Semiconductor Industry Outlook Panel — CASPA 2026 CEO Summit

[00:00:00] Speaker 1: Let's now step back and look at the bigger picture starting with demand and then supply. On the demand side, AI revenue is growing so fast but the pace is starting to moderate. The combined AI revenue today are only a fraction of what the industry is spending. So the market is now looking at every revenue number under a microscope. And on the supply side, capacity is sold out for years ahead and pricing remains firm. From a semiconductor analyst perspective, how should we read these two signals and who will capture the values along the way? To guide us, we have two leading voices in semiconductors. Dylan Patel is the founder and CEO of Semi-Analysis, one of the most influential research firms in AI infrastructure. Stacey Raskin is managing director and senior analyst at Bernstein, ran tomorrow Wall Street's top semiconductor analyst for more than a decade. Our moderator is Austin Lyons, chief analyst at ChiefStrat and former Intel chief engineer. Please welcome Dylan, Stacey and Austin. [00:01:27] Speaker 2: All right. Hello, everyone. Y'all ready for this? Yeah. All right. This should be good. This should be fun. Okay. So we are going to talk about the industry. We are going to look ahead to 2027, 2028 and hear these two gentlemen's thoughts about where the industry is headed. We will try to cover it all from financing to GPU, CPU, down to Foundry. So let's get into it. I actually want to start with the finance side. So my question for you guys, lately a lot of the news has been about the cost of capital. So we know, you know, hyperscalers are investing a ton and free cash flow is going toward negative. And now we've got a 10-year treasury over 5%. Even if you zoom out like CoreWeave, there's talk about them, you know, bonds at 15% yields. And so my question is, should we be concerned, is there going to be an impact as companies are investing more and more that it's becoming more and more expensive to finance this? So I'll start with Stacey. [00:02:42] Speaker 3: Yeah. I mean, look, there are a lot of constraints in general that are sort of moderating the pace at which this build-out can happen. And these are power and supply and semiconductor tools and now maybe financing. And if, again, I'm a Wall Street analyst, I mean, if you look at the stocks, there's still up a bunch on this, but I mean, they're clearly way down from the peaks and the market is telling you something. They're worried that we are, you know, that there's a lot of uncertainty in terms of like the pace and sustainability of the build-outs and maybe the peak is approaching. And clearly financing is one of those concerns. It's probably up near the top. And you mentioned a number of, you know, different reasons like the hyperscalers. These were the most cash-generated businesses in the history of humankind. And like they're burning cash now and probably going to be raising equity. You know, Meta just had a new share class structure that, among other things, that will enable them to do that. And you know, we saw this Google raised and others. You mentioned Corrie's bonds. They're, that's their unsecured debt. It's not like what they're paying right now. That's where they're trading in the market. But that would suggest if they go out to the market today to get new unsecured debt, that's in the ballpark of where they'll be spending. And it's a lot of money. So I think these are all broad concerns. Right now, you know, though, it's, I think at least for the next year or two, which is as far out, frankly, even than I can look, we're probably okay. But I do think it's pretty clear that we need to have revenues grow to the point where this becomes less of a concern. And maybe that happens. You know, I don't know. SpaceX reported earnings. It was this, this week, last week. I can't remember. And I mean, he's running out, clearly he's running out in the spot market. And so things are very, very constrained. The only thing we're hearing is nobody has enough compute, but he's running capacity out short-term contracts, but, you know, $50 billion a gigawatt, something like that. At those kinds of rates, I don't think it's actually that hard to imagine a scenario where this stuff like gets covered and then some. But again, it gets to the question, like how sustainable are those? And that has broader questions. It really depends on like what does demand actually look like relative to supply and how long does the supply demand mismatch last? And these are questions that actually maybe Dylan would probably have a better idea. I do, but absolutely it is a concern. It is one of many constraints that are out there that I think are moderating. I'll say one more thing though. The idea that these constraints are moderating growth and, you know, nobody likes it, but at the same time, it may be a good thing in some sense too, because it stops things from going nuts. Right? And, you know, we saw this during, you know, even a little before my time, but during, you know, during the tech bubble and like there's dark fiber in the ground today that's there that was over built that was never used. And some of these things actually may give us a stronger for longer kind of behavior because it almost forces, you know, the demand is there and it forces the supply to push out even more and allow things to be better matched rather than, you know, completely going nuts today. And then risking, having further risk of a crash who's going to the line. So I don't actually think it's necessarily just constraints in general necessarily a bad thing, but I do know how they're perceived and clearly they are impacting like the trajectory of the stocks like in the near term. [00:05:58] Speaker 4: I would say, you know, echoing Stacy's point that in any build out, everyone's going to want to build, build, build, and then it's sort of you get closer to the road. There we go. It's kind of