Can AI Avoid a Dot-Com Déjà Vu?

Is today’s AI boom echoing the excesses of the dot-com era, or is this cycle fundamentally different? Nasdaq’s Mark Marex joins Interactive Brokers to break down profitability, valuations, supply-demand dynamics, and what investors should watch to determine whether AI’s rise is sustainable.

Summary – IBKR Podcasts Ep. 337

The following is a summary of a live audio recording and may contain errors in spelling or grammar. Although IBKR has edited for clarity no material changes have been made.

Jeff Praissman

Hi everyone. This is Jeff Praissman with the Interactive Brokers Podcast. It’s my pleasure to welcome to our podcast Mark Marex from Nasdaq. Hey Mark, how are you? 

Mark Marex

Hey, Jeff. 

Jeff Praissman

I am doing well, I’m doing well. It’s great to have you in our studio, and you just wrote a great article. I kind of wanted to dissect it and go through it for our listeners. So yeah—Mark, your analysis compares today’s AI boom to the late ’90s tech bubble, but it shows striking differences in profitability between the two eras. Could you elaborate on why nearly 100% of today’s NASDAQ 100 companies are profitable compared to the much weaker fundamentals during the dot-com bubble? 

Mark Marex

Yeah, absolutely. So, I mean, just even before we start off, just as a bit of background—my team is responsible for covering really all the indexes that Nasdaq has out in the market across the U.S. and the broader Americas region. And, you know, the NASDAQ 100, as our flagship, is something we cover weekly. We put out at least one note a week for our weekly newsletter that really seeks to be the leading source of research and analysis for the benchmark on the Street. 

So whether it’s covering what’s going on during earnings season, recapping performance drivers—fundamentals, technicals, thematic, macro drivers—you name it, we’re trying to cover it, right? And so every quarter we have an opportunity as a team to sit down and look at what is the one major thematic driver contributing to index performance and evolution. And as won’t surprise most people, that’s been AI for a number of quarters now in a row. This topic really picked up in intensity toward the end of 3Q into 4Q this year, with people looking at a lot of the news that’s been coming out—some of the deal announcements between OpenAI and others—and starting to get a little bit skeptical. Skeptical about how sustainable all this is, starting to poke holes in some of the business cases and valuation stories. 

We wanted to look at it in as objective a fashion as possible and compare it to this mental anchor that a lot of investors still have in the late ’90s. If I had to summarize the difference in one word—probably, at least at the company level for the index—it’s maturity. 

In the late ’90s, there were a ton of companies—thousands of them—that IPO’d on Nasdaq that were part of, if not the dot-com wave itself, internet-centric companies. Then, in the more traditional hardware, software, and telecom spaces, a lot of these companies were brand-new companies in the index. And we point that out upfront. 

When we look at the distribution of companies that were profitable versus unprofitable then and now, it’s really a very different story. You had over 20 companies back then—a fifth of the index—not even earning a profit in 1999 for the full year, which, when you think about it, was as good as it got during the tech bubble. That’s when fundamentals were the strongest, when everyone was bullish, when all the exponential curves around adoption of the internet were peaking. 

And it’s interesting to me looking back on it. At the time of the strongest fundamentals, one out of every five companies was still unprofitable. Today, it’s one company in 2025 that’s on track to not turn a profit in this index. 

Jeff Praissman

And that actually is an interesting point you make, and it leads me nicely into my next question. Looking at the current P/E ratios—not bad, low 30s—versus the low 100s during the dot-com bubble, what other factors do you believe are keeping these valuations more restrained during this current AI wave, despite the tremendous market enthusiasm? 

Mark Marex

Yeah, so a lot of it has been fundamental growth. We’ve now seen 10 consecutive quarters of NASDAQ 100 top-line, index-weighted EPS growth at 15% or better. Earlier this year, in 2Q, we were over 30% year-over-year EPS growth, which is just incredible. When you think about the fact that this is an index whose aggregate market cap exceeds $30 trillion, it’s got the eight largest companies in the world—the Magnificent Seven plus Broadcom, which is one of the major players in this AI infrastructure buildout that we highlight in the piece. 

