Dissecting the AI Revolution's Economic Ripple Effects with Industry Expert Dylan Patel

Discover the intricate web of the semiconductor and AI industries as we sit down with Dylan Patel, chief analyst at SemiAnalysis. Peering beyond the surface, our conversation unravels the global complexities of the semiconductor supply chain, the geopolitical chess game impacting the industry, and the seemingly irreplaceable expertise of specialized companies. With Dylan's rich background in data science and a team boasting experiences from ASML to hedge funds, this episode is a treasure trove of insights for anyone keen on the technicalities and market dynamics that govern these pivotal sectors.

Venture with us into the depths of the AI chip market where the tides of power are ever-shifting. We scrutinize the balance between tech behemoths forging their own AI processors and the enduring dominance of Nvidia. As we dissect the pressures of chip design and software ecosystems, we also cast an eye towards the potential of market consolidation, and what it could spell for smaller entities and the future of innovation. This is a must-listen for those fascinated by the push-pull of economic forces and technological advancements driving today's digital revolution.

Finally, we zoom out to the broader economic landscape, drawing parallels with the internet boom and pondering the revenue and valuation implications of the AI surge. Are current market leaders investing wisely into AI, or will history repeat itself with newer, more adaptable players taking the lead? We offer our perspective on investment strategies within the semiconductor capital equipment sector and address the misconceptions surrounding NVIDIA's venture investments. Engage with us as we navigate the complex yet exhilarating intersection of AI, semiconductors, and the future they're intricately shaping.

Nothing on this channel should be considered as personalized financial advice or a solicitation to buy or sell any securities. 

The content in this program is for informational purposes only. You should not construe any information or other material as investment, financial, tax, or other advice. The views expressed by the participants are solely their own. A participant may have taken or recommended any investment position discussed, but may close such position or alter its recommendation at any time without notice. Nothing contained in this program constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments in any jurisdiction. Please consult your own investment or financial advisor for advice related to all investment decisions.

 Sign up to The Lead-Lag Report on Substack and get 30% off the annual subscription today by visiting http://theleadlag.report/leadlaglive.

Foodies unite…with HowUdish!

It’s social media with a secret sauce: FOOD! The world’s first network for food enthusiasts. HowUdish connects foodies across the world!

Share kitchen tips and recipe hacks. Discover hidden gem food joints and street food. Find foodies like you, connect, chat and organize meet-ups!

HowUdish makes it simple to connect through food anywhere in the world.

So, how do YOU dish? Download HowUdish on the Apple App Store today: Support the Show.

Transcript

Speaker 1: 0:08

My name is Michael Guy. I publish here off the lead Lagerboard. Join me for the hour at Dylan Patel. Dylan, the first time you and I are going to be chatting here, but introduce yourself to the audience and to me. Who are you, what's your background, what have you done throughout your career and what are you doing currently?

Speaker 2: 0:21

My name is Dylan Patel and I'm the chief analyst of Semi-Analysis. I founded this firm a handful of years ago and we mainly focus on semiconductors and AI. I myself have historically worked in data science, AI as well, semi, and then on the team we have some folks that are also in all the way from experience at ASML through to experience at design firms and things like that, and a couple of people who worked at funds as well historically, right. There's five of us. We're sort of international. There's two of us in the US, one person in Japan, one person in Taiwan and one person in Singapore, and so we mostly focus on semi and AI from a very technical and market perspective right, Because we have all the way from technical folks to people who have worked at hedge funds, and so we forecast supply and volumes and units. But we also flow through to what's the technical reasons why someone is gaining market share or losing market share.

Speaker 2: 1:17

What's happening on the technology side? We also do some technology consulting right, Like should a company do this or that? Right? What should they do on roadmap? What should they do on a lot of different things like that right. So that's sort of the stuff we generally focus on, and we work with firms both in the semi and semiconductor and AI space, do a little bit of venture investing, but mostly a lot of data research services as well. And then there's a newsletter that sort of popped off, maybe about a year ago that we started, and that's fun as well.

Speaker 1: 1:43

Yeah, congrats on the success there. I see like 70,000 plus people are getting your blast there. Ok, so look into it. So you said you really kind of started this effort a couple of years ago. What made you get into this side of the tech world in terms of just putting content out, because I think the AI narrative is really only one that took off in a big way around chat GPT Sounds like you were involved in this way before that.

Speaker 2: 2:08

Yeah, yes, I mean funnily enough, right, like I put out a blog on the cost of training GPT-3 the same day that chat GPT launched and that was just a coincidence. I had no forewarning that it was going to launch. I had a friend at OpenAI that were excited, but they didn't have anything like. I didn't have any specific forewarning that chat GPT was going to launch and be as big as it was, and so it just so happened. As a coincidence, I published a blog on how much does it cost to train GPT-3, which was a prior generation before GPT-3.5, which was for chat GPT release or GPT-4. But as far as what was the impetus to get involved in the content, public content creation space? Because I was doing a bit of consulting. Before I was creating content publicly, we were doing research and stuff as a firm, but before we built the newsletter, it was just really boredom during COVID.