like crack weed. [00:06:14] Speaker 2: Do you guys have a spare microphone for him? Yeah. We'll just stop off for a second. Thank you, Stacy. [00:06:20] Speaker 5: I'll never give it back. [00:06:21] Speaker 2: We still need one. Yeah. [00:06:24] Speaker 5: But anyways, I think when we look at any build out over the last few decades, we've always had the industry double order, triple order, and too much capacity get built out. And supply chains are really bad at forecasting demand. This is, you know, as Stacy mentioned, there is a bit of self moderation, especially on the finance side. But if we go back a year or two ago, any cloud that wanted capital could, they would pay a pretty high interest rate, but they could get it. Now, if you don't have really high grade offtake, you're not able to build it if you're some random cloud. And so ultimately, the hyperscalers, sure, they can take their free cash flow down to zero, they can raise some debt, and their earnings are obviously growing. So their capacity that they can build out grows, but they can't build, you know, they can't do north of two trillion capex next year, they can't do two and a half trillion capex next year. So that sort of moderates how much they can spend. And then when you look at the new cloud industry, which accounts for, you know, non hyperscaler demand accounts for nearly half of the stuff that's being built out, if they don't have offtake from a hyperscaler, how are they going to get it financed? And then, you know, the market's already saying, hey, if you've got offtake from Anthropic or OpenAI, we're going to price it at, you know, 13, 14, 15% debt. And so we saw that recently with Volta, they're making a cluster for Anthropic and their debt is basically a 12% coupon, but they sold the bonds for 95 cents. So it ends up being like $15 or 15% interest rate. And if we look out at everyone who's not getting offtake from these companies, hyperscalers, the major AI labs, how do you build that capacity? How do you build capacity to service, you know, smaller enterprises? How do you build capacity to service, you know, new startups? Well, you just literally cannot unless it's coming from a hyperscaler. And so that's Elon's whole gambit is he's making all this capacity that he can sell at very high pricing on short term rate rental contracts to startups to the AI labs. But ultimately, it is quite a dampener. And so we've seen, you know, NVIDIA recognize this is a very big problem for them because, you know, hey, they have five customers who can buy as much as they want or as much as they have capital for. But what about the other part of the market? And how do they enable that? And so they've gone and done really creative structures where with sharing AI and firmness and a few others, they've taken revenue shares on the cluster in order to guarantee that it's financed, right? So they'll say, hey, we'll rent the cluster from you at a guaranteed price of $3.50, which is the cloud making zero money. And then if you rented it to anyone else above that price, we get a revenue share. And we're starting to see more and more of these structures. In fact, even to build a cluster for Anthropic, Lambda had to do a revenue share with NVIDIA. In this case, it wasn't a 50% like Sharon and firmness had, but it was still a 10% revenue share. And so, you know, how these clouds can get the capital to over, you know, sort of build and get to that point where actually they build everything they want to build and get to the oversupply scenario is really tough just because of the financing of this all. And so we're seeing more and more turn from, you know, these long term loans and long term debt to something that's more shorter term. So Broadcom's done the same thing with Anthropic where they're basically leasing them the chips in a sense, right? It's a rent to own. They're stretching out the payments over a few years. And the whole name of the game here is if anyone wants to get capacity, if you're not the big companies, you've got to come, you know, with a deal, a structure that has, you know, NVIDIA or Broadcom or AMD backstopping these clusters. And so ultimately the financing of this all is a big dampener. You know, the hyperscalers are also, you know, very, very large off balance sheet and on balance sheet commitments. Like Meta has $1.4 trillion of commitments for compute and data centers and so on and so forth. And that's then able to get financed by the whole market where they go buy a data center or rent a data center from someone or rent gyps and then those guys can go turn around and get capital for it. But ultimately there is a big struggle for anyone to do more than that and also a lot of people worried about even Meta's financial structure. I mean Meta's completely fine but, you know, even the capital markets are getting spooked by that. So this is the, you know, financing is part of the holy sort of trinity. You know, it's Land PowerShell, it's chips and it's financing. [00:11:11] Speaker 2: That's good. That's interesting. So, okay, it's getting more difficult to finance, although the hyperscalers are fine, maybe a little bit harder for NeoClouds. We're seeing these new creative structures come around to help that. If we set finance aside and go back to the traditional maybe bottlenecks that are limiting ultimately the pace of the growth of the industry. Talk to me about where you see that 2027, 2028. You know, there's obviously, ASML says they're sold out in 2028. TSMC always has more demand than they have supply, even though they're doubling capacity for two nanometers. And then, you know, people talk about power as a bottleneck. Dylan, I know that your team recently this week maybe put out a post about up to 50 gigawatts coming online next year. So maybe I'll start with you. How about you stack rank some of these different parts of the semiconductor supply chain and where should we be looking at the constraints? [00:12:11] Speaker 