It really speaks to the difference in the fundamental nature of—if you’re going to call this a bubble today, and you can in some ways—you can call it a bubble in certain parts of the industry, especially on the private market side and even some publicly traded players that are not part of the index. Maybe we’ll get into that a bit later. 

You can find some bubble-like behavior in this ecosystem today, and I don’t think that’s surprising given how exciting the technology is and how much investor interest there is. But what you’re starting out with in the NASDAQ 100 is an index of extremely successful, mature, fundamentally sound companies—some of the most profitable companies in history. You’re talking about companies that, in some cases, earn $100 billion of net income over the course of a calendar year. The fact that you still see EPS growth compounding quarter after quarter, despite all the investments they’re making into AI—many of them spending a ton of money on R&D and CapEx, ramping that up year after year—really speaks to the underlying dynamics that are different about this bubble. 

It’s about the nature of the technology itself and how that investment is translating into profits and cash flow growth, as well as the very different way that the leading spenders and leading investors are approaching it. That was a very interesting theme to try to dig into in the piece. 

Jeff Praissman

Yeah, no. I like kind of, I don’t know watershed moment’s the right word, but kinda two, two main characters in these stories, right? Our current situation, the post ChatGPT market performance, which is, you know, up 115%. And you know, it’s less extreme though than if you, we go back in time, that post Netscape IPO where, you know, it was up 718%. So you kind of touched on a little bit, I think, where it’s, you know, more mature index and everything now, but you know. What would you say? Kinda looking at the investor sophistication, you know, versus the nineties as well, right? Like clearly people become more sophisticated and more informed throughout time and just being exposed to markets and, and trading. 

Mark Marex

Yeah, I think that, I think that’s a very interesting point, and, and we even talked about this as a group. Sort of while we were working on this research piece and, and, and, and wondered about how we could quantify that, and we ended up not going down that route. Because it’s sort of out of our, our general area of expertise. But I think you’re right. You know, I think in the late nineties, if, if people remember back to what it, what it was like back then, I mean, this was people who were on a computer, right, day trading for the first time in their lives. It was the first time you really saw this type of. Super active, you know, retail all in type of a trader looking to really make a quick buck, largely around that IPO scene, which was, you know, disproportionately centered on Nasdaq in the mid and late nineties. 

That’s where the vast majority of this activity took place. And, and I think it, you know, it’s not a stretch to say that not just the, the sophistication of investors back there was lower, but the, the availability and accessibility of information was much lower. I think, I think people were responding to. News of companies listing and seeing the, the first day pop in the share prices and getting really excited about, you know, how much money you could make just participating on those IPO days and shortly thereafter. And a lot of it was not based on fundamentally sound. You know, investment thesis, right? A lot of these companies were, if not pre profit, then in some cases even pre revenue. They were unproven business models. They were ideas that, you know, because of the nature of the capital markets back in the late nineties. Were still able to find a lot of capital, whether it was retail or whether it was institutional, to sort of bring them to market and at least get them trading and, and, and somewhat financed and funded for their business models. 

Very different situation today, where you have, you know, OpenAI is the biggest sort of private player in this play, in this space, that is. Seemingly going in that direction of an IPO, but they’re not gonna, IPO as a company with a, a market value of 10 or 20 or even 50 million dollars and trade like a penny stock, right? They’re gonna target an IPO valuation of around a trillion dollars, and Anthropic’s not far behind, several hundred billion dollars. And it, it speaks to the very different nature of what it means to. Come to market. 

I think in today’s investment landscape, the companies are staying private way, way longer, right? They have deeper pools of venture capital to sort of get them through those early growth stages and get them to a position where, okay, when they are ready to IPO, they actually can join the NASDAQ 100 on day one, as opposed to, you know, trading like a penny stock for months and then, and then seeing what happens. 

Jeff Praissman

Okay. Yeah, and you, and you you touched about these companies with real, you know, revenue. I mean that the Magnificent Seven, you know, plus Broadcom, they, they generated, you know, about 630 billion in net income, with 31% average, you know, margins, you know, versus if we go back again, back in time, just 27 billion, which would be about 52 billion, adjusted, with 14.6% margins for those top 10 companies in 1999. 

So, you know. Obviously this, this fundamental strength, you know, change between now and then, it’s gotta change the risk profile of today’s tech leaders. Correct. 