Speaker 2: 2:55

It's like seeing news articles and being like that's just wrong, no, stop, like this is the truth, is this Right? And it was like this innate need to want to correct people on the internet which I think we all have. We get sucked into that comment section and argue with someone, and it was like there's a more efficient way to do this Right. Especially, the impetus was really, was someone I respect like getting misled by one of these pieces and I was like, okay, I really have to put something out that's like tells the truth, right, because people who are smarter getting misled because they just don't know, right, and so that's sort of what was the impetus and a lot of the focus was, you know, those on semi and AI, right, like we talked about AI, chip startups, et cetera. You know well, before chat, gpt and all that. We mean that's where consulting is focused on as well.

Speaker 1: 3:35

Yeah, you mentioned that you've got a couple of people in Singapore and Taiwan. I'm curious how did you build out that network of people that are involved at seminalescom and maybe how important is that perspective having researchers, analysts and other countries that are maybe also at the forefront of this?

Speaker 2: 3:51

So I think that the semiconductor supply chain is one of the, like, most difficult to understand in the world, right, I mean collectively as an industry. Semiconductors spend more on R and D than any other industry, including pharmaceuticals, right. If you actually sum out, like all the equipment companies that are for making chips, through to design and analog and so on and so forth, there's more R and D spent in semi than anywhere else, even pharmaceuticals, right and so. And then CAPX wise, the same right. The electronics manufacturing sort of industry as a whole is is massive on CAPX, right, so there is a massive amount of stuff going on.

Speaker 2: 4:26

But people in the West, really, like you know, especially like when you look at a classic analyst, right, like they're like, hey, yeah, I'm a chips analyst, right, okay, well, I focus. That means they focus on, you know, in the US, right, then they focus on NVIDIA and AMD and Intel and various firms in the US, but they don't. Or Qualcomm, right, they don't go through to like, well, hey, like, what's this firm that they work with here and what's this random like Japanese company? Who's a tool supplier to this company? Who's who makes chips for this company? Right, like you have to float through the whole supply chain, right, and so you know, because a lot of the best information you eat oh, a lot of it's public is people don't go through to it, right, it's like from a random conference, right, it's an engineering conference for this, right? You know, one of my favorite conferences is for photoresist, right? So Myron's in Japan and he's.

Speaker 2: 5:07

There's a conference called. There's a photoresist conference, right? Photoresist is basically the stuff you put on a wafer, a silicon wafer, before you put it in a lithography tool, right? I think a lot of people have heard of ASML and their lithography tools, but there's a specific chemical that you have to put on the wafer before you put on it, right, and this chemical industry is worth billions of dollars alone, right? You know there's one of the major companies there is being taken private right now for $7 billion, right, and they only have like 30% markets here, right, and that's pretty much all they do is the photoresist and there's a conference for that, right, or it's literally 4,000 Japanese people, right, because this industry is dominated by Japan, this photoresist industry. And I was legit the only person who was brown, right, there were a handful of, like Dutch guys and, like you know, intel dudes, right, but I was the only brown person at the conference.

Speaker 2: 5:54

I was like, what you know, it's crazy how insular parts of the supply chain are, right, and it's sort of these people do. They focus on the exact mixture of chemicals and resin and all these sorts of things for photoresist, right, columners that need to be embedded, and every generation it changes. And there's, you know, and who knows about this? Right, there's people in the abstraction stack who know about it. One layer up, right, the lithography. People know about the photoresist and everyone else doesn't really care, right, the person designing a chip doesn't really care, doesn't matter to them, right, and so there's such insularness in the supply chain, and so it's really important that people sort of look top to bottom, right, like, hey, what's going on at this layer? Because that actually tells you about. Yes, those people and those people in the photoresist world don't know anything about designing chips, right, like, they know they need to be smaller and they know that, like, hey, like, well, my photoresist needs to be able to help you print lines that are four nanometers smaller, right, you know, than the last generation. But like, and where are all the technical challenges? They don't actually know, like, what's going on with the chips, right. Like. It's like kind of it's up and down the supply chain both ways, right.

Speaker 2: 6:56

So you go through and you look at there's industries like that in all three of these countries, right, japan really dominates certain chemicals, certain types of equipment, just like the Dutch dominates certain types of equipment, just like the US dominates certain types of equipment.

Speaker 2: 7:09

Us really dominates design, but then manufacturing is dominated by Taiwan, right. Like TSMC memory is dominated by Korea, right. You know, singapore has a few niche industries as well, right. So when you flow through to all these industries, it's a very sparse industry. It really does. You know, you do have to be embedded within each of these industries, right. To really fully get a, to get a good picture of the whole supply chain at the level of depth that we need right Now. Now, investors don't need that level of depth always, right, like it's, it's certain types of folks and certain catalysts and certain things that need to be at that level of depth. But generally, you know, you don't need to know anything about how, how the chip is printed to know that AI chips are selling like crazy and also like there's an interesting like things going on there.

Speaker 1: 7:53

Is there any consideration to sort of yeah, since you mentioned all these different countries that you know are involved up and down the supply chain that you know you don't hear about on the CBCs, the world, or is there any consideration to geopolitical risks, kind of throwing things off in the supply chain when it comes to AI, or are these things still largely isolated?

