:23

And that's true in the public companies. It's also true in the private companies. And we saw, by and large, if you look at our S&P Capital IQ data, you can see a 10x premium in terms of post-money valuation for AI companies versus non-AI companies. And this is funding levels, right? And by and large, these are speculative assessed valuations, right? These are just ideas that are starting light at a huge premium relative to even a premium category like technology. So this is risk fact on top of risk stack on top of risk. And while that was true, investors were willing to stomach that risk early on because they wanted to participate, right? And you had like Satya Nadella, the CEO of Microsoft, saying, the risk to underinvestment is greater than the risk of overinvestment. And by and large, Wall Street took him out his word and said, yes, then, if that's the case, he's a pretty smart guy. He's running the largest software company on the planet. I'm going to go ahead and back him on that. And so that was the feeling was it like this is the next big thing. I want in.

Joe Mantone   00:08:43

I anticipate there's a but coming.

Brenon Daly   00:08:46

You felt the change in the weather there just plainly. And -- but this is true for all technology or any E&D market cycle. Initial enthusiasm gives way to realism. We begin -- when AI first came out, we're coming up, by the way, on 4 years since the debut of ChatGPT, which is the iPhone moment for this whole movement. So if we think just 4 years ago, when OpenAI rolled out ChatGPT, and it became broadly available and widely known and people were experimenting with it and just finding all the great things that could do for them. It was amazing. I look at, oh, it's drafting recipes based on a picture of my refrigerator. It's working out puzzles that I never would have -- it would have taken me 4 sheet of paper in the 2 hours on a calculator, and it just doesn't automatically. These are amazing things -- they're great parlor tricks for lack of a better description. But those early days were just wow. And early days, a lot of people were experimenting, having fun and playing with OpenAI and seeing what it could do and different models and playing with different apps and different wrappers around different models. And then for many people, they brought it into the workplace. And suddenly, things slowed down a little bit. Then when you got into the workplace and suddenly, it met sort of business reality where it slammed into regulation and administration and business practices or what have you, it stalled out a little bit. People started to say, I'm not sure about AI, right? We did a trial program and it didn't really click some research, just to put some numbers on some research that we do at 451, and we talk to end users. So not people who are in the interest of talking their own book, if you will, but actual people who are rolling up their sleeves and doing AI. And when we talk to them, we ask them, okay, so you're doing AI, what are you getting out of it? And in our latest survey, half of them -- this is half the users of AI told us that they don't expect to see any return on their AI investment for the next 12 months. So when you think about, again, upfront expense, complex technology, every iteration is exponentially more expensive than the previous one. And yet we don't have that much to show for it because it is still an emerging technology, right? It's not priced that way, but it is actually just finding its feet, ascribed multibillion-dollar valuations to a technology that we don't honestly know how reliable it's going to be.

Joe Mantone   00:11:50

So is that part of the reason -- you mentioned there's a couple of big IPOs in the pipeline. So some of those concerns that you've just highlighted, is that part of the reason why we're starting to see some of these IPOs get pushed back?

Brenon Daly   00:12:05

I think it's also a matter that the technology is new and still emerging, I'm making the case. These companies are new and still emerging. They don't have the advantage of a conventional business model. And if you think back to software, 80% gross margins, 30% operating margins, like that's a very bankable asset. But when you think about a software company any time in the last 25 years, basically, in terms of technology industry, you had best-in-class financial model, again, highly profitable. fat and happy. Software companies are fat and happy in an industry, they just made money. They had 20% growth, didn't matter what they did. They just cash checks. And so when you look at -- when you contrast that sort of model with OpenAI, like OpenAI, it doesn't know what it is. Like it started out serving the consumer and has 1 billion users. So it's doing something right. But then you look at the monetization rate of 1 billion users and you think, wait a minute, why are they pivoting so hard to enterprise now? And are they -- and 1 week, they're trying to sell ads. They're hiring people from Facebook to sell ads and then they say, no, their CFOs here are talking about, no, no, more than half our business is enterprise now, whereas before, they didn't really serve the enterprise. So now they have a dedicated enterprise. They hired a person from Slack to run it. I think she lasted a year maybe, not even a new person in there. So these are fundamentally immature companies, and this isn't a knock on them, necessarily, their technology doesn't know what it's going to be. So how can a company know what it's going to be? And when -- I think when you look at from Wall Street's perspective, right, Wall Street likes predictability. They like certainty. They're willing to risk a little bit. But generally speaking, when it comes to allocating pension funds, for example, or retirement funds, like they're not way out on the risk curve the way a venture capitalist would be or should be, right? So this is a different investment criteria. And I think they're wondering, hey, before we commit to taking a position in this company, we really want to know what this company is going to be when it grows up. We're willing to put some money in, but we want to have a very clear line of sight to what you're going to be doing. And so that's why I think there's a new sort of cost consciousness that's coming into some of -- particularly on the -- I'm speaking mostly about the model side, the large language models, the LLMs, the OpenAI and Drop. -- primarily, there's a new cost consciousness. And there's -- instead of talking about growth all the time, they're suddenly talking about getting to breakeven. Now they're using all sorts of financial consortiions to get to breakeven and look through cost and adjusted this and -- but they're -- whereas a year ago, certainly 2 years ago, you never heard talk about any breakeven point. It was just all growth. Exactly.

