#100 - Rajiv Jain: AI Hype Echoes Past Tech Bubble Risks

0:05Welcome to the Insightful Investor Podcast, a weekly series that seeks to share industry, investment, investment, and market insights. We define insights as concepts that are counterintuitive, widely misunderstood, or underappreciated. In other words, unique ideas that you probably won't hear elsewhere. I'm Alex Shahidi, the host of the podcast and co-CIO of Evoke Advisors, a leading investment advisory firm. Learn more about our show at insightfulinvestor.org.

0:38Excited to have Rajiv Jain join us today. Rajiv is the founder, chairman, and CIO of GQG, which manages equities across the globe. The firm was launched about nine years ago and has already amassed$167 billion in assets as of the end of September 2025. Today, we have a special focus as we'll explore GQG's perspectives on the tech and AI-driven market, something very hot today, why Rajiv believes we may be at a significant inflection point and what investors can learn from the lessons of the dot-com era. Rajiv, thank you for joining us today. It's great to be here, Alex. Let's start with your high-level investment framework.

1:18Would you just walk us through your core principles? The core discipline revolves around buying what we call forward-looking quality at sensible prices. And I think the forward-looking quality is even more important today than ever before, simply because a lot of the names that we used to own, we don't believe are as high quality as the market perceives. But core businesses that we feel are in good shape and probably improve as time passes for kind of five-year horizon, but they have to be reasonably valued for the growth rate that they're delivering. And over the long run, what we also want to do is lose less in down markets.

1:55Obviously, we are long only, so we are fully invested most of the time. But losing less in down market matters for longer term compounding. So we're kind of conservative investors. As you know, Alex, I mean, we don't allow any personal stock trading at GQG. So a significant part of my personal network and Tim, who's our CEO, and everybody else is invest in the strategy. So we almost too much eat our cooking. and I think that also sometimes changes behavior sort of longer term compounding rather than trying to mimic the index. The big part of trying to minimize drawdowns is what you pay for a company, right?

2:34And how the future may unfold relative to that. Exactly. One thing that you're well known for is flexibility. You know, not sticking to a single style like growth or value, like you see a lot of managers. How do you balance flexibility with discipline? particularly when dominant market narratives take hold? It was always a good question, but I think it's even more relevant today because when people talk about discipline, buying growth at any price is in my opinion, rigidity, not discipline. If you can find cheaper growth, why wouldn't you buy cheaper growth? So our view is that at the end of the day, we need to compound our and our clients' money.

3:14And valuation is a critical component of that. So if a business, for example, is growing at, let's say, 15 % to 18%, selling at 20 times earnings versus a business which is selling at 26, 27 times earnings, but now growing at 10, 12%. The return profile would be very, very different. And by the way, this business essentially is Google or Alphabet. The return profile would be very different between the two. So our view is that we focus on sustainable growth, but at sensible prices. And that leads to sometimes, you know, areas which are actually fairly attractively valued. And you can say, oh, these are like value names.

3:52But I think most of the separation that is done by a vast majority of people tends to simply based on valuations. And high valuation does not always equate to growth, right? A lot of high, I mean, if you look at, for example, a company like Tesla, this will be the second year, back-to-back year of unit volume decline and probably revenue decline. But the multiples are very high. So the growth or well, is the growth stock or not? It's not really growing. It's declining as a matter of fact, right? So our view is that business have to grow. And if you can find cheaper, you know, a cheaper valuation, why wouldn't you buy a cheaper valuation?

4:28Because that ultimately dictates your long-term returns. So that leads to flexible and mime, you know, flexible sort of names. And as you know, we've owned from coal companies, energy companies, to Inmedia and everything else in between. One of the challenges is we can see the growth rates of the past. That's right in front of us. But you have to have some prediction of what the growth rates in the future will be and the sustainability of those past growth rates. And the high growth is really difficult to sustain over time. You're raising an important point, Alex, is that if a company dominates an industry, let's say more than half of the revenue or profits of an industry, if it's a major industry, there's almost no historical presence of them growing at sort of 20 % plus on a next five, seven-year basis.