Speaker 2: 8:09

Yeah, I would say there's the whole like US, china, you know geopolitical right, invasion risk, all these sorts of things, right, I mean, I think it's very clear that TSMC does have a lower multiple than many other businesses that are in the semiconductor industry. And you ask investors and there's like no. Like you know, you talk to some folks about you know, these massive funds and they're like, yeah, there is a bit of a valuation discount because there's a Taiwan invasion risk at some point. Right, maybe not you know super soon, but 2027 or 2030, right, like, there's enough of a risk that you know Navel, like commanders, will talk about it. Right, so there is a bit of a valuation discount. Now, when people say, hey, is there a huge geopolitical risk and 70 acres is like absolutely massive right, and one of my favorite things to do about this is, like you know, there's the trade spat between the US and China. Right, you know, block this block that you know the Dutch and Japanese are joining in with the US on blocking China from certain access to certain technologies and industries and vice versa. Right, german China is blocking like Gallium and a few things. But what's funny, right, and I wrote an article like you know, a couple maybe, like a year and a half ago or two years ago.

Speaker 2: 9:13

Austria, right, little old Austria in Europe, right, you'd think. Well, what do they have to do with the semiconductor supply chain? Right, and Europe. We all know about the Dutch right, asml. But what about Austria? Well, it turns out there's two different companies there that have market share north of 80% and north of 90% even in certain types of technology, right, but if you look at, like the leading edge, right, so you know.

Speaker 2: 9:34

Another funny thing is right, like there's, you know, again, ASML. They make EUV right, extreme ultraviolet lithography, right, the most advanced form of lithography. I think that's like it's really important. But it turns out there's this thing that you have to put inside the tool for a chip design. Basically, it's called a mask, right, you can think of it as a stencil, right? You basically put the stencil inside this $150, $200 million tool from the Dutch and then this mask has the chip design on it, right, and you cycle through that many times to make chips right, the stencil right, and so it's called a mask. And Austria has a company who makes a specific tool that writes these masks right, and they have a 90, 80 plus percent share. But on EUV, right On seven nanometer, on five nanometer, on three nanometer, they have a hundred percent share, right.

Speaker 2: 10:21

So it's like, okay, well, austria actually could shut down the entire semiconductor supply chain, right. It's kind of there's a lot of like independencies, like that. There's another company called EV Group. They're private and they also have a very high share, similarly extremely high share, and certain critical technologies and the whole semiconductor industry, parts of it, would shut down if it weren't for them, right. And so it's sort of you know you talk about is there a geopolitical risk with all these niches? Absolutely right, and as far as I know, there's no one trying to change it like entirely right.

Speaker 2: 10:52

And I don't know if it's even possible to go entirely right, because the universities in Austria focus on just that one little vertical, the semiconductor industry. Right, that mask writing, right, using e-beams, electron beams, to write these masks, which are stencils, right, effectively making stencils, electron beam, and the universities, they're only focused on that and the company's focused on that for decades. It's like, how would you build this institutional knowledge in America or in, you know, in China or wherever else, overnight? You couldn't, right, you couldn't build it overnight, you couldn't even build it, like, you know, if you gave it five years and billions of dollars right For certain aspects of the supply chain, right, it would literally take people trying stuff for decades to pick up all this institutional knowledge right, or just hiring those people right Like, and bringing them over right Like.

Speaker 2: 11:40

There's no really effective way to fix this geopolitical risk entirely right Now. There's ways to mitigate it, right, by having production, you know, in your country. So maybe you don't have necessarily every piece of equipment, but once you have equipment installed it's sort of a bit more safer, right, at least in terms of hey, there's no, you know, we can still make chips for a while. We can't expand production, maybe, but we can still make them for a while. So, sort of, there is that geopolitical risk there.

Speaker 1: 12:04

If you have the. Since you mentioned private companies, there are obviously plenty of public companies that are part of that supply chain, that value chain, but I don't really hear too much about sort of price action and some of the public stocks, whether they're in the US or internationally, that are involved, for example, in some of the things that NVIDIA is doing. Any observations on the narrative, maybe not focusing on those supply public stock companies that you know there's some maybe better opportunities there because people just not focusing on that part of the story.

Speaker 2: 12:31

Sure, sure. So I think there's, you know, sort of the upstream and downstream suppliers of, say, you know, specifically, nvidia, right are really important and the AI sort of stack, right, and so there's a lot of companies that people are really fully focusing on, right. So, for example, nvidia makes the GPUs but, like, who do they buy stuff from? Right, well, it turns out that people they buy stuff from are not benefiting that massively, right, because if you think about it, right, well, it's a sort of pain in the picture, right. So Meta has talked a lot about and Microsoft both have talked a lot about how much NVIDIA chips they're gonna buy. Right, historically, they used to spend about $20 to $30 billion on data centers. Right, Including the data center itself physically, the chips inside of it, et cetera, et cetera, et cetera, right, the power infrastructure, et cetera. Now they're talking about spending way more, right, microsoft's gonna spend, you know, $50 billion or even more right on chips this year. But you know, but it goes from, you know, hey, in 2022, how much did I spend on AI chips? You know a couple billion, right. How much are you gonna spend in 2024, based on the indications that they've said? Tens of billions, right, but what about the other types of chips? Actually, a lot of them I've reduced purchasing on, right, you know, for example, intel CP's I don't purchase as many of those, and you know you go down the list. There's a lot of stuff that I'm purchasing a lot less right, certain types of memory and et cetera. Right, really, because I'm increasing spending but I'm also shifting some of my existing spending toward to AI. And so then, okay, a lot of this money flows to NVIDIA.