Mark Marex

Oh, undoubtedly, undoubtedly, right. There, there’s, there’s a, there’s a number of different ways you can look at it, and the, the thing that I found sort of be the sort of, to be the most interesting and, and, enjoyable aspect of unpacking a lot of this data, comparing the two periods, was this idea of a mirror, right? The, the index today sort of being a mirror image of itself in the late nineties, in the sense of, you know, if you start to break this down in terms of like how the PEs are distributed, right. The vast majority of companies back then were sort of in a range of 60 or above in terms of PE, and that’s how you get, you know, an index weighted overall PE of a hundred, 150, even close to 200 on, on some days, sort of right around the peak of the bubble. 

Whereas, you know, nowadays, like you said, trailing basis, low, maybe occasionally mid thirties is, is the peak we’ve seen in PE recent years, that, by the way, is below the peak valuations we saw sort of during the COVID mini bubble in 2021 prior to AI breaking on the scene, right. So it kind of gives you a bit of context in thinking of like, well, not only are we not anywhere near. The valuation extremes of the late nineties, but we’re not even exceeding what we saw in 2020 and 2021 when people got very excited about some of the growth prospects of a lot of these companies due to work from home and, you know, entertain from home and play from home and all these things. 

But. When you layer on, again, the 15, 20, 25, 30% growth rates in earnings that we’ve seen for several quarters in a row for a lot of these names, and they still are outperforming a lot of these very, very bullish estimates. For, for growth, right? That means your trail, your, your forward PE is actually quite a bit below 30. 

It’s sort of in the, in the mid to upper twenties, depending on what day you look at this. And even when you drill into those top 10 names, right. Like Microsoft, one of the few, one of the few names that was among the biggest in the index in the late nineties continues to be one of the biggest today, has a lot of staying power, you know, is known for things beyond just the internet and beyond just AI. Certainly Microsoft is one of the, you know, one of the few names that. Has been the biggest name consistently in the index. A lot of those other names, they were trading at above a PE of a hundred back in the late nineties, whether it’s Sun Microsystems, which is not really around anymore, right, got absorbed. 

Cisco was at a PE of north of one 50, you had Qualcomm close to 500, you had Yahoo at, you know, off the charts basically at like a PE of 2000. These were all names in the top 10 of the index back then. 

And when you look at it today, there’s one name. One name that has an extreme PE in my opinion, and that is Tesla, which is, you know, again, depending on the day, somewhere in the range of two to 300 and has sort of always for a long time now traded to its own tune in terms of, you know, what do people think is gonna happen with robo taxis and self driving and all these things, and what’s Elon gonna say that’s gonna make the stock, you know, pop or drop 10 or 20% at, at every, you know, quarterly earnings report. So, you know. Tesla, setting Tesla aside, everyone else is sort of trading, even Nvidia after the runup it’s had, within what you would consider to be a much, much more reasonable range of, let’s say, 30 to 40 in terms of, you know, a, a trailing or a forward PE, where given that underlying growth rate, right. You be willing to somewhat of a valuation premium on a company like that with the level of exposure it has to the biggest growth driver in the economy today, you should be to somewhat versus the rest. 

Jeff Praissman

Yeah, I mean, I’m starting to feel a little nostalgic when you’re breaking out Qualcomm and Yahoo. I was, I was a floor trader on the floor of the Philadelphia Exchange back then. So those, those names were all, you know, near and dear to me. But your, your analysis points to, you know, actually a key difference between now and then that, like, actually excess capacity is not the issue at all with this current AI cycle. It is actually the opposite. The, the cloud providers are, you know, potentially struggling to meet this demand. So, you know, how might the supply demand, you know, dynamic affect the sustainability of, of the current AI investment cycle? 

Mark Marex

Yeah, I mean, that, that is, that is the million dollar question, right? And that, that’s something that, you know, at least from our, our standpoint on my team, is really, really hard to forecast with any level of confidence in terms of whether, whether and when we actually hit a point in time where there starts to be some excess capacity. Where there start to be chips that are not being utilized 24 7, data centers that are not a hundred percent leased out, right? 

