Digging into the AI Cycle with CFRA Research

01:43:12.747 — 01:43:37.547 · Jeff Buchbinder

Hello, everyone, and welcome to LPL Market Signals. Jeff Buchbinder here, your host for this week with a special guest. I am pleased to be joined by Angelo Zino from CFRA Research. He is the head of technology research over at CFRA. So Angelo, thanks for joining me today. Looking forward to talking tech with you.

01:43:38.907 — 01:43:41.667 · Angelo Zino

Great. Thanks for having me Jeff. And hi everybody.

01:43:42.947 — 01:44:58.867 · Jeff Buchbinder

Yeah, I think CFRA is a really good fit for our conversation today. CFRA is an independent research shop, and certainly LPL prides itself on its independence. So love to get the straight talk from firms like CFRA. Very popular partner. Many of you know them. In fact, the most popular partner for LPL advisors to get fundamental equity research to supplement the equity research work that LPL does.

So really, really appreciate the long term relationship. In fact, I'm pretty sure that CFRA was a trusted partner of LPL before I started at the firm, and I have been with LPL for 23 years, so we go way back. So thanks, Angelo, for agreeing to do this. Let's get into the conversation and let's talk tech.

I've got a few questions here, for you. So, I think we have to start with the hyperscalers coming off of earnings season. So my question for you on that is, is will the AI spend pay off and what gives you the confidence, that it will. If that's your answer that it will.

01:45:01.027 — 01:49:51.837 · Angelo Zino

Yeah. So you know Jeff and that's a good question. And you know, I think the hyperscalers are probably a good place to start off, really. As they're probably the best indicator of whether or not AI spent will and is paying off. And, you know, clearly they are the biggest CapEx spenders on the planet in terms of this AI spend at this point in time.

So, in our view, I think the way that it's the best answer to this is, that AI spend will eventually pay off. It'll be dependent on the demand environment in terms of, you know, how long some of this, you know, aggressive spend takes to really pay off, but in the same, you know, aspect of all this.

But I would also say is these hyperscalers are realizing, significant revenue at this point in time, it is translating to, you know, really high cash flow from operations. And ultimately it's allowing them to continue to spend at this aggressive pace, which we think will be sustained here for at least the next 2 to 3 years.

Now when we think about, you know, kind of how this spend is really taking place at this point in time, there's probably about kind of a 60, 65% split between what I would call your servers and everything that goes into the servers. So that would be your chips, your networking, you know, the memory, all that kind of stuff.

And then the other, you know, 35, 40% right now goes into the data center side of things. Now, when you start thinking in terms of building the data centers, right, the concrete and everything that's needed on that side of things. And, you know, I define that aspect of it, that 35, 40% in terms of sunk costs and high upfront costs initially.

But ultimately, the life of those datacenters are significant in nature. In fact, Microsoft just extended, I think, from 15 to 25 years. You know, companies like Amazon even have longer times on them, call it 30 years or so. So, you know, you're investing heavily on that side right now. That's not necessarily going to take up 35 to 40% of the pie here, you know, three, five, ten years down the road.

So overall, you know, over time you're going to kind of see this bend and you're going to see, you know, just a real good inflection in terms of, higher free cash flow over time. But for the time being, it makes all the sense in the world to invest heavily on the datacenters and I'd say, you know, more importantly, on the server side of things, which is, you know, more near cost, near-term cost oriented, and that's where you want to spend more aggressively because servers equal revenue in many respects.

So, but ultimately, when we think about all those early days in terms of the drivers as we inflect to more of an AI inferencing world. So if you think what's going to drive us, you know, into the AI inferencing world, there's three kind of key drivers. One, it's the number of users. And I'd say that's very early days at this point in time, you've only got about a billion users on platforms like Gemini, on ChatGPT and what have you. We think over the next couple of years you're going to be talking about potentially 5 billion users or so on these platforms. The frequency per use also extremely important. That's the second driver.

So as you kind of get more of these users, even the existing users, plus these new users are going to use these platforms a lot more frequently in nature. Not to mention as we kind of, you know, push more to an enterprise driven world. If you look over the last couple of years, most of the demand for AI has been more consumer driven, has been adopted by the consumer, not as much so by the enterprise space.

