Welcome back to Excess Returns. I'm Justin Carbon. I'm here with Jack Forehand as always. And today we are joined and have the privilege to be joined by Dan Niles. In my opinion, I think there's a small handful of people we consider elite thinkers when it comes to the technology sector. And for a number of years, I've admired Dan and his work. He is founder of Niles Investment Management, has been portfolio manager at Niles Investment Management for two decades, and you know, with you have over 30 years experience in researching and investing in the technology business. And it's a very interesting time to be someone that has your views on tech, Dan. There's a lot going on, and so yeah, just thank you for joining us. We think this is going to be a great discussion. A lot of it's going to be focused around AI, but we know you have, you know, thoughts on a lot of things when it comes to tech. So really appreciate you taking the time to join us today.
Oh, my pleasure, Justin, and thanks for having me on.
I want to start with something that you said earlier in the year, and I think it was along the lines that yes, we are 100% in a bubble, or maybe there was a specific pocket you were talking about, but that you specifically kind of keyed in on, you thought that, you know, semiconductor stocks were going to see somewhere in the range of, you know, 30 to 50% pullback. And since then, you know, semis have pulled back quite a bit from June. So just kind of where, let's use that as a starting point to where you think we are currently in sort of this bubble cycle, if we are in one.
Yeah. Maybe to look at it very big picture. So if you look at every great industrial revolution that we've had, and that's, you know, whether it's canals in the late 1700s or railroads in the early 1800s, or radio, TV, fast forward to electricity or the internet most recently. If you believe you're in one of those revolutions that's going to reshape the landscape of the world, then by definition, you're going to have a lot of people chasing that opportunity because they know if they're the last company left, they're going to make an inordinate amount of money. And so if you believe that AI is one of those technologies, which I do and I think most people do, then by definition you have overinvestment going on. Now the thing is that that overinvestment doesn't necessarily mean it's a bad time to invest. And in fact, it means it's a great time to invest. The problem is when that eventually breaks, you know, you're going to have a much bigger than normal meltdown on the other side of it. The good news about what's going on today relative to the internet is valuations aren't at these stratospheric levels that they were back then. And if you take Cisco as a good example, which was sort of the Nvidia of its day, it was trading at a PE valuation of over 100 times of earnings. Nvidia, in contrast, is at 15 times. So when the bubble does eventually break, you're not going to have the same kind of, or at least I hope you don't have the same kind of meltdown that you saw back from the peak in March of 2000 to October of 2002. And that's why I'm hopeful that any meltdown in semiconductors, which are sort of the tip of this AI sphere, will be, you know, limited to 30 to 50%.
How do you think about, because you hit on something interesting there. It's, you know, it's still a great, even if we're in the bubble, it doesn't mean, you know, you go to cash or anything because that bubble type of environment can last a lot longer than many people think. But what would be the things, and you kind of hit on valuation a little bit, but what would be the things you'd be looking at to make you questioning, okay, we're more towards the end here, this bubble than not?
The number one, well, there's two main factors, right? You've got the revenue picture, and within that you have two very simple things. There's number of tokens being produced and what you can charge for each of those tokens. And what you've seen since end of May is what you can charge for those tokens has gone down by about 50%. Why? Because you have this thing called open source or open weights that are coming out with all of these cheaper models. So what you can charge for each token being produced, which is sort of the unit of currency for AI, has gone down 50%, which seems terrible, except that the number of tokens being actually produced is up two and a half times since the end of May. So I'm watching both of those things very closely to see that, okay, is the dropping in costs being offset by a more than offsetting increase in tokens. The second part of that is how profitable is this? Because you may be generating revenues, but if you're losing tons of money, it really doesn't matter. The good news is profitability is also improving. And if you look at the big three public cloud vendors, and so that would be obviously Amazon Web Services, Microsoft Azure, and Google Cloud Platform, all three of them saw accelerating revenue growth from the March to the June quarter where revenues went from increasing 35% in March year-over-year for their cloud businesses to up 43%. But more importantly, the operating margins also expanded about 2 percentage points from March to June. And so not only were revenues accelerating, but the profitability of those revenues was also going up. So those are kind of the bigger picture things that I'm watching to see when I really worry about we've reached the peak of this bubble and we have to worry about the other side of it.