5: Yeah, so I think on the semiconductor side, the biggest constraint, of course, besides memory, is substrates. Substrates, PCB materials, substrate materials, all of these components were severely under-invested in. Much of the supply chains that create the materials in Japan, those firms are sort of a lot more conservative and unwilling to invest heavily in capacity. And so we've seen, you know, tremendous bottlenecks there in terms of what can be made over the next two years, you know, both for accelerators as well as CPUs and every other component in the supply chain. So substrates is, I think, the biggest bottleneck of all the bottlenecks, including, you know, land power shell, semiconductors, tools. I'd say the next bottleneck would be, of course, memory. And then following that, there's a lot of bottlenecks around the desire to ramp optics to a really, really large degree, right, with co-package optics or near-package optics. There's humongous constraints in ramping that, you know, we just simply cannot ramp. You know, even if everyone wanted to do CPO in 2028, or even one major company wanted to do CPO in 2028, they could not satisfy their volumes in terms of lasers, in terms of fiber attach, in terms of all these different components that go into it. And then, you know, as far as land power shell, power is definitely a big constraint. But one thing that our research sort of shows that I think is counterintuitive to much of the market is that next year there is, there is, you know, obviously there are data center constraints, but it's a lot less than people realize because there are nearly 50 gigawatts of data centers being built globally. You know, seven or eight gigawatts in China, and then most of that, you know, about 35 in the US. And then a bunch of rest of everywhere else, you know, Australia, India, Malaysia being some of the big places, but really all over the world data centers are being built at a tremendous pace. And while power is a huge bottleneck, people can find creative ways around that, whether it be, you know, turbines on site, or reciprocating engines on site, or fuel cells on site, all of these things are helping alleviate the power bottlenecks. So I think actually power, you know, as far as like a consent, you know, as far as what most people think, oh, data center and power are the biggest constraint, I think they're constraints, but I think they're actually a lot less than substrates, than memory, than lasers, and other things in the optic supply chain. Because there are things that people can do to ramp power capacity on a shorter time span, right? They are just significantly less complex than the semiconductor supply chain. [00:14:56] Speaker 3: Yeah, you bet. I think it's really interesting on the power side, you know, like, so Jensen and Hawk, like Envy and Broadcom, have given some pretty sizable guidances. And Jensen said he'll grow his revenues to at least 70%. Hawk said 100. And those are actually constraint numbers. And they're looking primarily, I think, at land power and Shell as their constraints. And maybe that's because they're semiconductor companies. They know what they can make, I guess, in terms of the semis. But on the land power and Shell side, I think I agree, there are ways to get around this. Although, even for some of the things that are going on set, I mean, I don't know what the lead time is on a turbine these days. I mean, Musk says he's going to go start building his own turbine blades to get around that. I think on the memory side, it's really interesting. You know, the memory stocks are all trading at two, three, four times price-to-forward earnings. And pricing has gone through the roof. I mean, I look at, like, I don't cover the memory stocks, by the way. It's a colleague of mine. But you take a look at, like, a SanDisk that, you know, they guided to a quarterly EPS that's higher than the stock price was, you know, 24 months ago. It's, like, really amazing because pricing has gone through the roof. And they're all trading at, you know, single-digit P's because that's how you trade things at the end of a, at the peak of a cycle. And everybody's worried that, especially as you get into 28, supply does start to come online and they're worried that the pricing collapses. And I sit here and I always wonder, like, what happens if it doesn't? You can look inside these racks, look inside an AI rack, 70%, 80% or so of the silicon area inside that rack is memory. It's mostly DRAM. You know, there are other issues when you're making high bandwidth memory, the memory that goes into the AI chips. There's something called a trade ratio, but, you know, it takes three or four times as many wafers to make a gigabyte of high bandwidth memories. It doesn't make a gigabyte of standard DDR5 DRAM that would go into a PC or go into a smartphone. So you're sucking up wafers from that. I mean, we need a ton of memory. And this feeds to one of the other bottlenecks, which, which is semi-cap, the actual tools to build this stuff. And, you know, we, we had a company report this week. They actually gave a preliminary outlook for WV for wafer fabrication equipment for next year. And he said more than 200 billion. I think a lot of investor view was, it was going to be a lot higher than that even, which I found really surprising because I just don't think it's possible. We don't have the fabs like there's your other constraint is we don't have to, it takes, there's no clean rooms. Like they're all full. They've actually been squeezing as much as they can out of the existing footprints. That's why, um, the WV estimates beginning of this year were, you know, 120, maybe we'll probably end the year 155. That's all finding other better ways to use the footprints. It's there, but now they have to build and it takes years to, to put this stuff into the ground. The clean rooms do start to come online a little in 27, more into 28 and 29. And at that point, I