What's happening in 26 is definitely more enterprise driven in terms of the growth. And then the third driver is the compute per use. So again, you know, AI agents take up as much as 100 x more compute capacity than some of the more traditional use cases like gen AI, you know, for instance, or even some of these basic reasoning AI models that we saw just 12, 18 months ago.

So, all those factors are going to be huge, contributors in terms of sustaining this demand and actually, increasing it. And then, you know, more. And then when you, when you kind of look at the cloud growth numbers that we just saw, what the street wanted to see was good cloud figures for Q2.

We saw it across the board in terms of, you know, Google, growing north of 80% year over year. You saw Microsoft at 43%, Street was looking at 40%. And then, of course, you saw a really nice acceleration out of Amazon in terms of the getting up to the mid 30s. And when you look at the guidance here for the, for Q3, these companies are also pointing to an acceleration.

So, all that new capacity that's coming on is getting completely absorbed by the enterprise space. And that is, you know, really a great indication that that AI spend at this point in time, it's starting to pay off.

01:49:53.837 — 01:50:22.437 · Jeff Buchbinder

Yeah, great points, Angelo. I've certainly been impressed with the cloud growth. Some of these companies are able to justify these big spending numbers pretty quickly. I know you've been looking at consensus expectations for just how big the spend will be. I just saw updated numbers this morning, actually, over $1 trillion for the next four years.

Is that possible? Or do you think those expectations are too high?

01:50:23.597 — 01:53:09.757 · Angelo Zino

Yeah. I mean, when we actually think about the actual CapEx spend here and we have, you know, we actually look outside of just kind of the big four, we look at the big four if you include Meta, you know, along with the three big cloud providers on top of some of these tier two players, the Oracle's, the Neo Clouds of the world, right?

Some of these China providers as well. Our view is essentially CapEx spend this year will double year over year versus last year, which is, you know, absolutely phenomenal. But that's an indication of all these companies essentially blowing out all their free cash flow to invest aggressively in CapEx. As we go into 2027, we are looking for a deceleration in terms of the growth. We're looking at about 25 to 30% growth. That does get you, you know, about to, you know, let's call it a $1.7 trillion CapEx environment next year in terms of all the CapEx spend, if you're looking at just the four big ones, they'll get to $1 trillion on their own.

They'll represent about, you know, 60, 65% of the total spend out there. And we do think there's a realistic probability as you kind of exit the decade we're going towards, you know, let's call it a $3 trillion annual spend environment, which is actually something that Jensen Huang had predicted, you know, 2 or 3 years ago.

And it seems farfetched in nature at that time, it no longer seems that far fetched if you kind of start thinking about some of the drivers, it is going to be more of a timing thing in terms of whether or not some of these demand trajectory, you know, some of the demand indicators really play itself out.

I'd say the biggest inflection that we do need to see is the shift to more of a physical AI world. Right? It's, you know, the more do we see more of these factories? Do we see, you know, some of these automotive companies out there really inflected, start going fully autonomous in nature inside their factories and start building these AI factories, which then kind of, you know, required you to continue to need more and more of these datacenters.

So that's going to be a big indicator and inflection in all of this. But at this point in time, yeah, I mean, I think there's no reason to think that we're not going to continue to see more growth in terms of the CapEx spend here over the next 2 to 3 years. But it's going to decelerate in nature. And I would also say expect some potential casualties out there.

And I think we have to be very mindful in what the, you know, the financing environment looks like, right? Will companies be able to continue the equity markets and also debt markets, and I'm not necessarily worried about those hyperscalers, where I'm a little bit more worried about clearly are the, you know, those tier two players, the Oracles, some of the Neo Clouds out there, which could see some bumps along the way here.

01:53:11.797 — 01:53:53.007 · Jeff Buchbinder

Yeah, sure. There'll be winners and losers as there are in any technology revolution, and this is certainly one of the biggest ones we've seen in our lifetimes. So thanks to that, Angelo, I know a lot of our listeners are just wondering, well, should we still hold tech, right? Or should we potentially be overweight tech?

And so with that question, I really want to turn to valuations because it's clear that the growth is there. I know you highlighted one of the risks at least, which is, you know, will the equity and credit markets help these companies fund these build outs which are just massive. But how about valuations?

