Jan van Eck: Everyone Is Missing the Biggest AI Opportunity

So I say stay invested in the market. I mean, my one exception is if you know monetary policy is going to tighten like crazy, that will hurt financial markets. Of course, if you could predict when the budget crisis is going to happen, then you get out as well, but that's hard to predict. So generally, stay invested. That's number one.

I'm great. Look forward to catching up. It's been crazy markets.

Well, there's lots of different things in that one question, John. So I'm going to just first of all jump to—I do these quarterly outlooks and there are a couple of charts. It's funny that you picked Korea. Look, this was clearly—there was a lot of froth in the second quarter. I tend to break up the world into quarters and then decades. And we actually sent out a note saying, look, this is probably not sustainable. I think the Korean market right now is around 6,000. Look, it went sideways for like 17 years and then exploded to the upside, and obviously there's nothing dramatic that changed. So clearly, number one, sign of froth. I would say Korea in particular, because it's dominated by the memory chip companies, was a sign of—so there's a whole AI ecosystem, obviously, and there's parts that I think are maybe softer and kind of stronger and more structural. I always ask the question: if we're looking back in five or 10 years, which of these companies in the AI ecosystem have a competitive moat, which are the survivors? And there were two areas that I was sort of wary about coming into the end of Q2. One is these memory stock companies, and the second one is the single model, if you will, companies—just the, what they call, the frontier labs like Anthropic, because both are just playing one role in the ecosystem rather than having a front-to-back connectivity between customers and compute. So, the memory chip companies, just to answer this specific question, really benefited. Their profits went up because they upped prices. They weren't able to increase volume that much, John. It was really, hey, you need me. There's a shortage of compute, and I'm going to raise your prices. And that's okay for a one or two-year game in corporate America or corporate world. But your customers are going to try to use less of your product, and secondly, you're going to have competitors. So even within the last couple of weeks, a very large Chinese memory chip company came to market. It's now the largest cap in China—it was for a while. So already this weak part of the market, even though it was a speculative excess, I think is being attacked in the ecosystem.

So this is a little bit dated, but I've updated numbers. So first of all, you can see that these are the big powerhouses of the equity markets, right? They're mega caps, and their earnings growth, I would call it, is justifying the market. Now, there are two caveats to that. One is that a lot of this profit growth in Q2—and I'll give you the adjusted Q2 numbers—but Q2 was driven by mark-to-market gains in their portfolios, because for example, Google was one of the biggest investors in Anthropic. So this just shows earnings growth. But if you do adjusted—sorry, just operating income that takes out these one-time portfolio gains that these companies, many of them do, you still had really impressive growth. So, Amazon's Q2 year-over-year operating income growth, 43%. I mean, that is just incredible. Microsoft up 18% operating income. Google up 30%. Again, taking those investment gains out, John. And then Meta is down 8% year-over-year profit. They just were spending a lot of money. They also had some legal expenses that are worth noting, but their revenue—they're still growing crazy amounts for a large company. They were growing 28%. So, there's two things we need to strip out. If you're worried about earnings, one is that earnings driven by just portfolio gains, gains in investing in other parts of the ecosystem.

And let's just double-click on that one more time, which is you could imagine a quarter where, let's say, Anthropic's valuation goes down by half. The accountants, the auditors, the regulators will require these public companies to show that markdown in their earnings. So I suggest that investors prepare themselves in one future quarter. I mean, Anthropic's not going to go up every single quarter, right? So one quarter we could have an earnings surprise, an overall earnings surprise because their portfolio holdings are down. I think if we start looking at just operating income again—the numbers 43%, 18%, 30% growth in operating income for Amazon, Microsoft, and Google—those are still very solid, backward-looking albeit, but very solid operating income growth numbers. So that's one thing that we need to do to kind of adjust this earning story. And the second is I just talked about the memory companies, where they weren't growing unit volume as much as they just jacked up prices. And so that's also something that we think is a shorter-term blip. But if you take the ecosystem as a whole, I think we're in really good shape.