5:12because if it's a mature industry and if it's, let's say, two-thirds of profits of the industry, it essentially means that if you assume they'll grow at 20%, you're basically assuming the industry growing at mid to high teens, right? That almost never happens because the industry dynamics for industry that have been around for decades is not going to start growing at, you know, 16, 17, 18 % while the past decade was 6, 7 % because you're a big part of the economy and there's only so many dollars to be had. And I think that math is an important aspect of what we're discussing here within software and semiconductor and so on and so forth, is that mathematically it's not possible for some of these companies to sustain the growth rate that some of the investors are assuming.

5:54And some of the models you see, they just can't do it because essentially what they're saying is that semiconductor industry itself will grow at 18%. Well, that mathematically won't happen. Yeah, and it is interesting when you have historical growth rates that are high, and you start with a smaller percentage of the industry, right, or the industry is smaller as a whole, and then it grows really rapidly, it's very common to extrapolate that historical past into the distant future. And what you're describing is there's some ceiling that prevents that from happening mathematically. Because the information technology is spent, I mean, there's only so much dollars to be had.

6:35that corporates are not going to say we're going to spend from 2 % to 5 % all of a sudden. It just doesn't work that way because something else has to be cut. And there's no evidence of that either, by the way. Because the other aspect is a productivity improvement because of technology. The question one has to ask is, what does the evidence suggest? What can the evidence on information technology's impact on profitability of, for example, large banks in the US? right? It's interesting. If you look at 30 years, and 30 years, by the way, talk about now launch of, you know, sort of introduction of emails and so on and so forth.

7:14Okay. So then email came around, internet came around, and cloud and iPhone and 50 other things. There's been no improvement. It's fascinating. So in 30 years, there's no improvement. All of a sudden AI will change things. And when we talk to a lot of investment banks, it's fascinating. I think vast majority are not seeing any signs of major improvement on profitability basis. So on what basis do you justify increasing the IT spend to 4 % or 5 %? It will not happen. And there are no signs of that happening either. But obviously, there's always a sort of spinning of the narrative, right? And narrative kind of becomes reality till it's not.

7:53And I think we feel that the narrative has taken hold that AI has got massive job losses, so on and so forth. When we talk to the companies, there's no evidence of that either. But the narrative takes hold, and you almost have to take a step back and say, does the math work here? That example of the 30 years in the banks and the profitability, why do you think that is? Is it because there's other factors at play that impact the final numbers? or is it something else that, because you would think that companies should be more productive, right? Technology has come a long way in 30 years. We are better doing certain jobs, but net-net, it doesn't flow through to the owners of the business.

8:32And that's a problem. So when you talk productivity, is the number of employees needed? Yeah, but if the compensation takes care of that from an owner perspective, it's a wash, right? And by the way, that's true of a vast majority of industries. and the margin improvement that we saw was not because of technology, because of outsourcing to Asia, which as we know is also coming to end in a lot of cases. The biggest product improvements came from outsourcing,

8:59tax strategies, shifting. If you look at MAC7, amount of taxes that they pay in the US is actually minuscule. They're paying between 5 % to 10 % max effective tax rates. So that improved profitability, right? It wasn't technology. and then lower interest rates. When you add these three up, that's almost between 70 % to 90 % of improvements in margins have come from these three reasons, not because of introduction technology, which is obviously very different from the narrative that you hear all the time. You've drawn parallels between today's AI-driven market and the late-stage dot-com bubble.

9:36What specific signs stand out to you in the current cycle? And do you see today's tech sector as perhaps even more vulnerable than 2000? First of all, it's way more vulnerable. And the reason is the numbers are much larger. The dependence of the economy on this area is actually startlingly high. I mean, let's start with market cap, et cetera, right? I mean, NVIDIA is almost 15 % of GDP. So it's approximately give and take Microsoft and Apple and so on and so forth. But their impact economy is actually far smaller historically in terms of number of jobs that they created and so on and so forth.

10:14But the market cap impact is quite significant, right? So technology itself as a sector is worth almost same as GDP, I am$30 trillion. The other aspect is the amount of dollars that are now going into data center, et cetera, again, as a percent of GDP and as a percent of overall CapEx spend in the whole economy is disproportionate. The problem is this is a very rapidly expiring technology. I'll give you an interesting data point. If you look at the GPUs, H200, which were launched last summer, if you go to the distributed network, which we have, then they're now selling between 40 to 50 % discount to what they sold 12 months ago.