guess, we'll know if there's a supply demand mismatch or not. But if the demand profiles continue at the rate that I think people seem to think they are, memory may still continue to be, to be a constraint. Um, you talked about substrates and PCBs and other things. Again, to go back to the tool vendors, they're, they're all complaining about short term supply constraints that are actually impacting their own ability to upside even in the near term. And it's exactly that because it's PCBs and it's actually semiconductor chips. And it's even other components, you know, valves and mass flow controls, everything. But it's simple stuff that's actually affecting other parts of the, of the space as well. We saw a little bit of this sort of thing during COVID. I do think that the, at least the semi cap players and most of the folks in the supply chain did learn some lessons from COVID. And then I do think they're managing that sort of thing better now than they used to. But at the same time, we've, we've got all kinds of constraints that are in the spaces that need to get fixed before the other constraints can actually get, get fixed. So you have these sort of layers, I guess, to things. Um, I actually don't really see, I think things are going to be tight for a while, frankly. Like, I think this year is fine. I think next year is going to be the same thing. 2028 will probably still be tight. 2029, I don't know. I mean, that, that turns into a demand question. Um, but I suspect we're going to be tight for a while. [00:19:00] Speaker 2: Okay. So if we're tight for the next couple of years, um, and we talked about prices just skyrocketing, you know, memory industries making margins in the 80%, 70%. Higher, higher. Yeah. Yeah. 88, 90. Right. Uh, so, um, do you just see, uh, you know, all those players, everyone's constrained up and down the whole stack. Do you see every industry in there just continue to kind of charge for their value? [00:19:28] Speaker 3: Well, you know, it's, it's funny. So this is an industry where, where pricing coming down was just sort of taken as a given. Right. That was just, and in fact, the semiconductor industry as a whole historically was a massive value destroyer. It never covered its cost of capital. Like all of the value flowed to the consumer. Um, I did a piece, I used to work at McKinsey many, many years ago. And, and, and I did a piece of work there. It was called value creation of the semiconductor industry. And, you know, we defined value very classically. It was economic profit. Roeck minus whack times your average invested capital. And we aggregated it up. It's like 150 different companies over like 12 different subsectors over like a full cycle. I think the initial work was like 96 to 04. I think over that period collectively the industry destroyed $5 billion in aggregate economic profit. And if you took Intel out of the mix, it was 50 billion destroyed. It was, which was a big number back then. Right. Um, so the industry didn't actually really start covering its, its cost of capital. Meaning that, that all of the, the economic profit you created was, was before that was flowing directly to the consumer. It was a wonderful time to be a customer for semiconductors. But post 08, there was a lot of rationalization and consolidation. At least we started to kind of like come in and, and, and kind of meet the cost of capital. Um, and then COVID happened. And what the industry discovered in COVID was that they have pricing power, which I think was something that they did not feel in their bones before that. And, you know, we had input costs going up and everything. And, and I didn't, I cover a bunch of companies and I didn't have a single one experienced margin compression because of input costs going up. They all raised their prices to offset it. And they raised their prices at their margin structure. So like if their costs went up by a hundred bucks, you know, they raised it by 160 if they had a 60% gross margin and they had no problem passing it along. And right now we're seeing, and by the way, I've said this right, inflation saved the industry during COVID. If you look at industry like unit shipments, like in 2023 or 2024, I mean, they were flat to 2019. And ASP is blended for the industry, we're 50% higher, 5-0, even if you take out memory, right? Um, and we're seeing that on steroids right now. So clearly memory pricing has gone through the roof. The industry, by the way, we're going to do a trillion and a half dollars this year, maybe more, we're going to be up probably a hundred percent. Seventy or 80% of the growth we're seeing this year is memory ASPs. Um, a big chunk of the rest of it is also like non-memory ASPs. And we got, I don't know, units are up 10% or something like that, whatever, whatever it is. But, um, the industry is now flexing that pricing power a little more. And by and large, I think customers are accepting it. They don't really have any choice. So, um, yeah, it may, it may, it may stay tight. I mean, so if it stays tight, we will probably continue to see this, I think. Yeah. [00:22:09] Speaker 5: I would say that, um, you know, the semiconductor industry is very interesting in that some people, uh, flex their pricing power, uh, in, in the good times. And they, they suffer really bad at the bad times. But then others, they just sort of eat margin the entire way through. So if you look at TSMC over the last decade, they've risen from 45% gross margin to now 60. Um, and it's been a very steady climb up. Yes, they've taken more when times were good, but they haven't given back much when times were bad. Um, you know, and you look at memory, right? They've gone from, um, basically negative margins just a few years ago to, to now, you know, 90%. [00:22:42] Speaker 3: People forget we're coming off the worst memory cycle since the tech bubble, like just like 24 months ago. Right? Negative gross margins. [00:22:49] Speaker 5: And, and, and so