Joe Mantone   00:15:39

And along with the fundamental issues that you're highlighting here, you also put out a piece not too long ago mentioning how they're thinking about going public at a time where there's increased public backlash, the backlash being risks of regulation and also just generally people pushing back against AI, whether it's the data centers or taking my job type concerns. So how much of a factor is that? Is that just another risk that in an IPO?

Brenon Daly   00:16:13

Yes. And I think public involves the public. And right now, we've seen, again, if we think about the initial excitement giving way to wariness, skepticism, and we've seen that both in the use and now the deployment of these because these are -- what people are appreciating now is how resource-heavy AI actually is. And by resource heavy, not just the hardware and the software and the infrastructure, but also in terms of like energy, data centers are going to be using -- pick a number, but internally, S&P says, somewhere around 20% of U.S. electricity is going to be consumed by data centers in 2030, right? That's our forecast, and that's 4x what it was before OpenAI came along. This is a real impact. And when you think about, again, if people aren't getting payoffs or payback from the technology and our research points to that, at least half of them saying, we're not getting anything from this. And yet the cost, both environmentally and in terms of your community, the costs are very high, then yes, there's going to be resistance. And the key to understanding also this resistance is that this brings into threat not only just the technology, but all of the financing around it. And you can point to what's happening with the post data center in New Mexico with Oracle is leasing it, right? And they've already asked for a way out. They already said, we want to delay payments on this. Now they have an Open AI, whatever, $400 billion, $500 billion agreement over 5 years that Oracle is going to be supplying some of the AI infrastructure, but they asked force majeure. They said, because there's been a slowdown community-based, folks said, "Hey, when you start pumping all -- when you light up this data center, right, we pump the brakes. We want to actually have a say in how you're going to use our electricity. And you can think about like that has a knock-on effect to Oracle, which is highly, highly levered in its AI bed. It has a knock-on effect to Blue Owl, which is financing this actual data center. It has a knock-on effect to the actual data center builder and all of the infrastructure that's being built in connection with that suddenly is in question. And so each of these decisions -- and again, this is one data center of thousands that is meeting resistance and is slowing down, which could underscore, could jeopardize hundreds of billions of dollars. These are not just 1 or 2 people in the community with a sign walking around saying data center is out. This is broad-based concern about what is happening in my neighborhood and what are you going to do to prevent this AI from really messing up not only my community, but also my job perhaps.

Joe Mantone   00:19:36

So what do you think the outlook is for the OpenAI and Anthropic.

Brenon Daly   00:19:43

I would say specifically in terms of their IPOs, I'll say Anthropic will go out in the fourth quarter here, fourth quarter '26. And I think it will be qualified as a generally well-received offering. I think it might out space 6 months long because I think for Anthropic, there is a defensible business model and there's incredible institutional support. Like people need -- they don't just want, they need Anthropic Claude, the maker of Claud, they need that to succeed because so many other businesses are using Anthropic as kind of their default model. And so I think if it goes down, if it fails, it's going to be really bad. But I don't think it will. I think with the problem offering is OpenAI. -- which is somewhat ironic because when OpenAI came along, we thought, oh, this is NetScape. This is the NetScape for the AI generation just as NS was -- what NetScape was for the Internet era, OpenAI is going to be for the AI era was the -- and the technology, they didn't meet it with a business model. They had the opportunity, but they didn't have the right business model. And Propic, which is more than a few people, including Dario, who used to work in OpenAI, they saw that opportunity and they matched a really well-established business model. to it and said, okay, we're going to execute against this business model, which is enterprise focused because enterprise has cash, it's dependable, it's predictable. It already knows how to write checks for infrastructure, right, technology infrastructure, software, hardware or whatever. So we're just going to be part of the CIO budget, and they've had success. I can't argue with it. They're more valuable now than OpenAI. I think when OpenAI, I worry for OpenAI's offering, I do. In best case, it comes in the first half of next year, maybe Q1. But I think of those 2 offerings will struggle. It would not surprise me to see OpenAI have a down round as an IPO. I'm not Yes, I'm not sure about that, but I certainly far more likely than Anthropic.