Mhm. I was recently talking to a colleague who I'm texting here to see because I thought he said, I want to think he said he used, oh yeah, 1 billion, this is an individual, 1 billion tokens in 24 hours. So to your point, this token usage is certainly going up, especially with the guy I know and Jack, you know who I'm talking about. So anyways, but so yeah, talk about, Dan, I wanted to get at this thing you brought up with Situational Awareness, and what's interesting is your take on it, on how sort of that solved, sort of that the failure of that fund actually kind of reduced some near-term risk. So can you kind of just explain that?
Sure. I mean, there's only two things that drive stocks, right? It's earnings. It's the multiple on those earnings. The multiple on those earnings gets overinflated because you have individuals as a group or funds as a group doing stupid things. And stupid things usually means you've levered up. And so you saw that with Korean retail investors who got absolutely destroyed when SK Hynix and Samsung imploded, but you also saw that with funds such as Situational Awareness. And there were some other funds that I heard that were in trouble as well. But luckily, Situational Awareness being pushed into that forced sale with Citadel helped solve some of those issues because the market obviously ripped the next day, especially in a lot of those positions that were getting unwound. So with Situational Awareness, the problem wasn't necessarily the positions they had. The problem was they were running it at 4x leverage. And so if you do the math and you say, well, if you have a 25% drawdown, but you have 4x leverage on it, then you're down 100%. At which point your fund is liquidated. It's the same thing we saw with Long-Term Capital Management in the late 1990s where they had a tremendous amount of leverage. At least in their case, they were trying not to have it be directional leverage. But similarly, you have a lot of leverage and you have your shorts going up. In this case, software was rallying, and your longs going down. In this case, semiconductors were getting crushed. You have both sides of your book working against you. You throw the leverage on top of that and you force this sort of cataclysmic bottom. But remember what we just talked about. This is all going on, but the fundamentals are okay because you've got a 50% drop in token costs, but you have a two and a half times increase in tokens being produced from the end of May till recently. And so at least the fundamental picture looked good. And so that's why I felt like, okay, you know, the timing was fortunate, but I put out that note on July 29th thinking we reached a short-term bottom because I had been hearing about funds in trouble. I thought we were pretty close to getting something, you know, cataclysmic to happen. And in fact, it did with the Situational Awareness forced sale that got announced the next day.
One of the big questions I think a lot of people are asking is like, what is the ROI to the end user here in the use of AI? So how do you kind of think about, I guess, that, I mean, we're all sort of benefiting from it, I think, in our work lives, right? But like, what is the, how do we, how do you think about like the end benefit to users of AI in terms of higher profitability or whatever it might be?
Well, I mean, it's really a combination of all the above. So if you look at the hyperscalers, part of the reason why a lot of those stocks have been struggling is because people go, okay, you are seeing accelerating revenues. That's great. Your profitability is improving. That's great. But unfortunately, your cash flows, which had been hugely positive, like for Google, are now negative for the first time since they went public. And so that's the part that people are wrestling with because if you've been around the block a few times and especially lived through the internet bubble, you and I think we can all agree, right? The internet was a pretty good revolution. It's not like things didn't keep evolving in 2001 and 2002, but the NASDAQ went down 78% from its peak in March of 2000 to the lows in October of 2002. And it was a punishing, grinding beatdown over two and a half years. The internet kept growing over that whole period of time. In fact, it doubled, you know, each of those years. But the problem was people were thinking, oh, well, it's going to double every three months or six months. And so that's kind of what's going on. And so there's no, you know, one answer to this because what'll happen is eventually somebody will have an issue and then they won't be able to live up to their commitments, and at which point you're going to see a slowdown in capex spending by one of the hyperscalers, at which point they'll all probably go ahead and slam the brakes on at the same time because they all kind of know they're overinvesting. I mean, you've had, you know, the CEO of Microsoft talk about that before. You had Larry Page say something along the lines of, I'd rather go bankrupt than lose this race. But it doesn't mean that, and I think we've got another year of pretty strong growth and profit growth in front of us. And so those are kind of the interplay between revenues, profitability, but cash flow, and then obviously valuations that I'm kind of looking at all together.
I'm wondering when you compare this to the late 90s, do you see, one of the ideas I've been thinking about is, are there more governors here in place that might prevent us from getting a huge bubble? And what I mean by that is we don't have enough compute, we don't have enough electricity, and it seems like right now we're kind of meeting demand, whereas in the late 90s, we were building out in advance of demand. I mean, do you think that fact that we can't meet demand right now might like prevent this bubble from getting as big as it otherwise could?