like when you, when you look across the supply chain, some of these folks, um, are taking tons of pricing because, you know, it, it is a structure of also like, what is the market like? Right? And so TSMC where they're a near monopoly and they're very long part-term partner has not really jacked up pricing even though they could. Whereas memory companies, you know, if, if Apple's going to screw them in the downturn, they're going to screw Apple over in the, in the upside of the cycle. Um, which, which is pretty clear, uh, clearly what's happening. Um, so, so when you look at this like sort of market, right? To build a gigawatt, you know, in, uh, in 2024, it was roughly $40 billion, right? Across chips, land power shell, everything. Um, now it's north of 60. So the price to build a gigawatt of compute has skyrocketed. Now, of course, not many people were building a gigawatt or really no one was building a gigawatt in 2024. And, and a lot of people were building gigawatts in 2026, um, in 2027. [00:23:43] Speaker 3: You're getting more out of the gigawatt now too, by the way. [00:23:45] Speaker 5: Yeah. And then that gigawatts profitability is huge, right? And so, um, I think everyone is recognizing profitability extends, um, you know, a lot of the mechanics I mentioned earlier, which is you can't build a cluster as a cloud unless you have offtake from a hyperscaler, which means you've locked in pricing ahead of time. Um, that, that is sort of being exploited by Elon in the sense that, okay, well, if chips and land power shell are a constraint, if I just build these and then I don't sign a contract until it's actually up, I can, instead of making 15, uh, you know, billion dollars per gigawatt, can make 50. Um, and, and then the AI labs who are making, you know, anthropic in their recent sort of disclosures is making on the order of 70 to 80 billion dollars a gigawatt off of their inference capacity. Um, you know, they're kind of forced to eat it, right? You know, yes, their margins are, you know, 75, 80% when they're spending 15, um, you know, billion dollars a gigawatt for a compute. But if they're spending 50, their margins are much smaller. And so it's, you know, not clear who's going to eat all the cost. Um, you know, Nvidia's had to keep their margin roughly the same, slightly decreased, but they've had to eat the memory pricing increase and, and other parts of the supply chain like substrates and optics and all these other things. Um, and, and so the prices that they charge their customers have soared despite the fact that their margins are flat. Um, and so this is, this is sort of, we're seeing really interesting, you know, who gets to keep their margin, who gets to expand it, um, and who's going to eat margin. And I think it's a, it's a very, uh, fun time to be on the pricing side, um, because you're real time discovering and yes, you can sign a deal at many different prices. Uh, but the, the best sales guys are going to get, uh, really insane pricing and, and hopefully they're rewarded for it. [00:25:28] Speaker 2: Hilarious. Um, so, okay, uh, Stacy, I kind of like what you said about COVID. It was almost like a dress rehearsal for AI for the industry to start to realize like, hey, we can charge for our value here. Um, and, and Dylan, you know, you make an interesting point about like a system builder like Nvidia having to decide of all the components as their prices are going up. Do we pass it on or do we keep our margins the same? Now I'm feeling a little tension of, okay, they're still going to be billed out all the way through, you know, we're not sure what's going to happen in 2029, but clearly, uh, supply is limited and demand is high for the next couple of years. And yet we started by saying that it's getting harder and harder to finance. And so I'm wondering if there becomes a point where those people at the top of the stack, the metas, um, in the model lab companies, if they're ever going to push back and say, this is just getting too expensive or beggars can't be choosers. [00:26:20] Speaker 3: Remember, you have to look at more than like the margins and the costs. You also have to look at what they're getting out of it. And so, I mean, I'll, I'll take Nvidia as an example. Um, you know, again, their, their pricing is going with every generation. So you, you, but the, but the, what you're getting out of it is going up more than the price, which I think is important. So you went from like an ampere to hopper. Yeah. It's roughly doubled when the performance went up 10 X or whatever. And again, Dylan would have better numbers for exactly what the performance are good, but it was a lot, right? Very NPV positive sort of thing. No, no problem. You can pass that along. You go hopper to Blackwell. It's a little harder to tell, like the Blackwell architecture is much different, but I mean, call it a 50% roughly ASP increase per GPU. Performance went up any, anywhere from three to 30 X. And again, maybe, maybe more given some of the, the later optimizations and software and everything. Again, no problem. You go black or Blackwell Rubin. Same thing that the cost of the racks probably doubling, right? And cost per, per gigawatt. I think Nvidia said for their own reference, I was like 25 billion per, for Blackwell and 40 for, for Rubin. So, but again, performance up 10 X. So that's what needs to happen. You need to keep giving your customer value because ultimately that's what's important is how much performance you're getting out of that gigawatt of compute relative to how much it costs. And as long as that gap is getting bigger, I think it's still very easy, but I mean, it's palatable to do this. I mean, hypothetically, just if, if, if they were to roll out Feynman and it's only 20% better than, yeah, then I would start to worry a lot about pricing and margins and everything. But this is always a huge controversy, by the way, for a company like Nvidia, because they do have margins