Joe Mantone   00:22:10

Sure. You still see Anthropic happening this year, be a good boost for the IPO market, which in general has been improving, but it's still not quite what it was.

Brenon Daly   00:22:21

When we look for signs that the IPO window is open, right? Anthropic is an N of 2 companies, right? In terms of model providers, really, it's OpenAI and Anthropic. Yes, you have other -- you have obviously, SpaceX used that to come public as well with their model and you'll have other models. But in terms of like major frontier models, that's it. So it doesn't really -- in my mind, it doesn't really -- I wouldn't consider Anthropic and a bellwether in the way that an enterprise software company, a big name like a Databricks came public. right? That would be a different signal than Anthropic coming public.

Joe Mantone   00:23:08

You don't see it being a catalyst for other deals.

Brenon Daly   00:23:11

I don't, not in terms of offerings now, also because the sector is so challenged on its financials. If you're Anthropic and you have brand name recognition like they do, you can convince investors. But other ones, the drop-off to the next 2 or 3 is very sharp. And I don't think they would have the same level of conviction to invest in you.

Joe Mantone   00:23:38

Okay. How about M&A? I mentioned earlier that a lot of these hyperscalers are choosing to build, not buy according to you, your research. Do you see that changing anytime? Do you see any of these firms getting into the M&A game?

Brenon Daly   00:23:54

Not really. And again, from our perspective, look back over the past quarter century of M&A, which is my day job here at S&P, -- pretty much every other era has been worked by acquisitions, by steady dose of acquisitions to get companies into the next era or to get them into the next adjacent market that to get them into the next revenue bucket or however they think about their strategy, M&A was always central, right? It was buy. And we contrast that. And you've mentioned that hasn't really been the case in the AI era. Yes, Cursor, $60 billion to SpaceX. But again, that was -- that's such a unique deal. It only happened because the company went public and had the currency. It was an all-stock deal. I guess. You don't have too many of those that have a disposable income of $60 billion of equity just to hand over. So that isn't -- I would argue, isn't a particularly representative transaction for the AI Heroes. But when we think about the commitments, the financial commitments, we talked about earlier, we opened the discussion with, hey, they're spending a lot of money on data centers. Every dollar that they're spending on data centers is a dollar that isn't going to M&A. It's zero sum, right? It is a finite resource. And in terms of debt capacity, Amazon, Microsoft, Google, they're fine. They could take on multitudes of millions of dollars in debt and be fine. Oracle, no. But again, these are companies that used to do a deal a month. like Microsoft would knock out 10 or 15 deals a year in and year out. They couldn't do that now, right? Just because financially, right, they've tapped out a lot of the debt market. They're not going to really want to do equity offerings too much more and their free cash flow is all -- was like screugecDuck. You had gold coins flying everywhere and you just grab a shovel and then throw it over to some venture firm and they gave you their best portfolio company. like that was a pretty slick model. Everybody wins. Now when you open up the door to the treasury and you look in there and they attempted. And suddenly, yes, even if you wanted to do a large deal and Microsoft and Amazon and Google, they did big deals. Google bought Wez for $30 billion, right? Microsoft bought Activision Blizzard for $70 billion.

Joe Mantone   00:26:35

Yes, it's an interesting observation because in other sectors, we have been seeing some really big ticket deals, but not seeing in the tech space.

Brenon Daly   00:26:41

Exactly because they're otherwise obligated. And even if they wanted something large, even if Oracle wanted -- Oracle bought PeopleSoft NetSuite, they bought Cerner in the downturn in the pandemic, right, $30 billion health care deal. So even if they wanted to do something like that, there's no lender that would touch Oracle at this point to underwrite a $10 billion, $20 billion acquisition. They couldn't find the money.