I mean do you see, you know, whether it's semi software,

01:53:54.087 — 01:54:05.327 · Jeff Buchbinder

networking equipment, whatever it is. Do you see the valuations as reasonable enough that they are providing investors with a little bit of a cushion if some of this spending ends up being wasteful.

01:54:06.007 — 01:55:04.287 · Angelo Zino

Yeah. No. And that's a great question too. So, you know, the way I kind of look at this market today is, listen, nobody knows what things are going to look like in nine, 12 months, 24 months from now. Whether or not some of this stuff can be sustained. And the big reason for that is we have a good idea of what the supply growth is.

What could the supply growth would look like with no idea what the demand side is going to look like, because we're in unchartered territory. What I would say is what you want the market to do is to continue to act in a rational way, which we think they are. You look at when valuations peaked last year.

It was October of last year, on a forward basis. You're looking at the S&P 500 tech sector trading about 32,33 times. It was essentially at a 20 year high kind of double topped in many respects from what you saw in June of 2024. So that was kind of that's kind of your limit, your upward band, let's call it in terms of valuations.

If you look at it today on a forward basis,

01:55:05.367 — 02:00:17.617 · Angelo Zino

trading closer to 23 times on a forward basis. But I think more importantly, you know we're analysts, so we typically look ahead. We're looking at valuations already on the 2028 basis. Because as you get towards the end of 26 and early 27, the street is very forward looking in terms of valuation.

So that's 16 multiple or so in terms of the tech sector is extremely enticing if you just look at it on a valuation perspective. Now, I think it's important to kind of look at how the sector is constituted at this point in time. Overall, we continue to have a positive fundamental outlook for the tech sector.

I mean, again, it's because the earnings growth is there and the valuations, at least at this point in time, is discounting a lot of the concern and potential risks that are ahead. It's not probably completely dismissing it because things could get ugly really quick, but it is discounting some of it, which is very different than what we saw, let's call it in 1999 and 2000, where valuations kind of ran up along with fundamentals.

So this is actually doing the opposite. It's actually trying to find a peak and discounting that you know those peak earnings. When you kind of look you know sub overall tech sector we kind of break it down into three components. We look at it as semiconductors, then we look at hardware and then we look at software.

I would say on a semiconductor side of things, again we continue to have a positive outlook more because of, the earnings growth trajectory. And also when you look at valuations, look at the three biggest components within the tech sector of the semiconductor industry. It is Nvidia, it is Broadcom and it is Micron. Nvidia about 13 times our 28 estimate.

Broadcom about 15 times. And then you look at Micron about 4 to 5 times. Those are typically what you would see in what I would call a peak earnings environment. And if we can kind of continue to see a sustained higher spend environment, let's call it to 29 and 30, you're going to continue to see higher prices for the semiconductor industry.

As long as and it's probably not going to be dictated by earnings anymore, it's going to be dictated on the sustainability of this, you know, the spend. So, I think what investors are going to look more at is some of the derivative factors, right? It's going to be the AI monetization stories from the cloud companies as well as, you know, how the spending environment from the enterprise space overall.

So, that's what's going to dictate whether or not these, these chip stocks go higher in nature. But overall the earnings is there. And the valuations are, you know, at this point in time acting very rational in nature, I would say, as you kind of very similar on the hardware side of things, hardware is more concentrated with Apple, obviously.

But and we do we're big believers of the whole kind of AI at the edge phenomenon. But also when you kind of think about some of the other players out there, I'd say more along the lines of the Dell's and what have you, the server providers. Again, the risk reward looks, you know, fairly, it looks it looks okay.

I think on the software side, it's a bit mixed in nature. I'd say we're more positive on the infrastructure side, more neutral on the SaaS side. So the infrastructure side, think about cybersecurity. Again, think about some of those cloud providers with Microsoft in there. Think about, you know, the Palantir's of the world, those that area of the world.

I'd say there is a the data infrastructure providers. Right. So examples would include the likes of a Datadog and what have you. I think there's very kind of real, good visibility in terms of the long term trajectory of, you know, those trends here looking out over the next couple of years. The SaaS space is going to be, you know, it's there's going to be a lot more bumpy in nature.