up in the, in the seventies on a massive revenue base. And so people just wonder, you've got competition, you've got ASICs and you've got other GPU vendors and other things. And how can they hold the margins in the, in the wake of all the, and this is always in my view, as long as they can keep improving performance at a significantly higher rate than their ASPs are going, because they have to raise ASPs because their own costs are going up, right? And the things are getting more complicated. So you have to keep that performance to cost equation working for the customer. And as long as they can do that, I don't really worry about the margins, but if the pace of that, of that innovation slows, then I would worry an awful lot. And they, they're aware of this too. I went to the hot chips conference. This would have been in like 2023. And I didn't understand anything there. Like I'm not a computer, I'm a, I'm a, I'm a process engineer by background. So all the computer science and chip stuff design stuff goes over my head, but they had Bill Daly there from Nvidia as their chief scientist. He gave the keynote as the one thing that I understood. And what he was talking was exactly this GPU performance improvements. And he kind of made the point over the last 10 years, we've improved GPU performance by a, by a thousand X. It was a whole bunch of different levers that they pulled, but they got there. And what he said was over the next 10 years, we want to improve performance by another million X. And here's where we're good. And so further on the trajectory, and if they can do that, then I don't worry as much. If they fall off or if anybody else falls off, then, then I worry a lot because ultimately somebody has to pay for it, right? In a perfect world that somebody is the customer and there's a lot of value from it. But, you know, if the costs go up, you know, at a rate that's unsustainable, like the suppliers have to start absorbing some of it. And then it, then it's a dog fight to see who, who does and who doesn't. Yep. Gotcha. Dylan? [00:29:26] Speaker 5: I mean, I mean, that's certainly the trend is all of these companies are trying to make chips, right? OpenAI showed, you know, I think for the first time ever in the industry that you can make a first generation chip that's really good. It looked good. They let, they let us into the lab. We tested on our open source inference X benchmark running all these open source Chinese models. And the performance was better than Blackwell. And that's a first generation chip and they utilized AI heavily in design and, and programming process. And, and Anthropic is starting their chip program and they want to do it in half the time that OpenAI did. So they want to, they want to get a chip. I mean, these are aggressive goals. But if, if, you know, models are able to help them, you know, really accelerate this progress process. You've got Google selling TPUs externally. You've got Amazon's training them really quite good now and, and fully usable in PyTorch and such. Meta is very aggressive roadmap, Microsoft, et cetera. If you look at the, you know, I said, I said it, it went from $40 billion to build a gigawatt to 60. If you remove all the margin out of the chip, you know, design houses, right. The Nvidia and Broadcom Marvell, et cetera. You're actually at like $20 billion a gigawatt. And so, you know, if you were to able to do that, then you could, you could pay for five gigawatts with one gigawatt worth of compute running inference. And so obviously the economics here are very aligned. And as Stacey said, these companies just have to run like crazy, right. You can't, you can't afford to be 2X performance per dollar per year because you're going to get completely smashed by these, these other companies. Right. So, so like, like, like in, um, and so if, if, you know, Nvidia can keep up this incredible pace. Sure. They, they deserve to continue to earn margin. But if the other competitors, you know, and their, their main major customers are able to, um, design chips, deploy them, scale their manufacturing, you know, unmargined memory, TSMC wafer, substrate, optics, et cetera, is, is really, really cheap. So those guys, I think those guys have a lot of room to run on pricing, especially if things get disaggregated more. And Nvidia's sort of loses their position or Broadcom loses their position. And, and that's why these two stocks trade at their lowest multiples basically ever. Ever. Um, you know, is, is a lot of people are worried about that, right? Is Meta going to do, is Meta going to, uh, disentangle Broadcom? Is Google going to disentangle Broadcom? Are both of these guys going to stop buying Nvidia and just in-source completely? Um, and so, you know, I think the component vendors are going to be able to jack price. Up like crazy. Uh, the question is if, um, you know, Broadcom Nvidia can continue to earn these incredible margins. And at least for the next two years, it, it looks like, yes. Um, but beyond then, like, is, is, is, is that off the table now? Because at the end of the day, while chip design is, is very, very complex, um, models have solved software, um, and programming accelerators is actually really easy now. Um, it took us six months to get Nvidia inference working for inference X, our open source, the largest open source inference benchmarking. Um, it took us like six months or nine months to get AMD working. Uh, we got Google working in a few months and we've got Amazon working in a few months, right? Um, we got open AI working in, in like a month and a half. Um, so it's, it's, it's sort of the pace of this from, uh, you know, chip works, you know, to, to, you're able to run software on it is, is rapidly compressing. And now these companies are all turning their eyes to chip design. And, and can that actually work? It's, it's similar in the sense that it's very verifiable like software. Um, obviously there's a lot of challenges on data