I don't know. I think a lot of the froth is now gone. I mean, we see parts of the ecosystem, kind of with exuberance and then being pretty quickly punished, right? There's this recycling. So, I don't have the chart here, but Korea will serve as a good example of this. In the fourth quarter of last year, Oracle—there was a lot of worry as you may recall about the finance ability of Anthropic's compute needs and OpenAI as well, right? And Oracle and CoreWeave were the public stocks that kind of took the pain there, and they were both down 50% in a quarter. So over the course of July, right, you had semis being down 20% and again some of the more marginal players down even more. Some of the Bitcoin miners that are transitioning to compute companies got hit a lot more. So I think a lot of the froth is already out of the market.

So as I said, you want to look at the ecosystem. I like to talk about history a lot, and people keep talking about railroads as kind of the quintessential bubble, or the internet bubble, right? So again, which companies are going to survive? The one thing you need to think about, I think, is the market structure of the industry. So railroad stocks in the United States peaked before the technology was built out. So before the transcontinental railroad was built in the 1860s, railroad stocks kind of peaked in Europe—sorry, in the UK and the US in the 1840s, just sort of big picture. But a lot of those railroad companies, they weren't really a complete ecosystem, right? It was only like the railroad from one end of Pennsylvania to the other, or New York up to Connecticut. These weren't companies that really provided the full ecosystem. So that to me is—the transnational railroads, if you will, are that full solution end-to-end for customers. And so I think of Google and Amazon from that perspective. This is all about the American consumer, right? So, do these companies own the American consumer or have a big market share? Absolutely. Right. Google, Amazon, Meta does as well, right? Then are they vulnerable, or do they own basically the compute that they need to provide those services to those customers in a sustainable manner? Then do they have leading hardware—I call it leading software, these frontier lab models—and they also have some of the physical infrastructure. They're starting to design their own chips. And so I think those companies are just with that vertical—I call it vertical integration—have more of an economic moat, they're more protected. I mean, unless they lose their touch with the consumer at the front end, I can't see any technology change where they can't adjust and continue to service the US consumer, which is the biggest market in the world. Right? I can contrast that with companies like the memory companies, or just the chip companies like an Intel or a Cirrus, where they're only really providing part of the ecosystem, and I personally think they're just much more vulnerable. And then also those single LLM models. And in the third category, they're software, but they're also right now—they're trying to reach into the device, and they clearly really have the customer, at least OpenAI does. So the question is how quickly can they build out the compute?

Yeah, I mean, I call it the IBM mainframe of the AI era. So Nvidia used to just be a graphics chip company. And if you've ever, as I have, run a company in a commodity business, right, where you're very vulnerable on one product and you're only one part of the ecosystem, you're just dying for that strategic opportunity to be more front-to-end. And so a lot to me of what Jensen Huang does at Nvidia is to be sticky to customers. So probably the most important thing is the CUDA software, which is the programming of their chips, which their customers—I mean, I'm not saying there aren't competitors, there are, but that's one of the things that's made them more competitive. They're obviously at the cutting edge of chips as well in terms of efficiency, and they continue to innovate there. And then they're also—the chips oriented towards the training of the models and the frontier labs, or the kind of I'll call it use of trained models, what they call the inference—and Nvidia stretched into those parts of the ecosystem as well. So, their growth, earnings growth is fantastic. Their sales numbers are fantastic. The stock has become very unloved, and I'm sure it's driving Jensen Huang crazy. But the valuations—I think I still have a chart of that here, just to show what that looks like. A lot of charts here. But you can see the forward PE ratio on the right-hand side of Nvidia. So, it's as attractive as it's really been in the AI era, but its revenue and profit growth is still at the high end of any one of these peers. So, I still think it's one of the core companies of the hardware stack, and I'm comfortable. It may not be the top high-flyer, John. It may not go up seven times in a month, right? No longer giving it large market cap. But it's still very core, has a strategic, I think, competitive advantage.

And the other thing I wanted to mention about their behavior—now, this chart is extremely dated because it's like two weeks old—but the point of the chart here is that in the black line is IBM's share price. Now, IBM has been core to corporate computing for decades and decades, and was basically very slow but able to pivot from hardware only to software to services. Its stock is not that exciting, which is the black line, but if you add reinvested dividends, it's very competitive with the NASDAQ. Now, of course, it got slammed two weeks ago because they pre-announced a revenue miss, but the story is still there. And Nvidia is now paying out some of its dividends—doing buybacks and reinvesting in the ecosystem—but its cash flow is tremendous, and it's returning some of that to shareholders. So, anyway, that's my kind of pitch on Nvidia, John.