You're not going to be able to prove. I'd say from a high level perspective that these SaaS providers won't necessarily see some of their business prospects erode over the next couple of years, because if we're moving towards a more consumption oriented type of world where enterprises are creating, you know, more of their own capabilities in house and kind of renting out this compute capacity, the money's got to come from somewhere.

It could potentially come on the SaaS side. It could potentially also mean lower headcount from, you know, across the enterprise space as well. So there, which also indirectly or will directly impact the fundamentals on the SaaS side. So, you know, I'd say you kind of look at different aspects of the tech sector and the risk reward is very different in nature.

But what we do like and why we continue to have a positive fundamental outlook overall for the sector is because unlike what we've seen in potentially other cycles where we've kind of gotten this irrational exuberance, that's not necessarily happening in this period of time. And a lot of the potential concerns out there as far as the AI build goes we do think is at least discounting a good amount of those risks.

02:00:19.216 — 02:01:18.177 · Jeff Buchbinder

Yeah. You kind of touched on a question I was going to ask you a little bit later about the AI disruption. I completely agree that, you know, you want the mission critical software, you want the software you can't really live without. You want to be sort of a layer below the SaaS layer or, you know, the application layer.

On that note, though, I want to ask you, I mean, how hard, you talk to, you know, IT managers. How hard is it to replace software with AI? I mean, because I've, I mean, I've heard all different things and some, you know, some tech companies are obviously ahead of the curve, right in sort of building their own or vibe coding their own applications.

But your typical company that's not tech, you know, how long is it going to take for these companies to figure out what they're going to be able to vibe code and then to actually embed that into their workflows?

02:01:18.537 — 02:05:21.937 · Angelo Zino

Yeah, I mean, I think it depends, right. One it depends on the industry. I think, you know, there are some financial institutions, you know, from what I've seen, that have moved a lot faster than others out there. Some of it may also be, you know, the how they kind of manage their budgets at this point in time.

But, you know, one that you kind of have to look at all these at this point in time when you think about just the overall SaaS environment and the different type of companies that are out there, you know, listen again, we're shifting away from the we're going towards this structured to unstructured world.

In many respects, I almost kind of think about, you know, other types of disruptions we've seen across the tech space over the last 20 years. I think kind of the whole, you know, the shift to this mobility world and, you know, the impact it had when Apple kind of did what they did back in 2007, creating the iPhone, but more importantly, what they did when they created the iPhone.

The impact from the App Store that they created and what that did to other companies out there, you know, to think about the research emotions or blackberries of the world and what have you, where people thought, you know, having a massive installed base was good enough. Where those, you know, where that company, you know, was going to be just fine.

And ultimately we realized, you know, how quick technology changes and how quick, you know, and how it erodes certain businesses that don't kind of keep up with the times. And I think in this environment, as we kind of again, move towards this, you know, structured to unstructured world, there are going to be certain type of businesses.

I mean, I think if you're a content creator out there where there are clearly being a ton of use cases that are being created, a ton of competitive offerings. Um, not to mention, you know, it's probably also easier to create some of those replicate some of that stuff in-house as well.

So, you know, the this isn't going to be I think on the software side, as in, the pie gets a whole lot bigger and everybody wins for a couple of years, like we're seeing on a semi site in some respects, it's going to be along the lines of, you know, there's going to be greater opportunities, especially as more compute capacity comes out there.

But the companies that are moving to a consumption driven world that have real good capabilities and can offer that manage the data and then kind of offer new use cases of value to that enterprise space will win. And then others that kind of keep chugging along and maybe, you know,  they don't, you know, manage the data out there.

It's going to be a little bit harder for them. But, you know, as far as, you know, budgets go are concerned. Listen, if you're if you don't have an offering out there that's going to increase the revenue of, you know, enterprise X by ten, 15%, that enterprise company is going to probably have to, you know, find ways to increase their IT budgets while you know more on the compute side and find a way to maybe adjust some of the spend in other things that they do.

It's probably going to come on the SaaS side of things potentially, again, come from, you know, rationalizing headcount as well. So, but that said, I think there are a good amount of winners, especially some of the bigger software companies out there. I think they're going to find some great use cases and be able to, you know, they're going to use AI to be able to empower themselves.