Speaker 2: 13:57

But NVIDIA has these massive margins, right, they have these margins that are 80% right On these data center GPUs, right, gross margins, basically, roughly, you know, effectively a 5X markup on the cost of manufacturing to them, right. And then they don't manufacture themselves, right, they design chips, design systems and get it manufactured by others, right, like TSMC, like SK, hynex, like Fabernet, you know, go down the list. And there's all these contract manufacturers or chip firms that manufacture for them. And you think about it with like, okay, well, nvidia has $10 billion of revenue, but because their margins are 85%, right, 5x markup, it turns out they only spread $2 billion downshifts, right, because the chips are so high margin, whereas historically, right, when I was buying data center chips, right, my suppliers had 50, 60% margins right. So 2X markup right. So if I spent $5 billion then I had $2.5 billion downstream manufacturing revenue.

Speaker 2: 14:56

So it turns out, even though NVIDIA is sort of soaring and there's huge orders from at least the biggest companies and many other companies, but especially like Microsoft, meta, as the biggest sort of individual customers, even though there's humongous orders and sales, the downstream supply chain is not always benefiting as much.

Speaker 2: 15:13

Now there are pockets of it right for certain types of chips that are benefiting massively. Right, because it's a shift of what type of chips. But overall the pie, even though it could double, right from five billion to ten billion dollars because of the margin, nvidia just eats a lot of that margin right, and so the actual number of chips being purchased is, or you know the sort of manufacturing volumes, sort of stay roughly flat. But in reality there are some parts of supply chain that are sort of eating, hurting really that right, and there's some parts of the supply chain that are benefiting massively, and so when we think about that, it's like companies like Marvell and Fabernet and SK, hynex and you know so many more coherent, and you know the list could go on and on. It depends on which part of the supply chain we're going to talk about, but we could sort of break down an AI server and talk about companies in each segment.

Speaker 1: 16:32

So the margin point, I think, is obviously a powerful one. To come to Nvidia. But you know one thing I know, at least from business school, and just logic is that over time margins get reduced because of competition. So that led to a discussion around moat, right, when it comes to different semi-companies and video, obviously, in particular, how sustainable are these types of margins? I mean, that is the bare case I often hear, at least on mainstream media, it's like all right, competitors come in, that's what ends up creating lower margins and things like that. Talk through that, because that seems to be a critical component of the long-term argument for some of these stocks.

Speaker 2: 17:06

Sure, sure. So I think that the margin component is very sorry. The moat of Nvidia is very strong today, right, like there is absolutely no contest in terms of, hey, if I were to buy a chip today, have it installed. What would I want if I wanted to train a large-language mod, if I wanted to run a large-language mod, if I wanted to generate images? You know all these sorts of use cases that are popping up, right, and there's no question there that Nvidia has the strongest, but you know, that does change over time, right, and a big part of this is, hey, who are their customers? Who are their biggest customers? Well, microsoft is, meta is, amazon is, google is, you know, these are their biggest customers, right? Alibaba, tencent, of course, as well, but you know, I guess all of these companies are designing their own AI chips as well, right, and that's the real crux of it, right. And then, furthermore, amd is coming with their own chips as well. They're finding some success, and so, when you add those things together, it's like, oh my gosh, like there is a big risk here of Nvidia being sort of competed away, right, if your biggest customers are designing their own chips, and then you know, sort of your historical competitor, amd, in the gaming space, who's always sort of had 30% share, whereas you had 70% share, sort of, in the gaming space, right Now makes these GPUs for AI and right now Nvidia has 99% share right for AI GPUs. Right, like, is it possible that AMD, even if they get only to 30%, right, that really does make a lot more competition. And so this is the big risk that everyone's worried about. And then, and to break it down like there's, like for the few reasons why this is a legitimate risk, there's also, you know, I think on the surface it sounds like a massive risk there's a few reasons why it's not like, maybe as as scary as it seems on the surface to right. And so so, starting off with like, why is this? Why does Nvidia still have a moat in a year or two, right, you know, once these chips are starting coming out, well, it turns out.

Speaker 2: 18:58

You know, google, for example, right, google's had AI chips longer than Nvidia has had AI specific chips, right, which is pretty crazy to say. But, yeah, they've been designing AI chips for longer than Nvidia has been and yet, right, and yet Google still purchases a lot of Nvidia, right, why? Because Google is as good as they are at design, as good as they are at you know. And then their partner right, they work with Broadcom a lot and Broadcom's been a stellar stock, of course, and it will continue to be. Because of this Google AI chip business, plus the VMware sort of savings and earnings are squeezing out of that. But the you know, when you look at Google and Broadcom, their design partners right, they're making this chip but turns out it's great, but not for everything, right? There's still a lot of use cases where I still need GPU, right, because I'm simply not as good at designing chips, right.

Speaker 2: 19:45

The other aspect is the software, right, google maybe internally can use more of their own chips, but externally, our customers still want GPUs. They don't want as many of their chips. Right, because the software is a lot easier, right. So there's the software component and there's the chip design component. The chip design component breaks down into two pieces as well. Right, there's networking, which is connecting chips together, and then there's the design of the chip itself, right, and so on. Both of those points, like for the compute side, right. So on both of those points there's also a sort of differentiation there and long story short is that, like, hey, google, even though they're on their fifth generation of AI chip? Right, they're on the fifth generation. They make two different versions of the fifth generation. Right?