Yeah. No, I think that's an incredibly, I think that's a great point because I put out a piece, you know, a couple of weeks ago and I said the one really negative development that surprised me is that data centers are getting pushed back on really hard. And in the latest Gallup poll, I think when people were asked about having a data center in their backyard or nuclear, it was 71% against data centers and only 53% against nuclear. Like that's really bad, right? Like I couldn't believe it when I saw it. And it's like the Democrats and Republicans can't agree on anything. But what they all seem to be agreeing on going into the midterms, which surprised me, especially from the, you know, the red states like Texas, is data centers. They don't, you know, they're all pitching running against data centers, which I, by the way, I think is wrong because I think it's great for the communities that have gotten it, but the hyperscalers have not done a good job of, you know, doing PR around the benefits for these communities. And so the other side, and let's call that, you know, the socialist movement, which by their own definition is against big business, has done a good job of kind of selling this as a bad thing. And so that's the part that really could cause a problem, not so much as you said from the demand side, but from the ability of actually building these things, which, you know, there's enough people out there who've been negative on the whole AI build from, you know, beginning of last year that will jump on that as all the bubble's breaking, etc. But I think again, it's something that should help prolong this buildout for even longer because you don't have the land to actually put these data centers on as we approach the midterms. And so that's one of the reasons why now, between now and the midterms on November 3rd, I'm pretty negative.
It does seem like the political risk is maybe the biggest risk here to AI, like in terms of AI growth. That could, that's a very high level thing that could turn against AI and already has turned against AI, and it seems like that might be the biggest risk. Do you agree?
Well, that's a short-term risk. I mean, I always want to believe that people will do the rational thing. And when you see the data coming out from the communities that have had some of these big data centers come in, the amount of jobs that's created, the amount of tax revenues, they don't use that much water at all. If you got rid of all the golf courses, you'd do a much better job with that. You know, they put in their own electricity. A lot of it is they're trying to make it renewable going forward. And so I want to believe with more education that that political risk, especially after the midterms are done, will hopefully die down. And now this is a big if. That's assuming that you don't end up with this huge landslide in favor of, in particular, socialists that want to make these companies forcibly give part of, you know, the share ownership to the government and all these other things that, you know, if by your own definition socialists don't like big business, that's not good for the big businesses that are involved in AI. And so that's the part that, you know, I do worry about going forward, but I want to believe rational, you know, thought will prevail, but we're going to have to wait and see what happens.
How do you think about across the AI stack where the benefit will occur? I mean, if you look at history, a lot of times the builders of these technologies have not been the ultimate beneficiaries, and it seems like so far here they have been. But just, just how do you think as you look up and down the AI stack, like where the most benefits will occur?
Well, you know, it's a great question and you're 100% right. How many railroads are in the top 10 market cap in the US? What about electricity companies? What about canal companies? What about fracking companies? Right? And, you know, Cisco, which used to be the most valuable company in the world, just recently got back above its market cap from 2000. You know, that's a pretty long wait. So when I think about this, I do believe open source is going to be increasingly used. I'm sure that person you talked about said he used a billion billion tokens in what did you say, a week or a day or something?
24 hours. In a day, I'm sure he's not paying a dollar a token, right? So it's, you know, you're going to have your cost continuing to collapse. And I always like to say you don't need a Ferrari to go to the corner store to get milk, right? A Ford will work just fine. And so people aren't going to be using Anthropic to go summarize their emails, right? They'll be using some open-weight open-source model to go ahead and do that, and they'll use Anthropic or ChatGPT to do that front-end, you know, replication of some software point solution tool that they don't want to pay all this money for. And so I think ultimately the value is going to move from the model providers to the infrastructure providers and then ultimately the companies sitting on top of that that create great businesses. Because if you think about the internet as an example, all that bandwidth and capacity that was put into place, when it collapsed and imploded on the other side, it gave you the infrastructure necessary to have companies like Google and Amazon and Facebook, you know, Netflix all show up that were built on the back of all that cheap bandwidth. And I think on the back of all the cheap tokens we're going to have, you're going to have some really great companies coming around because of that.
How about within the model layer? Like I would think what you said is not great for OpenAI and Anthropic. And it's been interesting, like you're seeing more and more usage, as you said, down at the lower-end models, but they still have captured most of the value with the high-end models despite usage coming down. Like do you think that changes and maybe the value starts to trickle down away from them?