I would say it might be a little bit different on the services side of things. So think about some of the IT consulting companies, your IBM's, your Accentures of the world, where it might be a little bit of a little bit tougher for them to kind of start realizing the benefit, they may have kind of more of a direct attack in terms of, you know, the AI, you know, the AI impact or where the AI disruption can maybe potentially impact them a little bit more, I'd say relative to this, the SaaS side of things.

02:05:23.377 — 02:07:01.867 · Jeff Buchbinder

Yeah, that that makes sense, Angelo. It is, you know, such an uncertain environment in such a dynamic environment that there are some companies that are just kind of taking a step back, I think, in assessing what they can and cannot do and how they're going to deal with all this change. So, yeah, in some cases you're going to probably have big tech projects put on hold I would imagine maybe whereas in the past they would have gone forward. That's actually my old space. I covered IT services 100 years ago when I was on the sell side. So let me let me go back quickly to the valuation discussion, because it kind of relates to the question about the trillion dollars of spend.

And can we really get that annually for the next few years? You know, I like to value things based on future peaks and future troughs. And right now I just I don't know where the peak is right now. Maybe it's 2030 and I don't know how deep the trough is after we hit the peak, whenever it is. So how do you think about, I mean, this really this discussion of what's going to happen in 2030, really comes to light when you're talking about the memory names, right?

These things have just been moonshots this year before a recent collapse. They're still up nicely, but the moves have been tremendous as the market debates you know what 2030 looks like and what 2035 looks like. Okay, so anything you could share with our listeners to help think about how to how to value those types of cyclical businesses.

02:07:02.187 — 02:14:34.637 · Angelo Zino

Yeah. You know, it's memory has always been kind of the most frustrating area of the semiconductor supply chain, and rightfully so. It's because it's the most cyclical and it's got the most earnings volatility among all the different segments of the semi ecosystem. You know, when I think about memory I think about different phases of the memory cycle.

Right. So there's, you've got these boom, historically you've had these boom and bust periods when you hit a peak, you kind of go through some, you know, massive, you know, rationalization. You know, average selling prices start rolling over as average selling prices roll over. You kind of you start cutting your spending.

And then after that, what you'll do is you're kind of cut some of your capacity because you'll think you have too much of it. That actually happened in 21, 22 into 23, where a lot of these d-ram providers cut as much of as 30% of the industry capacity or took about 30% of industry capacity offline thinking, hey, we built too much during kind of the Covid environment.

And, you know, clearly they had gotten to a point where pricing had collapsed. We had these companies were actually unprofitable in nature, very common in prior memory cycles. And, you know, gross margins went negative because they had a write down. Um, you know, a ton of their investments.

But when you think about an up cycle. Right. Eventually you find that bottom, that stabilization. The first year is usually the best year, right? It's that recovery year. That's what as an investor, when you want to first be in it. Then you've got that profitability phase where once you're recovering, you don't necessarily want to aggressively expand capacity because you're getting more profitable.

Average selling prices are starting to come back because that supply and the market is getting filled and that's helping you. So you want to continue to get more profitable because you had lost so much during the downturn. Then the third phase of the up cycle is usually the investment phase.

It's when you significantly increase CapEx spend. And that's the phase you're probably at right now. And the question right now is how long does that investment phase take place? And do we get to an extreme point. What's happened in the past is because these cycles have been more consumer dominated with PCs and smartphones, let's call it, over the last 15 years or so, the boom bust cycles one have been the up cycles have not lasted that long. They lasts about 8 to 12 cycles and then the downturns will last, let's call it, you know, 2 to 4 quarters, maybe a little bit longer than that. And you just kind of rinse and repeat, right. This is a little bit different because of who the spenders are and what's driving it.

So, this investment phase right now is as long as, you know, it doesn't get out of hand in terms of, you know, the three big players that being SK, Samsung and Micron on the d-ram side don't get out of control in terms of the spend then we'll be okay. At this point in time, it doesn't look like they're going to get out of control because they're to an extent also capacity constrained in terms of the equipment that they can buy and put into these facilities.

Not to mention when you think about the memory that's actually getting built right now, it is high bandwidth memory. It soaks up 3 to 4 x, the type of capacity that you would use for traditional memory. So in traditional use cases on the consumer side. So that helps a lot in all of this. So you know this is one of those situations where most of these companies are sold out through 2027.