Joe Mantone   00:27:10

Are we seeing smaller AI-related M&A transactions?

Brenon Daly   00:27:14

So when we look at AI, there was a period where sort of we fell back into the acquihire phase, and that was just more a reflection of the AI era is so specialist the data scientists are so hard to find. And it's -- so it just made sense instead of -- so you had Microsoft and you had NVIDIA as well, doing almost like tens of billions of dollars worth of acquihires. You also had the advantage of if you do an asset deal, it's less likely to get any regulatory scrutiny. You can just hire a few people, they bring over their bras, you put them down and you say, what you're doing over there, but do it on Copilot or do it on claw or do it on our Blackwell chips or whatever you needed smarts into, you just went out and effectively acquired a couple of really smart people. And so we saw a few of those early on, particularly again when the move was a little bit more mania, a little bit more optimism around AI. Now we're seeing really like there's not a lot of AI deals. And that's true not only for, again, the hyperscalers, but also more broadly. We did see Salesforce get back into a few AI deals. Salesforce again, a key acquirer in the market. It's the largest SaaS company on the planet. That didn't happen and somewhat sanity returned because that's just not the way technology works. Like we still have trillions of dollars a day in technology flowing over mainframes. -- but they're not going anywhere. like every time you book a flight, it goes through a mainframe computer. And those -- you can dust those things off from the 50s, 60s, some of those old machines, they're still humming along. People don't change that fast. And I think that was an underappreciated aspect of technology is the adoption isn't anywhere -- broadly speaking, right, the mass adoption isn't anywhere as quick as the people who are out on the bleeding edge of technology and who are developing this stuff. And most users, they don't care. They don't even know like you give them a 2-year-old model and they would never ever know that they were using a deprecated model, it's fine.

Joe Mantone   00:29:41

So are we still seeing the hiring taking place?

Brenon Daly   00:29:45

Not so much. Right now, what we're seeing is, honestly, like when we look at Microsoft and when we look at Google and when we look at Amazon, they're trying to basically rationalize the offering. And the drive now isn't to accumulate assets or accumulate employees or data scientists or PhDs or anything like that. It's to actually make what they have now work commercially. They've looked at it and they've said, Microsoft is a $200 billion revenue company. And they've pivoted around on Copilot 10, 12, 20 different ways. They still don't know how to price it. Are we doing consumption-based? Are we doing just license, all you can need, like how do we do this. So it's now in that phase where we're rationalizing the offering with an eye towards commercialization. And I think we're seeing that in a lot of other vendors as well, including Salesforce, including ServiceNow, including Workday is they're now beginning to say, okay, AI is part of our offering. and the very expensive part of our offering. We need to make some money on this because we can't continue to subsidize the unlimited use of this high-cost offering. So that's the focus now. And I think a lot of the -- both the product development and in the commercialization within those large hyperscalers, that's what they're focused on, not necessarily optimizing the offering itself, but optimizing the commercial offering.

Joe Mantone   00:31:33

I'm sure once more companies figure that out, then we'll maybe see the deal flow pick up.

Brenon Daly   00:31:38

Because again, right now, everything is speculatively valued. There isn't really enough revenue to associate with any of these companies to justify the prices. And -- but markets, all markets settle at some point. And then I think it will go beyond just the few deals that we have seen in the AI area from software companies, non-AI native software companies buying into the space. They've been huge multiple deals, 50, 100x trailing sales for these tiny little vendors. There's a big world of acquirers out there who want to get some of this technology, but they simply can't afford it. So they're waiting for the market to settle. They're going to come in and take out a bunch of really good technology companies that just AI native technology companies develop some slick, maybe it's a vertical application for farming or maybe it's a supply chain management agent that is super slick and works in multicurrencies and can real time navigate through 6 different languages or whatever, that just didn't make it commercially, really valuable technology on the right platform. than that. So the larger established vendors are just going to come in and say, cool, thank you for doing the R&D for us. Here's a check, but that's not a 2026 story. That's maybe 2027, probably more likely 2028, honestly.

Joe Mantone   00:33:01

Okay. Really interesting perspectives here. I think this is a good spot to end on. Brenon, thank you very much for taking all the time with us here today.

Brenon Daly   00:33:10

Glad to be back with you, Joe.

Joe Mantone   00:33:14

And thank you, everyone, for listening.