Speaker 2: 20:27

They still purchase billions of dollars of invidio Right Now, and albeit, google, despite being a company it's roughly the size of, you know, quote, unquote, the other meta and Microsoft. They buy a lot less than Microsoft and meta for shirts when they buy their own right, they're buying about $10 billion of their own chip, right, and then they're buying billions of dollars of invidio chip as well. So there is definitely market share that's been eaten by Google, right? And so the question is then what happens when Amazon's chip gets good enough? Right, because Amazon's chip is garbage today. Right, the AI chip very bad. You know what happens when Microsoft's chip gets good enough, right? They just released it or announced it. It's okay, but it's nowhere near good enough, right? They're not even going to buy a billion dollars worth of it, right? What about the meta chip? Likewise with meta, right, they're not even buying a billion dollars worth of it because it's not that great. Right?

Speaker 2: 21:14

It's hard to design a really good chip, right, you know, especially out of the gate.

Speaker 2: 21:16

Right, because, again, google has been doing it for five generations with Broadcom, with a partner, and they're still not objectively, you know, even as good as invidio in many regards, right, so sort of you flow through to that and it's like, okay, then this is. You know, there is a legitimate design thing there, right, in terms of designing a chip. And then there's software, right, much in the way that, like Intel, sort of won quote, unquote data center, because everyone on the customer side, like like desktops and things like that, were buying Intel chips, right, and then developing on them, is the same way with Nvidia, right, everyone is developing on Nvidia chips. Today All the existing software is on Nvidia chips. So the portability of that software to other chips is an important concern, right, it's a major concern and there are a lot of efforts around that, such as PyTorch 2.0, openai, triton, things like that. That makes software more portable. But it's still, you know, a big most for Nvidia going forward. So we can dive into that a bit more. The software, put it, I think, is very important.

Speaker 1: 22:18

Just reset the room for their meeting. 20, 25 minutes here. Please make sure you follow, tell, as you know very knowledgeable about what it comes to this side of the world. If any of you want to come up and ask questions, click that bottom left micro-request button and, as always, this will be in podcasts under Leedlag Live. Okay now, so a lot that you cover there. There is this other element which is aside from you know, microsoft's and of the world and others doing their own chips. There's also the argument that a lot of this demand is being pulled forward right with this kind of manic ordering on the Nvidia side. Any proof to that from your perspective, that maybe things have just gotten too extreme too quickly on that end of things.

Speaker 2: 22:59

So the funny thing about the sort of overordering narrative is that semiconductors do this every time, right, and people and I'll explain what I mean right, the semiconductor industry always overorders, always goes through massive booms and busts, right, so most recently, of course, this AI boom. Before that, in 2023, and especially 2022, there was a chip glut. Right, but then and you can still see it in certain companies like analog devices and Texas Instruments and Microchip when they report right, all of those companies are, you know, tens of billions of dollars, hundreds of billions of dollars companies. Right, and all those companies still have, you know, chip gluts right now. Right, but when you go back historically, you know this has happened a lot. Right, the semiconductor industry, because the supply chain is the most complex in the world, has humongous lead times to increase production. Right, and it's very hard to judge demand. Right, it's really hard to gauge what it is right, and every single time, there's this concept of overordering, if you will. Right, where chip companies will build and build and you have these orders that come in from firms. And every single time, lead times extend. Right, lead times for chips grow, they get so long and users are ordering so many of them, right, they're like hey, yeah, we need this chip for that purpose, you know, and anyways, the long story short is, they end up getting far more than they needed to, right, and you end up with a glut at the end of it.

Speaker 2: 24:26

And so the question is that going to happen with AI chips? Yes, it absolutely will, right, the supply chain is growing capacity so fast, right? It's like do you think Nvidia will have $100 billion of revenue this quarter? Very possible. Do you think Nvidia will have $200 billion of revenue next year? Or not this quarter, sorry, this year? Do you think Nvidia will have $200 billion of revenue next year? Like, no, probably not. Right, like you know, at least $100 billion this year is possible, but $200 billion next year is, like I just don't see how. Right, but that's what the supply chain is building for. So there will be a glut, right, like that's without a doubt. And so then the other question is sort of when and what's the sort of rebound? Right? Certainly, people are in a manic craze, right?

Speaker 2: 25:07

Well, if I will afford, if I, you know, let's just call it simple right? If I want one GPU and I know the my part, my, who I'm buying it from can buy, can build 10 of them, right, and maybe there's 13 buyers, right, and everybody wants one. Well then, if I order one, I'm possible I won't get it. But if I order two, right, if I say, hey, I need two, then I'll definitely get one, right.

Speaker 2: 25:29

And then everyone so sort of a bit of game theory sense, right, where you kind of double water so that you're you get your order in right, sort of, and you get the amount that you need sooner, so sort of. There's this whole game theory sense around it as well, where sort of hey, like, if I double order, then I'll get my chip sooner. And then there's also a bit of exuberance, right. It's like, well, I don't actually know how many I need, but let me order two anyways, right, maybe I only need one, right, I don't know, but I certainly need them now. But I don't know how many I need, because it's a brand new thing, right. So sort of. There's also that aspect of it that people sort of have to account for as well.

Speaker 1: 26:00

So that's an interesting point, right, that, even if you're going to make estimates that future demand is going to look somewhat similar to today's demand, the more time that goes on there's more clarity to your point about how many of these things you actually need, which, if it's a lot lower in reality than what people or companies are currently expecting, then alone is a deceleration of revenue.

Speaker 2: 26:22

Yeah, exactly, and I don't think that happens like this week, by the way, or like this the first half of this year, right, I think the party will go longer than some of the bearst thing, but also, like for the bulls, right, it won't go as long as you think either, right, so a bit of both ends.