When you look at the purchase commitments of the hyperscalers, we're looking at close to $2 trillion from just, you know, four companies or so. Yeah, it's pretty insane. So and that number just continues to go up and a lot of it is going towards memory. I think what's also important that people need to understand is when these companies came out earlier this year and they blasted out their free cash flow and they put it on the CapEx side, every dollar that goes that went from that free cash flow, that was going to be used for buybacks now gets deployed into the supply chain and that has a huge multiplier effect.

Right. So if Microsoft is now buying a server, let's call it, you know, a good chunk of that or most of that might go to Nvidia. But Nvidia's also then you know they've got it. Then there's a multiplier effect below Nvidia right. Some of that is now going to your memory guys. It's going to your networking guys, your CPU vendors or what have you.

Then when you know, Micron gets that, you know, whatever portion of that dollar, they've got to spend it on that CapEx dollars. So there's a huge multiplier effect. And that's why, you know, the earnings trajectory across the tech ecosystem has absolutely shined this year, it's because of that multiplier effect that we're seeing.

And a lot of that multiplier effect is going to go, you know, continue to take place into 2027, especially as it trickles down across the lower end of the supply chain. So, it's really important. But, you know, as we look at these cycles and the valuations, at least for memory right now, again, you're trading close to peak multiples. In prior peak cycles, for the industry, you were looking at about 4 or 5 times, which is where Micron is at on a forward basis at this point in time. Mid-cycle for us would be closer to 15 times. So if you think this peaks out at, let's call it, you know, a company could you know, see earnings like you know Micron sees it closer to 200 or so and maybe you know mid cycle.

Is it theoretical where you kind of get closer to 100 or something like that. I mean we don't know what that looks like at this point in time. But what I do know is, one, the free cash flow over the next 3 to 5 years is going to be absolutely insane for these companies. Two, you know, I do think in the trough cycles for these companies, I don't think we're going to be talking about, you know, losses anymore.

I don't think we're going to talk about the need to raise capital or, you know, whether it be through debt or equity for these companies like they've had to do in the past. So, these companies are much better financially positioned to go through boom and bust cycles. And I think, you know, also on top of all of that is, I think, the margin trajectory of these companies, we are going to be significantly higher because of the type of chips that they're now producing.

It's very different. It requires it's not just putting up four walls and pumping out, you know, some basic memory chips. It is a lot more sophisticated and tougher to build this stuff. So it's not like the Chinese are going to come out tomorrow and be able to pump out high bandwidth memory 4 or 5, you know, shortly they're going to continue to be 3 to 4 years behind us in terms of that side of things.

So, from that trajectory, I think there's going to be a price discovery phase at some point in time. It may not happen until you get to a downturn, unfortunately. So you may have to sit with these type of valuations. And the upside may be limited in terms of memory. But in the same trajectory you could potentially see a price discovery moment over the next 18 to 24 months where people realize, hey, the free cash flow potential of these companies, you know, during cycle up cycles as well as potentially down cycles, might look much better than, you know, the market has given it credit for.

So I know it's a long winded answer, but overall, you know, I think you've got to be pretty optimistic given where the valuations are for memory and the fact that they're already discounting, I think in many respects peak value earnings.

02:14:35.437 — 02:15:58.957 · Jeff Buchbinder

Yeah. If these things end up being less cyclical than they've been historically then you have the potential here for higher earnings to move these stock prices quite a bit higher. That's not necessarily something you can have conviction in now. But there's certainly an opportunity there no doubt.

So, I agree with your sentiments on that on that space, Angelo. Thanks so much for that. Tough questions. You really have to look into your crystal ball. But I appreciate a great job with those answers. Really good perspective from somebody who covers tech all day, every day. So I think we'll wrap it up there.

But and I also know some of you who are listening are probably at the LPL Focus conference. So for those of you who are out there, hopefully you are having a great conference in San Diego. It's Monday, right around lunchtime here in Boston as we're recording this on August 10. So, I also want to mention that CFRA is available to all LPL advisors by subscription.

So if you're interested in getting all of Angelo's work and the work of his many colleagues, you can do this for help if you need it. And finding out how to subscribe to CFRA. So thank you again, Angelo. Appreciate your time. Thanks, everybody, as always, for listening to another edition of LPL Market Signals. We will be back with you next week. Take care.