and scaling that, but this is the big question for the semiconductor industry is, uh, does design become more commodity like software? In which case everyone can design chips. Obviously there's the pesky fact that you have to pay $50 million for all the fixed costs of masks and EDA tools and all this other stuff. But, um, you know, the number of people designing chips is going to explode. Um, just because programming them is much easier. And then if designing them becomes much easier than margins have to fall, right? We've seen SAS companies get obliterated in the market and we've seen, you know, Nvidia and Broadcom really trade sideways despite that their earnings are going at 2X plus year on year. Um, you know, we'll, we'll see what happens, but that's, that's the big, uh, question. [00:33:36] Speaker 2: Yeah. So, uh, Stacy, I liked your point that it's not about the fact that it's going from 25 billion to 40 billion to 60 billion. It's just, are you squeezing out more tokens per dollar? And that's all that really matters. Um, and then Dylan, you brought up XPUs and, you know, OpenAI, very interesting example. So, how should we think about over the next two years, um, the growth of GPUs, the explosion of XPUs, there's all sorts of AI accelerator startups that are taking different architectural bets. Um, is it going to look any different in a couple of years? Yeah. [00:34:09] Speaker 5: So, so I think, um, for XPU companies, there's sort of two paths. If you're a hyperscaler, you can take a more conservative design approach and do similar things to Google or Nvidia and, and, and it works out. Um, but if you're a startup, you have to do just really crazy things. And that's why in the first gen of AI hardware startups, the ones that did the craziest things, um, Cerebrus and Grok were the only ones that were like the most successful. Um, as we look at the current crop of all these startups that are popping up, only the ones that do really, really crazy things that are very different and can't immediately be copied by Nvidia or Google or whoever, or AMD. Um, those are the ones that are gonna, um, those are the only ones who have a shot of succeeding. The nice thing is because there's so much capital flowing and there's a lot of really crazy ideas being tested, um, and, and companies that are trying to build some really insane formats for, uh, building XPUs. Um, you know, I think, I think 23, 24 Nvidia swallowed up as a percentage of, you know, total HBM, total COOS, et cetera. Um, in the industry, as we looked at 25, it started to moderate, but then in 26 Nvidia's lost share, right? They have, they're using a lower percentage of the total, um, COOS, uh, wafers, lower percentage of total HBM. And again, next year they lose share. Now, ultimately the number is growing so fast. Their earnings is still, you know, doubling year on year, uh, next year, um, or more. But ultimately, um, this is, this is a big sort of, uh, you know, it's a thing that's already happening is Google's taking share. Amazon's taking share. AMD's taking share. Um, unit basis. [00:35:43] Speaker 3: That's what you're sorry. On a unit basis. Yeah. [00:35:46] Speaker 5: On a, on a, on a, yeah. [00:35:47] Speaker ?: Yeah. [00:35:47] Speaker 5: I mean, units are always difficult. Who makes a big chip, small chip. But COOS wafers and HBM are kind of hard to, and, and revenue, right? Dollars, dollars spent on froundry wafers, on HBM, on these sorts of things. Um, these guys are taking share. Ultimately Nvidia, again, will the, the base is growing so fast that they're still going to double earnings. Uh, and so, you know, as we go up to 28, 29, um, you know, maybe growth rates aren't as high, you know, that maybe they're not, uh, you know, a hundred percent increase year on year of revenue. Um, and then, and then what happens, um, you know, does, does the growth moderate to the point where Nvidia actually goes? Or are they, you know, they're making some crazy bets for their next generation, you know, architectures past Rubin. Um, so, you know, maybe that, that's to, to Stacy's point, right? They just have to run really fast. And, and that's how they defend themselves. [00:36:32] Speaker 3: I mean, this is one of the reasons you Nvidia has moved to an annual cadence for, for, for chips. It wasn't just for them. It forces everybody else to try to run, to keep up. And, and, you know, they've, they've got a scale that makes them hard to, to catch. And that's one, I mean, Nvidia's got a lot of, a lot of people always talking about moats. They've got a lot of moats. There's the CUDA software that we can, we can have a debate on whether or not CUDA is actually a moat or not anymore, but, but that's there. And, you know, the, the, the ecosystem and, and, and install based infrastructure. And then even their, their balance sheet and the financing is, is, is a moat, right? Um, the pace of their design, they're, they're leveraging AI like to, to speed up their own design. They're not, the industry is not doing stuff there that like they're not aware of and they're not using. Um, but it's a race, right? And like Jensen, at least is, is paranoid, right? He took the old Intel lessons to heart, right? He's looking over his shoulder every day. Um, yeah, this is absolutely why. These stocks are trading where then you're, you're absolutely right. They're, they're trading cheaper than they ever have. I wrote a note. Oh God, this was probably nine or 10 months ago, but we were looking at like Nvidia relative to the socks index, which is a, a broad index of semiconductor socks. And as of, when I wrote the note, I think there were only 13 days in the last 10 years that Nvidia had traded cheaper relative to the broader semi industry. And that was like nine months ago as cheaper today, right? So it's, it's, these are, these are the things that investors, that they, that investors worry about. It's mostly, it's beyond just the, you're worried about like