Speaker 1: 26:36

Outside of NVIDIA and all the sort of major headline plays on this semi-M that you hear about, because the stock prices have gone ballistic. I'm assuming you're always looking at other companies again touched on the supply chain value side. What are some of the more interesting or exciting things that maybe are getting as much play that you're focusing on?

Speaker 2: 26:53

Yes, I think a lot of the stuff that I'm focusing on is a lot of the downstream and upstream suppliers, rather than necessarily like, of course, like we study NVIDIA and the supply chain a lot, but there it is sort of you know, when you go a couple layers deeper, right, where is capacity being expanded and where is capacity oversupplied and where could that be leveraged and what can companies do to better adapt to the current times and what's going to happen once AI chip demand cools down. And so it's hard to disentangle because, at the end of the day, these AI chip companies are also selling many other chips. So, for example, I think Marvell is a really interesting company right now. Right, marvell, yes, they are selling AI chips. You know they're. They're going to ramp their premium and infrastructure for Amazon, but that's not their main business, right? Their main business is, historically has been storage chips. It's 5G telecommunications chips for infrastructure, right, it's networking chips for telecommunications as well as now, ai, right. And so their companies like Marvell, have a very, like you know, interesting viewpoint, right, and that, oh wow. So certain aspects of their business are doing horrible right and the business is doing awesome right, that's for sure. Everyone knows that.

Speaker 2: 28:13

But what happens, like, when the AI sort of craze, kate cools down. If it cools down, right, well, turns out, like you know, telecom and storage will come back right, and those, historically, have made over half of the business for Marvell. And you know, at least in like the second half of this year, estimates for us and you know, are, hey, they're much less than half of the business, right, so sort of flow through and it's very clear that companies that maybe are they selling only to AI? Do I want only AI exposure? That could be a possibility, right, those companies are going up more, right? Do I want AI plus, you know, other business exposure?

Speaker 2: 28:44

So, when AI cools down, right, like you know, what exactly do you want as an investor is important, but companies like Marvell are super interesting. It's like, what's going on in telecom? It's like, well, no one cares, right, right now, right, and in fact, right, like you know, there's a lot of if it's not serving AI, right, and so if you look at a lot of companies, actually they're actually reallocating engineers, right, they're like, hey, like, if it's not AI, we want to reallocate them. Right, well, then we'll move you back when we need to, but today, right, like we said only focus on AI. Why would we focus on anything else? Right, for those engineers, right? So sort of a bit of a catch-22 there, right, and exactly what you're looking for. Then there's companies that have no AI exposure at all and they're doing work, right, you know Texas Instruments and Microchip and In-A-Log devices. The stock prices kind of are down stop, but the revenue and their guidances are horrendous, right, and so you know what's going on. There is also really important to watch and look at.

Speaker 1: 29:36

Are there any energies in terms of the companies that are pulling away on the AI manufacturing side, with those that are not, like just mentioned Texas Instruments, you know? Could there be a maybe some of these non-AI chip manufacturers become acquisition targets by the AI chip manufacturers?

Speaker 2: 29:53

So for companies as large as Texas Instruments no, texas Instruments is far too large to ever be acquired and they do way too much military business. Same with Microchip, right. But I think there's a lot of stuff in the small cap and mid cap side that could happen to right. So, for example, you know it's very possible that Coherent, which is a company that makes optics for traditionally telecom, but now they're really booming in AI, right you know, because AI is such a booming category, so you know there's possibilities that companies like that could get acquired right. And so if there are about $7 billion right now, so imagine like, hey, like an acquisition would be maybe in the $10, $12 billion range, right? So it's not a you know otherworldly to see an acquisition like that happening, right.

Speaker 1: 31:06

Yeah, I clearly think this will be too large, but my point is that it just it gets me intrigued to think about consolidation, If it's sort of if AI is now becoming the difference between sort of the haves and half-nots in the semi-industry.

Speaker 2: 31:16

Yeah, absolutely so. We've actually seen, like you know, for example, you know a lot of this Nvidia's lead is due to their acquisition of Melanox in 2019, right, and specifically Melanox being a big leader in the networking space. Right Now it's called Nvidia networking, so you know a lot of. There's a lot of sort of hunting for networking assets by their areas.

Speaker 2: 31:37

Right, rodcom and Morvelle were the other two leaders in networking and, of course, Morvelle acquired a handful of companies over the last few years Infi, inovium, cavium, etc. They've rolled up all these sorts of companies to build a very legitimate, like networking contender, which is why they're able to do some of these AI chips. Right, and likewise, rodcom has had this partnership with Google because they've had this historical networking, you know sort of business, right, and so there's, you know, companies like Max Linear who are really in the dumps, right, like I wouldn't even recommend like attaching it because of you know some of the dynamics that are going on, but like they could be. You know they have some good networking assets but someone want to buy them potentially, right, like, so there's there are companies like that that are really in the dumps. That you know, hey, like, they have networking assets, they don't have enough to really launch into AI, but someone who has some or some of those other aspects of the AI supply chain could right, and so that's sort of always the question as well.

Speaker 1: 32:32

What are some of the threats when it comes to demand for AI chips manufacturing side of things? So you know there's a recession possibility, there's competition right, but what are some of the real threats from your perspective? That, andy, overbuilding right and over demand?

Speaker 2: 32:47

that you mentioned.