Yes, I do. And I worry more about OpenAI than I do Anthropic because if you look at the kind of the genesis of both companies, ChatGPT was launched for consumers. Anthropic started focusing on enterprise, and I've been saying this now for, I don't know, a year or two at least, that I think Anthropic is going to, and on the enterprise side, and then Google on the consumer side, are going to squeeze OpenAI between them. Now obviously OpenAI was much bigger than Anthropic early on. Now Anthropic is much bigger than OpenAI early on. OpenAI has now switched from consumers, which, you know, we've been all trained that, hey, we get answers for Google for free. So why on earth would you pay for it, right? And so that's why OpenAI has switched to enterprises where enterprises pay for their stuff. And so that's why Anthropic's going public first. But I think as this continues to evolve, OpenAI and their ecosystem is what I worry about the most because of the fact that if you look at the history of technology, you typically have a winner-take-most market, right? If you think about, well, you know, who's the market share leader in e-commerce? Well, it's really only Amazon, right? Then you have a bunch of smaller players. Well, what about in social? Well, you only really have Meta and then you have a bunch of smaller players. What about in streaming? Well, you really only have Netflix and a bunch of smaller players. So do you think you're going to have 10 models that are all thriving? I mean, anything is possible, but the history of technology tells you that's probably not likely. And so we'll have to see what happens on the other side of this. But my belief is that, yeah, you're going to have some one of the large players, right? If you think about the internet era, I would have never guessed AOL would not have thrived, or Yahoo, or you can pay, you know, Lycos or Netscape, like these are all public companies that don't exist. So when I look at today, just because ChatGPT is what started all this doesn't mean that OpenAI is going to be ultimately the biggest winner. And I think most people would agree that Anthropic is already in better shape than OpenAI is.
I want to shift to semiconductors, and sometimes I ask a question that I know the answer to, and I'll do that here. But you have seen a lot of talk about this idea that semiconductors are no longer cyclical. And I know you're going to tell me that's not true, but I am wondering, like, do we think about it differently in terms of that cyclicality? Like, do we think about that cyclicality in a much longer time frame? Like, all the people that are saying, you know, semis are no longer cyclical will be wrong eventually, but this might carry on a lot longer than people on the other side think. I mean, is that a fair way to think about this?
What I would tell you is it's the same answer as the AI trade. When do you think the bubble breaks? Actually, it's not. You know what? I take that back. You could argue, I'm actually starting to wonder if the semiconductor bubble breaks before then for one simple reason. I think a lot of investors are vastly underestimating what China is doing. And I hear these arguments and I go, this is what people were thinking, you know, back in the 1980s when Japan entered the semiconductor industry, and that was state-sponsored as well. And, you know, some of your listeners may not know, but like Intel at one point had 75% market share in the DRAM industry. The US was the dominant force in that industry. And then Japan started, and they were obviously several generations behind, but they caught up. And then by the early 1980s, Intel was facing bankruptcy, and they switched to this thing called microprocessors, and Japan took over the DRAM market. And then you fast forward to the 1990s, and Korea, also state-sponsored, entered the DRAM industry, and the same arguments were made then. Now don't forget you also had an internet buildout happening at the same time, but Korea entered that market, and then by the 2000s you had the Japanese companies being driven out of business. What you had recently is CXMT, which is China's DRAM company, go public. And then YMTC, which is their NAND company, is going to go public very soon. Also, CXMT plans on increasing their wafer starts from 300,000 to 500,000 by the end of next year. And YMTC, which is their NAND manufacturer, is saying they want to be big or they plan to be bigger than Samsung or Hynix in the NAND business by the end of next year. Now the combined market, if you look at CXMT, they have about 8% of the global DRAM market, which is nothing, but it is ramping incredibly fast. And for commodity memory in particular, where a lot of people are making this argument there's going to be shortages through 2030 and all of this stuff, if I go, well, if the Chinese meet the goals that they've set out for themselves, that's, there's no chance that it's going to, or there's always a chance, but there's very minimal chance you're going to get there because China is obviously a much bigger country than Japan or Korea ever were in global GDP relative to, well, the population size, land mass size, ability to put up these massive factories relative to Japan or Korea. And from their perspective, unlike with Japan or Korea, they've been cut off from chips from the US. So from China's perspective, having their own semiconductor supply is as important as having an aircraft carrier or nuclear weapons. It's more of a defense technology in some ways even than a semiconductor technology. And so this is why I look at semis. And the good news is, as I said earlier, right, Nvidia is trading at 15 times. Not that I think Nvidia is where the problem is. And the memory companies likewise are trading at valuations that the people who believe that this is no longer cyclical, oh my god, they're very undervalued. But that's kind of what I am thinking through right now because the data, and you even saw it today where CXMT said they had produced a high-bandwidth memory that was actually yielding very, very well, and it's maybe only, and this one's only like a generation behind. And as I said earlier, sometimes a Ford will work just as well as a Ferrari if it's not the bleeding-edge applications.