The fascinating thing to me when I read about—’cause I wasn’t trading or sort of like actively following the market in the late nineties—but I’ve, I’ve read a lot of these studies of, you know, what went wrong back then. And you look at some of the stories around, you know, companies like, forget about even MCI WorldCom for now, which was, which was a, a very sort of prominent story of fraud and, and, and misleading investors, but companies like Global Crossing, right? Who sort of, like, only business model was let’s lay as many miles of broadband and, and fiber as we can all around the world and connect the entire world to the internet. And, and sort of, you know, adoption and, and, and demand for that service will follow, right? Well, what actually happened in the early two thousands when the bubble popped, there was anywhere from 95 to 97 percent unused excess capacity of broadband, of fiber, right? That is a massive, massive overbuild of capacity that was based on, you know, to some extent, a misunderstanding of, of how quickly demand for internet services could ramp. 

Like, we didn’t have things like Netflix back then. Netflix obviously uses a ton more data than, you know, a news site with just text on it. And even news sites today have video and ads and tons of other things that just didn’t exist when the internet was really being built out in, in terms of this infrastructure, right? So part of it was a, in my view, from what I’ve read, a misunderstanding of how long it would take for that demand to ramp and meet all the supply that was being created. And, and part of it was also, you know, there, there was this sort of gold rush mentality of a lot of companies in the telco space went all in on this, leveraged to the hilt, acquiring, you know, smaller players as much as they could, all thinking, you know, as long as we provide the most of the infrastructure that, that can enable widespread internet adoption, we’ll sort of, you know, recoup a lot of the, the benefit from that. 

And so look at it—what, what, look at what’s happening today. You, you don’t have dozens of companies in the hyperscaler space sort of all competing with each other to build out all of this CapEx. You really have five companies, right? In terms of the big four, NASDAQ 100—Microsoft, Amazon, and Google—sort of the, the, the dominant cloud providers, plus Meta, which doesn’t have its own sort of external facing cloud business, but is still a hyperscaler at this point given the, the, the amount of data center capacity they’re building. And then Oracle sort of in, in the, in the fifth slot, not part of the index anymore, not NASDAQ listed anymore since 2013. That’s it. And as you say, they are all, every single quarter, talking about how much bigger their backlog is growing, right? Their sort of contracted obligations, performance obligations to deliver capacity to OpenAI and, you know, whoever else is looking to run either training or AI inference within one of the data centers that they’re building and managing. So, like, it’s a totally different, at least at this point, situation in terms of the supply and demand underlying dynamic. And I wanna stress that it’s very highly unknown still. Is there going to be some technological efficiency breakthrough in a few years, maybe, that vastly reduces the amount of compute that we will need to have to train and run these models on an ongoing basis? That is possible. No one knows the answer to that, though. No one knows when that’s actually coming. So it’s purely a theoretical. 

And in the meantime, where we are is a very, very different space than where we were in the late nineties in terms of capacity being built out with the hopes of it’s gonna get all filled in the very short term, which never happened. It took decades for all that capacity to get filled. Not today. It is all at max capacity and can’t be built fast enough. 

Jeff Praissman

Right, and, and you mentioned the, you know, these major AI investors like Amazon and Microsoft and Google Meta. I mean, they’re already seeing, you know, direct return on investment, you know, with their infra infrastructure investments, you know, again, unlike back in the nineties. So, you know, you touched on before, but like, you know, does that, you know, vertical integration really change the investment thesis for these companies, for these AI? 

Mark Marex

I mean, to me personally, that is the most interesting part of this entire dynamic in this debate, and it’s one that is frankly not spoken about and debated enough, I think, in the, in the public arena, at least from what I’ve seen, right? Which is this notion of, again, if you’re a telco company in the late nineties and you’re on this physical CapEx—the internet fully built out and delivered across the U.S. and across the world—what is your personal, your individual benefit as a business, as a telco company, once that gets built out, right? Like, in theory, sure, you’re able to use the internet a little bit more and maybe you can do things a little bit more efficiently in terms of running your own business. It’s not, in my view, it’s not changing the nature of your business model. It is maybe tweaking things a little bit on the edges. But in terms of the ROI on that, the, the vast majority of the ROI is taking place outside the telco companies that are making these investments. 