Speaker 1: 32:47

What are some of the sort of major threats that would worry investors? Take the mega trend out of it right, Just in terms of where things are going, and stock price.

Speaker 2: 32:56

Yes, I mean, in the 90s it was very clear the mega trend of the internet and all this sort of stuff. Right, like people fantasized about everything that we're seeing today. Right, you know, massive automation and people's whole life being lived through the internet and all these sorts of things. But it's like, well, it took decades, right. And likewise, right Like the big risk with AI, you know, you know chips and stocks. There is, okay, there's going to be a period of exuberant building. There's going to be a period of like, hey, like revenue, you know could do this or that on the end application. But at the end of the day, right, if Nvidia sells $100 billion worth of chips, their customers, you know, depending on how long they depreciated over and all that sort of stuff, right, they depreciate over six years. That means, and then they have a margin target of, say, 50%. Right, that means they need they margin target of 50%, 100 billion, they need. You know, if they buy 100 billion dollars of chips, they need like $250 billion of revenue. Right, because there's other costs there as well. Right, they need $250 billion of revenue and they need that over the six years. Right, probably more than 250,. Right, because electricity and all these and development and all these other things, right, so, and their own costs, right so, really they need, you know, a multiple of Nvidia's revenue, but that multiple of Nvidia's revenue has not popped up anywhere, right? In fact, the only company making revenue on AI, right More or less, is Microsoft and Meta, right, microsoft, because they're selling, you know, the opening IAPIs and they're doing, you know, chat, gpt and co-pilot. And then Meta because they have a humongous amount of advertising spending being driven to them and being able to make ads much better, right, like hey, you know, you know I sell ads effectively. That's what Meta's business is for companies to advertise to my users. But you know what, if I could make a you know, target it explicitly at that one customer, right, with generative AI, right, well then, that makes the ad much better, right, and so there's money that Meta's making there and they're going to continue to make money there.

Speaker 2: 34:43

But a lot of these AI use cases, you know how long is the revenue going to take, you know so it's the real big threat is revenue, right, and so when you look at, like hey, in the 90s we built so much fiber that even in like 2009, google was still buying fiber. That was put in the ground in the 90s, right, that was just like sitting their dormant, right. And so you know, likewise with AI, right, maybe fiber is not the best analogy because fiber lasts a lot longer than AI chips. You can upgrade the sort of the modulators and such on each end, but you know, and encode data, throw it at a higher signal rate and so on and so forth. But with AI, right, like it's.

Speaker 2: 35:18

Like, yeah, people are going to spend a ton, but you know, if the revenue doesn't come, they can't continue to spend a ton. Certainly people can make the bet, but you know that's the real big risk there. Right, is okay, if Microsoft spends, you know, tens of billions this year to continue their growth. Right, at some point it's going to be $100 billion of spend, right, just from Microsoft, you know, at some point. Right, and you know five, ten years, right, if the growth rate, you know even, you know could be even next year if the growth rate continued. But of course not. Right, of course not, but at the very least in a few years, right, if you want the growth to be even solid. But that requires massive revenue and we haven't seen that yet. Right, and that's the biggest risk.

Speaker 1: 35:56

At what point do you think I mean it's hard to, I guess, know this, but at what point do you think that becomes sort of the boogeyman, right, there's some time limit under which it's like, all right, well, all these chips are being ordered, all this talk about AI is there, but it's just not showing up from a monetized, monetizable standpoint with all these mega cap companies, let alone the productivity gains that should be helping every other company. Because that's really been sort of my, my own personal gripe with this AI narratives. It's that you are not seeing at least the stock side react from a discounting perspective of the future way that you would think they should, if AI is supposed to be so incredible, for it's sort of the next inductor revolutions in the US.

Speaker 2: 36:37

Well, the funny thing is, when people talk about the valuations, they keep focusing on the historical or they keep mispricing the forward. But in reality, what are the biggest invinias making the most money from AI? Well, it turns out they're only trading at about 25x EPS for this year, right, based on our estimates. Right, broadcom is making the second most from AI, especially because they're Google partnership. Right, turns out, they're only trading at like 25x EPS, right, or even a little bit lower than Nvidia actually. So it's, you know, the pricing is kind of. It's kind of you know, you could argue either way, right, you could say, hey, you know, if this is as big as it's going to be, why are these guys only trading at that much, right? And the other side is, you know, hey, this is you know, you know, yeah, they're only trading at that much because people don't expect it to continue to grow this fast or grow much, you know, in the future, right? So, sort of there is a tale of both cities in regards to the valuation of, you know, high flying stocks.

Speaker 1: 37:34

Right. But what about the companies that should be benefiting from AI, right? Like to me, it's more the disconnects, right? So, again, if it's about productivity enhancement, then you should see, I think you can go on with this. I think at these three, we can argue that AI in the perfect world should make the margins for every company higher, right, because it's in high. You see, I think that to me is more the disconnect. You've got the narrative around AI. You've got clearly the demand is there on the chip side. Fundamentally, you know we value it as a different animal, but it doesn't seem to be reflected in those that you actually benefit from. The output is my point.

Speaker 2: 38:08

Yeah, I mean that's a good point. But there's also the sort of like point of like, well, you know, did you know? Did G&E benefit hugely from the internet? Well, it's like, yeah, sure, I'm sure they benefited, but they lost so much, right, because their core business fell apart. You know, in many ways, right, they had to sell off assets and this and that, right, did IBM benefit much from?