Yeah. Yeah, one of the things I always have to remind myself about is like everybody gets caught up in these bubbles to some extent. And so one of the arguments you hear people make is, well, the people running all these tech companies are incredibly smart people. So they're throwing, you know, these massive amounts of money for a reason. But you kind of have to take a step back and say, in the late 90s, those people were incredibly smart that were running those companies and they were throwing the money for a reason. So I sometimes have to check myself and like think that through.
Well, for anybody who believes that, they should go on to Cisco's investor relations website, pull the earnings release from May of 2021, if memory serves me correctly, where the CEO of Cisco Systems says we've gone from 70% year-over-year bookings growth in several months to negative 30% year-over-year bookings growth. And they were at one point the most valuable company in the world. So yeah, to your point, you know, these, yes, are these really, really smart companies? Absolutely. Are they the best companies that we have right now? 100%. But do massively smart big companies get it wrong? Yes, all the time. And that's the thing about bubbles is, you know, they're great to be invested in. You can make a lot of money if you stay with them, but obviously you have to get the timing pretty well because you never know, well, is this the actual break or is this just a pause that refreshes? And when I put out that note on June 20th, I said, look, you know, I think we have a short-term problem here. We have a speed bump. And then, you know, I put out that note on July 29th saying, hey, I think, you know, we're near the bottom. We should be able to see things starting to rally again. The backdrop of all of this is agentic AI is new. We first heard about this thing called OpenClaw being formalized on January 30th, right? That's what, seven months ago. So it's brand new. We didn't have that with the internet if you think about it, right? There wasn't like a new thing that came out for the internet. If you look at AI, you've had kind of distinct phases, right? You had the training phase initially. Then you had the inference phase, and then on January 30th of this year, you had the agentic phase. And the agentic phase uses 10 to 100 times more tokens than the chat-based AI phase. And so I think you've got a long way to go, at least another year for stocks to go higher. But I do think it's getting, you know, like what I said about the Chinese memory companies, I do think you're going to have to get more selective and then watch the data like a hawk to make sure that you're not deluding yourself either by being too negative or being too positive.
On that idea of watching the data, we don't talk about individual stocks a lot, but I know like in a period like this, looking at the reports of individual stocks can tell us a lot about what's going on overall. And I'm just wondering as an expert, like when you looked at Nvidia's report, I mean, it seemed like a blowout report. Was there anything you learned about what's going on like overall with AI?
Not really. I mean, the only, I mean, I thought it would be bullish. It was quite honestly the only thing in that report I didn't like is when they gave long-term revenue guidance because Nvidia for its entire history has, you know, guided out one quarter in advance. I know why they did it. I understand they're like really frustrated with, you know, the way their stock's performing with the multiple relative to other semiconductor names in the industry. But, you know, and I think Jensen is amazing. He's actually a great human being to boot on top of that, visionary as well, but that's the only part I didn't like about it. And I know they're trying to give confidence by saying, hey, we think we can grow revenues at 70% in calendar 27 and we think demand is actually at 100%. But at the end of the day, you know, there's a lot of other factors that's going to determine how fast AI is going to ramp and, you know, that stuff evolves over time. I mean, Cisco Systems in 2000 was saying, hey, we think the internet can, we think our revenues can grow 30 to 50% sustainably in the future, and then two years after they made that proclamation, their fiscal year revenues were down for two consecutive years in a row. So things change and, you know, the Nvidia report made me feel good in that, okay, yeah, I do think there's another year for this to run, but the magnitude we'll have to wait and see how that shakes out. But, you know, trading at a 15 times PE off of calendar 27 numbers with the S&P...
Up at 19 times, you go, this isn't Cisco in 2000. Even if they're wrong, it's not Cisco going down 20,000 in terms of the downside.