If you look at all of the major hyperscalers, whether it’s Amazon, Microsoft, Google, Meta, they’ve all been laying people off the last few years. They’ve all done multiple rounds of layoffs, not because their businesses are struggling. Their businesses are growing at a record pace, whether it’s, you know, Amazon retail or the core search business at Google or, you know, the Office 365 Suite at Microsoft, with Copilot now being added onto it, Meta family of apps. 

Like, when you look at the underlying metrics of the strength of their business—subscribers, time spent, engagement, all, all these different things—they are firing on all cylinders. And what they’re telling us every quarter is they’re finding ways, and in many ways they are the leaders globally, in finding ways to deploy and get internal ROI on artificial intelligence technology. 

Whether it’s, you know, Amazon launching a, a shopping assistant called Rufus and saying they’ve already seen this year $10 billion incremental revenue growth just attributed to that one enhancement. Not to mention they’re using, you know, sort of their own AI model to optimize the flow of robots across their warehouses and reducing, you know, at least 10 percent travel time, which is a huge number when you think about the scale of Amazon, right? It’s the biggest retailer. That’s just one example with Amazon, right? Like Google has talked about this with their ads business, Meta as well, in terms of using AI to get better at predicting what different types of users on their platform are going to respond to in terms of the ads that they see, what their click through rates, what their purchase rates are gonna be. And so, like, they’re telling you there is an RROI on AI already. We are leading the way in terms of top line growth and in terms of bottom line growth in being able to get more efficient and cut, cut costs where they see the ability to do so. 

And so, to me, it’s a really powerful argument for the, for this, for the theme overall, that the biggest spenders, the biggest investments on the technology that are sort of driving it forward are demonstrating, right, that there is a lot of value to this. And in a way, they are helping pay for and fund a lot of this investment because they’re becoming more profitable in the meantime. That’s a very, very different dynamic from, I would say, you know, what you saw in the late nineties with a lot of the companies in the telco space that got in trouble over investing and, and, and sort of doing it for a very one dimensional reason, which was we want in on this gold, gold rush and we think, you know, we’re gonna make a lot of money ’cause there’s gonna be a lot of demand in the future. 

It’s, it’s a very, very different story, in my view. 

Jeff Praissman

Yeah, I mean, and that goes, you know, we already mentioned that there’s this, you know, significantly lower leverage among today’s tech giants, you know, compared to the nineties telecom companies. So, you know, how important is this balance sheet strength in determining, you know, if we were gonna be in a bubble, but also what, you know, what leverage levels would concern you if it starts to creep up? You know, at what point is there a concern that we might, you know. 

Mark Marex

I mean, it’s, yeah, it’s tough to say. I mean, I will sort of cheat a little bit on this and say if the big four approach levels that le, that Oracle is at—and again, emphasizing that Oracle’s not part of the index—they’re at a, they’re at a debt to cash ratio of about eight to one today, right? Eight times as much debt as they have as on their balance sheet versus their cash levels. 

You know, all the rest of the other four have net negative debt ratios, debt positions. They have more cash on their balance sheet than they have debt today. And as a percentage of market cap, it, it ranges from about one to two percent for each of the 4% of market cap—that’s actually debt—one to two percent. So when you look at how much leverage some really large companies in the U.S. are able to carry on their balance sheet and still be investment grade and still be profitable, I’m thinking of, you know, like the AT&Ts and the Verizons of the world, there is a ton, ton of excess leverage capacity that all these companies have. And to an extent, you’ve seen them start to tap that in recent quarters in terms of issuing, you know, bonds and doing some bank loans, and also, you know, taking advantage of some private capital and trying to finance some things off balance sheet and making guarantees there. 

And basically just using an all of the above approach, right? Figuring out, hey, we don’t actually need to pay for all of this with internally generated cash, given how cheap it is for them to borrow, given how small of a piece of the, of, of the market cap that they’re actually looking to finance versus, versus how much they’re worth today. So, totally different story, I think. Yeah, once you get to Oracle levels for the others in, in the, in the big four, maybe, maybe a bit of a cause for concern at that point. Yeah. 