Speaker 2: 38:27

You know, you know cloud computing. Well, you know they have a really crappy cloud computing business. It sells some, right. But it's like if you go back to the 90s and you told someone about cloud computing, they'd be like, course, ibm, right, like it's like. So. It's like there's always that aspect of it too and that like, just because a company can benefit from it, should benefit from it, doesn't mean a new player doesn't come in. And disruptions like this are always like how new companies come in and how a company could sort of, you know, be displaced in any capacity. Right, and certainly like S&P is up, but not like crazy, for you know, many companies who you know, like you said, should have productivity boops.

Speaker 1: 39:05

I'm curious for your own investing portfolio. Right, because I'm going to make the assumption that you're like others. Right, you're putting some money to work in public companies as well. You mentioned the ventures. I'm going to make the assumption you're not all in on NVIDIA, but if you are, god bless. But what do you do from a sort of investing allocation perspective? Are there, you take the approach that this is just a boom across the entire industry, so by a broad base like ETF, or are you going very concentrated in individual positions?

Speaker 2: 39:31

Yeah, so historically I used to run extremely concentrated. Actually, at one point I think 80% of my portfolio was NVIDIA, right, so just you know not to be too crazy, but that one point it was right. But over time you may concentrate at bets when you have extreme conviction around an idea, right, I think individual investors should either, you know, buy the buy, you know an ETF, and be like very spread out just buy the S&P 500, or they should make extremely concentrated bets, right, like I don't think individual investors who are not full-time investors should ever care about you know, spreading their money like very broadly, right, either don't pick stocks or pick you know very few, right, because that's how much time you have to make a convicted bet and leave, like buying like 20 stocks or you know 15 stocks to sort of you know, institutional investors. But as far as, like, my personal portfolio, on one one, I did stop, I did sell actually all my semiconductor stocks on one two, actually one two, that whole first week of January, because we're sort of, you know, gets into like sort of compliance related issues because we hire, you know we're doing more and more work with certain funds and they don't want you to trade when they trade off of your information, right, and all that kind of stuff. So there's these sorts of things, and so that's why I've been doing a lot more venture investment more recently.

Speaker 2: 40:46

But still, the semiconductor in the eye space is just not public market firms, but generally, like, I think the one area that people sort of neglect way too much is the equipment space. Right, so you look at semiconductor companies and they're very boom bust and they're capex light businesses, but they're also like high margin right. Think about NVIDIAs and things like that. But the best businesses of the 2010s and they continue to perform like crazy this decade are the suppliers to the manufacturing space. Right, so these are called semiconductor capital equipment firms. They have crazy cash flow right. They have some capex, but not a ton.

Speaker 2: 41:20

They make the equipment for the chip right, so they don't have to buy the equipment. They sell it right. They're very R&D intensive right. 20, 30% of the revenue goes straight into R&D. They spend tens of billions on R&D collectively. Right, and they return almost all their cash to the shareholder in the form of either buybacks or dividend. Right, because they don't need a ton of cap and generally they've grown pretty much every two years, right. So there is a bit of ebb and flow but if you average out revenue for two years and two year increments, they grow right and of course there is cyclicality in the semiconductor business. But companies like Applied Materials and Land Research and KLA and ASML these sorts of companies are by far the best investments that I think that individual investors sort of have ignored right and a lot of funds picked them up but a lot of individual investors have ignored them despite their stratospheric performance and continued performance.

Speaker 1: 42:11

Dylan, for those who want to track more your thoughts, more your work and just engage with you. Again, I give you a lot of credit, built all the sub-stack following and obviously very knowledgeable with you and your team. Putting that content out there but probably people are aware they should go to and maybe just sort of parting thoughts on what you think people get wrong since you mentioned this point about kind of wanting to correct the media what people most get wrong when it comes to this space.

Speaker 2: 42:34

I think one of the things so I said you can follow me on Twitter, dylan5gdp, or the team on semianalysiscom, where we publish some of our research, especially the newsletter. Right, the newsletter is very timely and aware and everyone gets that. At the same time, right Time stopped and moved just on our newsletter, which is pretty cool, and people have had to refute us on our niche calls or trying to argue against us, which is really cool because we've published research. That's like technical as well as market-related, right. Anyways, in terms of what people are getting wrong now is, I think there's a whole narrative again that like this like most recently, I think Financial Times posted a article about this one credit analyst I can't remember if it was JP or Goldman saying like, hey, nvidia is scamming people, right, or not scamming people, but like more or less like, look, nvidia is the stock is doing really well, but look at what they're doing in the venture side. Basically, they're arguing, hey, they're investing in their customers, right On the venture side, but it's like people kind of are ignoring, like, hey, like CoreWeave, right, they got to invest from NVIDIA, and perplexity, and you just go down the list, right, there's so many companies that have gotten investments from NVIDIA.

Speaker 2: 43:43

But those checks are all very small. Right, it's not like NVIDIA is investing hundreds of millions of dollars in a company who turns around and buys hundreds of millions of dollars of GPUs. Right, well, coreweave is buying billions of dollars of GPUs, but their investment from NVIDIA is not even $100 million. Right, look like yes, nvidia invested, but they didn't invest like the, they didn't induce the demand by investing, right, which is like sort of one of the narratives that is sort of starting to pop up in the media.

Speaker 1: 44:08

Everybody, please again give a follow to Dylan. I'm very good, I appreciate it. Cheers.