Yeah, I mean, I'll show you two slides. I know I'm on page 14. Q1 was a huge breakout for the Frontier Labs because Anthropic demonstrated that corporate America really wanted their product and they were showing tremendous revenue growth. But as I show through my very uncool bullseye, once you know you're grabbing that much of corporate America's IT budget, people are going to come after you. And I just want to articulate the threats that came out in Q2. So number one, and I won't go through all the charts here, but number one, obviously at some point corporations are going to try to optimize their spend on AI. I would say they don't really at mass scale have the tools to do that right now. But the idea is that there's a term called 'harness' which gives corporations and gives users more flexibility to pick different AI LLM models given the tasks. Like just at VanEck, I'll throw in one example. Like half of what people use Claude for at VanEck is just what you could use a Google search for for free, right? But once that prompt is up there, they just use it for everything. That's not optimized. I think over the next year, corporations will try to optimize their use of these models and not just maximize.

Secondly, they have a lot of competition, right? And it's not a secret. The Chinese open-source models are very competitive with Claude's quality. At some point, I think that will dawn onto the market that despite their tremendous revenue growth, they may not have a competitive moat on the pricing of these models, and they may not be that different from each other, and you may be able to use older models for different tasks in corporate America. There's only a subset that really needs the latest greatest. Third—I won't go into it—but you know, there are some voices out there in Silicon Valley saying that when a corporation puts all its data into these Anthropic models, that they shouldn't be trusted, that they end up being competitors, and it's better to wall off your corporate data from the model companies themselves, which is another version of bullet one but it's not exactly the same point. So, anyway, with those points, John, I just think they look to me like—I wouldn't want to have a lot of free competitors with VanEck ETFs, right? Just as a starting point. I wouldn't want to set up my customer base to be really mad at me and trying to move away. And I wouldn't want—I'm going to fight very hard to control, if you will, the entry point or the optimizer of these AI software models. But that's a risk out there.

I would say we're still, to use a phrase, token maxing. I still want people to kind of understand and use these tools as much as possible, because adoption, you know, technology adoption by humans is uneven and unobtrusive sometimes, right? Projects fail on Wall Street like half the time, and the reason is users just don't adopt them. So we've been trying to encourage what I call our champions—those who are building their own citizen software or using these models—and then kind of getting proliferation. I would say there's some—just to give you an example of one of the issues that we have—we're starting to create a lot of agents at VanEck. So how do you name agents and separate them? Interestingly enough, we're using Zoom. Well, we give an agent an email address and then it has used Zoom as my interface with my personal agent. So all I have to do is do a Zoom chat with my agent saying add this skill or whatever that might be. So I thought that—I just don't know if that's repeatable. One of the inefficient things of everyone writing their own software programs is that a lot of people are writing the same software programs. So there's lots of things that—but to me, I always say there's data and there's software and there's people, and at the end of the day, data really, really matters. So our data organization, I think, is pretty good on the customer end. We buy a lot of data to understand our customers better. And on the investment side, obviously data is super important as well, but it's hard to collect the notes of individual analysts, which I would consider to be proprietary data, and then to propagate it across the firm. So push the champions, don't limit the spend right now is kind of our overall philosophy, and then try to, I guess currently, really communicate more across the firm and try in terms of adoption. So that's where we're at.

Stop that. Because of course, you know, two clever colleagues of mine started having their agents try to talk to each other and saying like, well, what are you doing that's better than I'm doing, and our IT department put a stop to that right away.

Not yet. Not yet. I think we need to adopt kind of an interface or a harness. Like I would just say we're at this sort of simple stage where we moved from OpenAI in February this year basically to be an Anthropic shop, but we're definitely going to take that next phase. But to move faster, we'll probably 3x our Claude usage or something like that before we can kind of build that infrastructure. Of course, some of our geeks and researchers are doing open Claude and all that kind of stuff and creating their own environments, but there's a big learning curve and the protections on cyber and again some of the stuff I was talking about need to be there before we can unleash it.