11:00In fact, some of them are even suggesting even the Blackwell might be available for a discount. We haven't gone the numbers yet, but they are suggesting that you should be able to get that. And these are authorized distributors, by the way, of NVIDIA. The question is, something that I can depreciate that rapidly, why there's such a surge in investments when we do know that the Chinese open source models can be very effective? In fact, no other country is doing that. And there might be a good reason why other countries are not investing at this point, because technology changes are so rapid. So unlike the dot-com era, those optical fiber, et cetera, lasted decades.

11:37We're still using them. These may be a lot of stranded assets. Number three, the amounts that are going in are happening on the private side, and hence are more opaque. But some of the valuations are completely in Alice in Wonderland. For example, if you look at OpenAI at$500 billion, it's got a finger in the air, whatever they're valued at, for a company that last quarter, third quarter, apparently, based on Microsoft's numbers, had an$11.5 billion loss on approximately$4 billion revenue. And now they're saying they're going to spend$1.5 trillion. I mean, the math, what's fascinating, nobody's questioning the math.

12:19And this one company, right? If you look at X.ai, being valued at$200 billion, but under$1 billion revenue. Pets.com was$2 billion, right? People don't Pets.com. Oh, by the way, what's interesting is Pets.com was more than half owned by Amazon. And then it didn't work, right? But it was$2.5 billion. Inflation adjusted is not$200 billion. So the numbers are completely astonishing. The other part is the circularity. I mean, the whole venture financing issue, which was heavily criticized in 2001, but there are a lot of articles if you go through Wall Street Journal, et cetera. Now that thing is on steroids.

13:02I mean, today there was another deal announced of Microsoft NVIDIA investing in Anthropic. And Anthropic then is going to go around by compute from Microsoft. So they're going to borrow more money and buy back compute from these guys. Almost 40 % plus of startup capital is coming from tech companies. So the circularity of these, these are like maybe a dozen companies which show, you know, they come as usual suspects in everything. And the size and scale of this is actually massive. Because one thing which is very different, the other thing, by the way, is also different from what happened in.com.

13:44the underlying economic growth was actually stronger. People forget, U.S. had a fiscal surplus. In 2000, there was a New York Fed paper discussing what will happen if U.S. government has no debt. Think about it. This actually Fed research study, we had no deficit. In other words, there was a lot of fiscal flexibility in case there was a recession. And by the way, 2000, 2001 was one of the shallows recessions. So when you add all these things up, S &P growing a lot faster at almost 17 to 18 percent compounded annual growth rate going into the peak EPS growth. Now you're looking at high single digits, by the way.

14:28Right. And that's along with fiscal surplus, because that has obviously massive stimulative effect. So the question you have to ask is, if there's a sort of recession, is the deficit going to go to 10 percent? If you assume 300 to 400 base point increase in deficit, what happens in that case? How would the bond market react? Because in 2003, Fed basically cut the rates to zero. So as you can see, there were a lot of fiscal flexibility. The Fed had a lot of flexibility. The corporates had a lot of flexibility. This time around, it's a lot more tougher environment and the valuations are higher. So when you're doing your analysis, how much emphasis do you put on historical analogs like the dot-com era in shaping the way you view markets today?

15:12It's important to sort of look at the historical periods, but also see in which way it could be different. So you can't, you know, as they say, it rhymes, but it's never the same. And it always is slightly different. Otherwise, it'll be too easy, right? I remember in late 90s, there were a few gray hair who said, oh, they're similar to this with Nifty 50, and they were obviously dismissed. That, oh, this is not same at all. This is completely different. Those guys were idiots. We know better, right? And the funny thing is, it was exactly 25 years ago. Today, we are 25 years again. So there's something that we said about passing a full generation before people kind of forget the lessons, but there are differences.