Jeff Praissman

Yeah. And then looking, you know, at let’s say like Nvidia and Broadcom specifically, your article notes that they have margins, you know, roughly three times higher than Cisco did during a dotcom era. What, what structural advantages did these, you know, semiconductor companies have that just weren’t present for the network equipment makers, you know, in the late nineties like that, that has such this this such advantage right now? 

Mark Marex

Yeah, I mean, those, those are the two big names, and in particular, right, a lot of people have tried to analogize Nvidia as being the Cisco of today. And, and again, it’s sort of like when you look at the numbers, when you stare at them, you know, whether it’s valuation ratios or profitability ratios, you see very quickly that they’re, that they’re in very different spheres. 

I would say, you know, Cisco wasn’t a poor, fundamentally poor company from a fundamental standpoint. It was profitable. It had fantastic growth rates. But a lot of their growth, you know, it was tied to companies that were building out this infrastructure, right, the physical infrastructure for the internet, and making that fundamental miscalculation of, well, we see the demand catching up to all this excess supply that we’re building. So from a very sort of high level standpoint around business case, very different situation for Cisco and its peers back in the late nineties versus Nvidia and Broadcom today. The other major difference, though, being is that Nvidia, right, at least in terms of GPUs, no one has been able to catch them from a technological perspective, what they’re doing from a technological perspective in terms of designing chips and also supporting them with things like CUDA, which is their proprietary sort of coding platform that all the developers have been working on to build on top of GPUs for two decades now, right? 

That is an ecosystem and a level of technological advancement that is way, way harder, I think, in my opinion and many others’ opinion, to try to catch up to and replicate if you’re an AMD or a Chinese semiconductor company or whatever, versus a lot of the sort of like switchgear type stuff that Cisco and others were doing in the, in the late nineties was, was just not as, as technologically advanced, let’s say. 

It, it’s, it’s much, much harder to, you know, design and, and produce a leading edge GPU chip than it is to, to create something like that. Oh, and by the way, they have the perfect outsource partner in, in TSMC in terms of their lead of, for decades now, perfecting, right, becoming the go to source of actually manufacturing via foundry the chips that these companies are designing. So they’re very sort of, you know, R and D heavy in terms of expenditure given the, the, the cost and effort it takes to design all these chips. But they, they have an advantage, I would say, in the sense that, you know, a lot of that manufacturing cost, that CapEx cost, is outsourced to TSMC. And because of their position, you know, they found a way to maintain those really incredible margins, right, 70 percent or higher gross margins, because, because they are, you know, so far ahead of the rest of the pack and they, and they focus on the high value add work of design as opposed to manufacturing. 

Jeff Praissman

And you know what? One thing you hear in the news, and your article mentions this as well, is the skepticism around so-called circular financing and investment deals that are involving probably OpenAI and Nvidia. Could you expand for our listeners on what these potential concerns are and whether they really represent a potential warning sign for the broader AI ecosystem? 

Mark Marex

Yeah, I mean, that is the thing I think that started to get people nervous in the last three months or so, and you’ve seen a few of these sort of mini pullbacks in the market. I would say the most recent one, that started in early November, mid-November, was closely tied to this news of Google Gemini sort of leaping up the leaderboards in terms of model performance, leaping ahead of ChatGPT, which of course is OpenAI’s product. 

And so what you have is OpenAI obviously being, in many ways, the locus of a lot of this circular deal activity. It’s unique in history in the sense that its scale is so massive as a private company. They don’t quite have the same levers to pull in terms of raising money as a really big, public, established company like Nvidia or Google have. And so they’re in a tricky situation, right? Because for the last three years they’ve been the leader in this space in terms of adoption, broadly speaking, for consumers. You can argue maybe Claude, Anthropic’s model, is better adoption-wise for enterprise use cases, but in many ways ChatGPT is the flagship of this AI wave. And so what they’re trying to do is stay ahead technologically of Google, among others, as a private company, knowing that they need a ton more compute in terms of being able to train the models, but maybe even more importantly deliver the models to end users in a way where there’s low latency, you’re not waiting a long time for an answer to be given to you, and that answer is still somewhat high quality. 

And so they’ve been trying to solve this really complex equation for most of this year. Meanwhile, Google has limitless resources, pretty much, as one of the top three or four biggest companies in the world. They have Google Cloud, which is one of the top three cloud platforms in the world. They have many of the leading AI researchers, so from a model design research standpoint, as well as training and ongoing compute and inference, Google’s got the whole package. 