overall demand sustainable. Like if the demand's not there, we're all screwed. Like regardless of whether it's Nvidia or anybody else, but within the group itself, it's clearly competition is their biggest threat because they're, they're the, they're the forerunner by a mile and everybody's trying to eat their lunch. But I mean, at the end of the day also that's semiconductors. There's always trying to somebody trying to eat your lunch. Like it's, it's up to you to make sure that you can stay ahead. And that's something that they will have to do that will probably get harder and harder as you go forward. I mean, just, just naturally. Sure. [00:38:24] Speaker 2: Sure. I mean, we're, we're talking about GPUs and XPUs and unit share and all these competitors gaining share. We're really focused on the data center and hyperscalers pretty much. What do you think about enterprise on-prem as an opportunity for someone like Nvidia? [00:38:40] Speaker 3: Well, actually, even today Nvidia splits their revenue out and they split by hyperscale and non hyperscale, although it's not exactly clean. Right. Because, you know, they got the, the labs are in the non hyperscale, but I mean, some of the hyperscale spending is going to support or vice versa, you know, so it's not clean, but they're about 50, 50 right now. I think the labs directly and indirectly for Nvidia are roughly 20% of the revenue right now. So it's, it's, it's sizable, but I would say with Nvidia, their revenue base is maybe even a little more diverse than, than maybe we had thought, which is why they're giving that presentation now. Um, at the end of the day though, I mean, you clearly need enterprise adoption. I mean, that's, that's where, that's where this is. If we don't have enterprise adoption, I mean, there's, there's, there's no industry. It's, it's still early days, but I think certainly we've got at least one killer application of enterprise, which, which is coding, right? And, you know, those revenues, agentic coding revenues have gone vertical. And, um, I don't, does anybody write software, but by hand, like artisanal code anymore? I don't, I don't know that they do. I think everybody uses it now. What's coding? I mean, I, I, I don't know. Um, I also think on the enterprise side, you know, people are looking at revenues, but it's, it's, some of the, the, the, the actual saving may not be revenues. It actually may be productivity, right? And, and, and, and it's harder to measure, but I'm, I'll give you just a dumb personal anecdote. You know, I, I, I'm a financial analyst. We build models and, you know, one of them, one of the guys who works for me for his own personal account, he's been using Claude to build models, like for himself, right? For, and he, we, we tried this among the team, like a few months ago, it was, gave us garbage, but he said, I think it's working. And I said, okay, like, I want you to, we're going to get on a team's call, you know, I want you to throw it up and say, we're going to read, have it, have it build a, for a company we already cover, have it build a model for us from scratch. It one shot at a model in an hour and 20 minutes, 10 years of history, went through all the different resegmentations. It captured it all. Um, it even forecasted, it, it listened to what, like the CEO said on earnings calls and forecasted trying to use logic rather than random. It was, so, I mean, it was something that would have taken one of my guys a week, a week to build and would have taken me, you know, three months to build, right? Um, in an hour and 20 minutes. And so, like, I could, I could, I could, we can do more, I can do more with less. Now, there are other implications of that. Like, are we going to be in a world where AI, you know, adoption of the enterprise is rampant, we have 20% unemployment? Like, I, I don't know. Like, that's, there, there are other, you know, factors to think about here. But, um, I'm, I think we're still early. And I think most of the anecdotal evidence that we're hearing right now, because this is the, this over, is the overarching debate. Is there ultimately a return to use this stuff or is there not? I think anecdotally we're coming down pretty solidly on the sides of yes, there is a return. But, but it is absolutely still early. A lot of companies are still in the experimentation phase. They don't have the expertise to know what to do or what they need to do. And so it takes time. Yeah. But I, I've been encouraged by what I've seen. Totally. [00:41:29] Speaker 5: The only thing I would say is, um, for enterprise adoption is, is I don't think that enterprises are going to adopt that much on the hardware side, um, and run their own models and such. I think, I think the, the efficiencies of running batch, the, you know, the scale economics of models mean that most people are just going to be buying tokens, um, and implementing them, you know, combining them with their own data. And ultimately that's, that's the real way that most people are going to consume AI. Um, and so it still leaves this sort of stratification where, you know, I think, and I think the industry has gone towards that more and more, right? If you go back two decades ago, everyone bought their own infra, infra and deployed it themselves. And now most of the spend is in the cloud and most of the spend is going to be buying tokens. Um, even pharma companies, even hedge funds, like all these guys, most of their compute and spend on AI is going to directly for models, not buying compute and running open source models. There'll be providers who take that and do that. But I think, um, yeah, that I, I would love to talk more about this, although we're, uh, we're a bit over time. So I think, uh, it's very fun. [00:42:26] Speaker 2: Yeah, I see a big zero minutes left. Okay. I guess that means we're done. Um, okay guys. Well, this was awesome. Uh, I learned a lot. This was great. Thank you so much. [00:42:35] Speaker 3: You bet.