15:54So there's some similarities, some differences. So I think it's important to learn from them, but you still have to sort of think through in which way this may be different, better or worse. One of the similarities is humans, which are hardwired to create narratives. They can get greedy. They can overpay and be overly optimistic. And then also on the other extreme, they can be overly pessimistic. E-Trade has now been replaced by Robinhood. E-Trade of that era, right? It's similar. It's different, but it's actually similar. Crypto may be far more harder. We say, well, there's not that much speculation.

16:27Well, yes and no, right? It just shifted its color and shape. If you look at how much Koreans are now investing in the US equity markets, how many Chinese and how many Thai and Taiwanese are investing in US markets, right, in terms of tech names, and all of them own tech, by the way. It's really gone global. If you look at what's doing the best, what is performing the best in Asia, is basically an extension of the AI trade here, right? So in a lot of ways, it has similarities. But again, the size and scale of this is huge. But to your point, one thing which doesn't change is human emotions. And we like to believe that this is going to be

17:12truly transformational. The problem is it could take a lot longer. Earlier, you described big tech as showing backward-looking quality rather than forward-looking quality. Would you talk about that a little bit more in the metrics or patterns that you're seeing? So let's talk about some of the names. Because as you know, these are kind of household names. We owned them before in a big way, very recently. But as you know, we tend to be valuation sensitive and our view is that the fundamentals are deteriorating. Why do we say that? So if you look at, for example, a company like Alphabet or Google, their core advertising business is now growing at low double digits.

17:49Last few years, by the way, like between 10 to 12%. And that's without economic slowdown. In economic slowdown, like in the first half of 2022, their revenue barely grew on the advertising side. It was low single digits. Okay, so this is cyclical. Same thing happened in 2020 during early parts of COVID. So you see they have cyclicality. Okay, the other big part of Google or Alphabet is cloud, Google cloud. Now, the Google Cloud issue is that what used to be a basically oligopoly between a few players, kind of three-player market. First, it was monopoly of AWS, part of Amazon, then Azure came along, part of Microsoft, and then Google came along.

18:35The problem with that is that now there are hundreds of players. So on Google Cloud, the competition has gone from basically Azure and AWS to Oracle, to Corvi, to Lambda, to Nebius. This at least 200 plus, you know, public cloud, vast majority on AI. So if you look at the incremental market share, it is shifting away from the legacy cloud to AI cloud. And there, the NeoCloud, as they call it, the new players, the Lambdas and the CoreVs, et cetera, they have more than a third of market share. So now you have two things working for them, of working against them. One is their new competition, because AI, it has kind of leveled the playing field.

19:19Okay? And the other thing is that they don't have the embedded legacy cost, so these guys can actually offer cheaper. In fact, I'll give you a couple of data points that we ourselves have asked in terms of find out what pricing can we get Okay? And if you look at pricing for GPU rentals for the blackmills, it is under$4. And we got written offers, by the way, under$4. If you look at what AWS is, is north of$10. I believe it's$12 last. Google Cloud is also the same thing,$10 plus. So the new startups, whether they pay$3.5 or$12, it makes or breaks their economics. So what is happening is that they have new players.

20:04Now, Google doesn't have to compete, but they're losing market share then, right? But why are they growing so rapidly? The reason they're growing rapidly is they're also investing all these startups like Anthropic and OpenAI and all these companies and they themselves buy compute. So the problem is, to summarize is, A, you have more competition, B, you have price wall growing, and C, it is much more capital intensive. See, historically, Google Cloud was basically a lot of hard drive type stuff, which had long shelf life. Now, these GPUs, basically every year, every 12 months, there's a new one.

20:37And the old one depreciates 40%, 50%. Second thing is these things run hot. I mean, that's why they need water cooling, right? Literally run hot. So the expiration of these things is going to be a lot faster. The funny thing is that companies have actually changed their depreciation policies. Think about it. Google's margin improved once they changed the depreciation policy. Before that, they were still making high single-legend margins. But for that, now you're paying almost 25, 26 times earnings. So think about this way. Slowing growth, multiples are the highest or second highest in last 12, 13 years.