At this point they were a little bit slow out of the gate, but it’s very clear that now they’re sort of in a leadership position. And so OpenAI, I think, is feeling a lot of the pressure. And you will see, as these models continue to jockey with each other every couple of weeks or every couple of months when new versions get released, this tension of whether OpenAI will be able to get back to a steady leadership position and keep the foot on the pedal in terms of their expansion, or whether they’ll have to start relying on more of these sort of, call it circular whatever, creative financing deals. Where obviously if Nvidia takes a stake and they give them a bunch of money to help them keep iterating and keep expanding, it’s beneficial to Nvidia at the end of the day, because that means they’re going to have more demand for chips as the models get more sophisticated and more data centers get built out. 

So I think the way to distill it down is there’s a lot of cold economic logic as to why OpenAI is making these deals for their own survival and their own benefit, as well as why names like Nvidia want OpenAI to survive and keep competing with the likes of Google and Meta in the long term. We’ll just have to see how it plays out. I don’t know how much room there is globally for five versus ten versus maybe one or two leading models to be the ones that everyone uses at the end of the day. There were a ton of search engines out there, and Google sort of won that race at the end of the day, even though they weren’t the first and weren’t as big as Yahoo and others in the late nineties, but they ultimately won out with the best technology and the best distribution. 

Jeff Praissman

Yeah. So, Mark, in the interest of time, I’m going to skip questions ten and eleven, if that’s right, and go right to twelve, where we’ll edit that part—what, me just talking out. But if you had to identify one metric or indicator that investors should watch closely to gauge whether the AI boom is becoming unsustainable, which would it be and why? 

Mark Marex

That’s a great question. I’ve thought about that question a lot in recent weeks, and I think it may surprise folks to hear that maybe the P/E ratio is not the best one for this cycle, given just the amount of fundamental strength that you keep seeing in a lot of these leading companies, these leading AI investors, because they have so many different business models. I mean, you think about Google. It’s not just search and the advertising business. It’s YouTube, it’s Waymo, it’s Google devices—like all sorts of things layered on top of it. You may not see the warning signs from a valuation perspective for the NASDAQ 100, at least the way that you did in the late nineties. I think instead you have to look at other things like implied valuations of the big private-company players in this space—OpenAI and Anthropic and others like them. If they start seeing down rounds, I would say that’s a major warning signal. That’s a flag to pay attention to, the private-markets activity. 

As well as if you see the pace of data-center construction, announced investment, and planned construction start to slow down. If you see that year-over-year growth rate—in the U.S., for now—still ramping up, but if it starts to peter out and plateau a little bit, that’s going to impact names like Nvidia. 

Because that simply means their forward growth rate in selling things like GPUs is not going to sustain itself at the massively high levels it’s been at in the last few years. And so that might be a period of indigestion, so to speak, for the market, when the Nvidias and Broadcoms—and to a lesser extent the AMDs of the world—maybe have to adjust to fewer chips being sold than expected. But that doesn’t necessarily mean that the Googles, the Metas, the Microsofts, and the Amazons of the world, who are spending a lot of the money on the CapEx to build the data centers, are going to suffer. Because all of a sudden that means maybe there is enough capacity for them to take the gas off the pedal a bit and not have to spend quite as much on CapEx, which is eventually going to be supportive of earnings. 

And maybe the ROI question on AI will start to get more reasonable, and people will get more comfortable with the level of spend versus the level of benefit that different players in the ecosystem are getting. So that’s what I’m going to be looking out for in the next year or so. And hard to say, obviously, when that moment happens. It probably will happen at some point in the next few years. But for now, from our standpoint, it’s a very, very fundamentally solid thematic investment strategy that we’re tracking day in and day out and trying to stay ahead of. 

Jeff Praissman

Mark, this has been great. And for our listeners, you can find more from Mark Marex, the Head of Index Research of the Americas for Nasdaq, on Nasdaq.com. You can also go on our website, interactivebrokers.com. Go under Education to find more great articles and podcasts. Once again, thank you for stopping by the studio, Mark. Really appreciate it. 

Mark Marex

Thank you.