21:12These are mature companies with slowing growth, not to mention capital intensity is through the roof. They don't have that much free cash to keep spending that much money. Hence, they begin to tap into bond markets. So now they're getting leverage too. early days, but you see in Oracle, right? Oracle bond yields have begun to widen. The CDS has begun to spike because people are nervous on Oracle's balance sheet. So it is leveraged. Some of these companies are getting more and more leverage, getting more cyclical, pricing power and rolling, but the multiples are as if there's no cyclicality. So for, and by the way, if you backtest companies that are increasing capital intensity, or reducing or declining free cash for margins, the outcome is not good.

22:03It's not probabilistic. It is not in your favor. So, and by the way, same thing you can say for so many other companies, Amazon, AWS is declining margins last couple of quarters. The revenue growth is being held by a lot of startups using compute from these companies. And the startup, as you know, it is a bucket that is filled every year. Hence, you have to go to the bond markets. So this is actually quite a unique setup where the market narrative is completely not appreciating how these businesses are becoming, in our opinion, worse than telecom used to be in the late 90s. And I think a lot of that is supported by the narrative because people use the AI and they see how good it is.

22:51And they're very optimistic about what that means for the future. Well, I think one could, by the way, there are more and more doubts even of that. Chad GPT-5 was hardly any improvement for 4.5, which was barely anything better than Chad GPT-4.0. Right? Why is that? Because they've already optimized all the data publicly available. And actually, the Chinese open source is becoming a mainstay. In fact, Anderson Horowitz had a data point. The 80 % of the entrepreneurs, AI entrepreneurs now are using Chinese open source. They are much more frugal in terms of needs for compute. Right? And by the way, that's the companies that some of the larger companies we talked to saying the same thing.

23:35so what exactly are we building LLMs for? Not to mention the huge issue of hallucinations. I mean, I don't know if you saw, but there's been a famous case of Deloitte in Australia where the Australian government had asked them to do a project. And what it found out when the project, they submitted the report, the report had fake ministries, fake court cases. If I'm memory serves me correctly, even a fake name of the minister, right? I said, hey, even a 12-year-old, a high schooler would not make those mistakes because they're not going to make up stuff, right? So hallucination is a real problem.

24:13And there have been surveys which have clearly suggested in service industries that sometimes using AI actually increases work, not reduces work. So we are far away. This is powerful technology. But if you have a tax accountant who's 75 % accurate, it doesn't work, right? So these tests, they are 70%, 75%. Yeah, good luck using a tax accountant who is 75 % accurate. The other big headline is the sustainability of the mismatch between AI infrastructure spending and revenues. How do you think about that? And do you see any historical parallels to that spending revenue gap? Yeah, see, the funny part is when people compare to dot com.

24:54So we actually went back and read some of the old annual reports from 99, 2000. And there was an interesting nugget from Cisco's annual report of 99. It said in the US, the internet economy, as they called it, in 1998 was$300 billion. This is Cisco's annual report of 99, which came in early 2000. Today, revenue from AI is under$25 billion. And if you look at, let's look at OpenAI. More than half of the user base is from emerging markets. India, Indonesia, Brazil, Philippines, Turkey, do they really have pricing power? Let's look at that. So if you look at the telecom operators, for example, in India, a company called Bharti, the average revenue per user per month is$3.50.

25:46I'm not sure that OpenAI would be able to charge$20 for 5 % or 10 % of the users in India or Brazil or other places. It just won't happen. Not to mention, there's no stickiness because you can change from one to the other in a pretty quick fashion, right? And third thing is that they've already lost market share to Anthropic in coding, etc. So this is a very fluid space. One of the biggest problems, by the way, which I think is a real Achilles heel for this industry is it doesn't scale well. What do we mean by that? So if you look at every incremental user that OpenAI gets, it may be equal or more expensive from a compute perspective than the prior one.

26:33There's no real economies of scale. In a software business, if you get a new software client, fixed costs are already embedded, right? So the incremental margins are very, very high. In this case, incremental margins may be worse. well because if you get a power user they may completely blow through your your your budget for that individual so then they end up throttling them so think about this way you're actually trying to control your best users the other thing by the way is when any company tries to hide numbers you know it's not good news right i mean i've yet to come up with a company which which hides because they're so good that we want to tell everybody else right these companies is bullentropic, are bending over backwards not to disclose any numbers.