The Next AI Trade | Liam Dalton | Ep. 4

Liam welcome on the Miss Price assets podcast. A good friend of mine. We argue every single day over the markets. Usually when we're on different sides of the trade in the short term, I'm on the wrong side. Liam's a macro first investor. He decides what the macro is and recently that's been AI. I would love you for you to introduce yourself and sort of give a description of how you for your entire fun's life obviously you've been trading for longer. But what is the market personality today? Thanks for having me on Nick. Brief background on me started up Will O'Lake a couple of years ago coming up on year two in November. Previously with Axiom Capital, my father's firm he founded in 1990 and I continue to be there as a registered rep so that'll inform some of what I'm able to talk about and not talk about in terms of things that might sound like recommendations or financial advice. But Nick and I have been working closely together on markets now for it's got to be a couple of years now and a little bit before that even and so it's great to be on the on the Miss Price assets platform and we've done some research and some work together with the memory war and so lots to cover. When we reconnected two years ago about two years ago we were just arguing over China and Taiwan and I was like I like this guy we should talk more but what's the past two years been like from your perspective? Gosh well you know I think that and I think I've told you the story before but recently Stan Druckenmiller you know the fame Ducane investor he was on a podcast I believe and he was talking about late 2022 where a lot of market veterans like him you know David Tepper on CNBC actually I think in in the fall of of 2022 he was leaning short saying that you know the macro was pretty clear the Fed was on a tightening path and you weren't supposed to fade that even though at the time the market was off you know 25 or 30 percent but then Stan said that he went to a conference and he was telling people his views and he came away from this conference and it was right kind of around the time that that chat GPT the that kind of groundbreaking you know kind of generation of GPT model came out and he came away from this conference just realizing that AI was the macro and just you know so long as nothing came along to really derail the you know the financial system writ large there was this transformational realization happening at some of the biggest the most well resource firms in the world and that they were just going to have this priority of spending to build infrastructure to experiment in this space and support inference eventually and so you know between the guilt crisis the UK guilt crisis in 2022 which kind of took the Fed initially off of that tightening path at least behind closed doors and then Silicon Valley Bank Corp once that kind of didn't become become a full-blown regional banking contagion in crisis and the bank term funding program was put in place and financial stability kind of became more clear of a goal of the Fed and of the government in early 2024 that just in hindsight was the all clear that this was going to become the AI theme and it's been the most animating thing for the business cycle it's where the most dynamic things are happening in the economy you know I kind of said to people you know imagine being a long short equity manager at a consumer pod right now or in financials where you know you're trying to find okay what is the what is the edge and what is the alpha and you're basically sniping against other headstrong managers and other pods but really in the economy like the big dynamic thing happening right now and there's nothing really close to it has been this surge of spending and the priority that is placed on it and where that capital is going to break different bottle necks people trying to understand what the next architecture of models might unlock in terms of enterprise or consumer or even physical AI capabilities and so that's I mean listen that's set the that's set the pace of the market now with some interruptions you know some greater policy risk maybe with the Trump administration being quite a revisionist administration whether it's with the global trade order or with the the state of the Middle East you know they clearly want to rip down certain parts of the world and remake it in their own image and so that's created some risk events but really the main event has been this spending cycle the way that it's coursing through the biggest businesses in the world that's a good table setter I would say financials energy industrials it's all AI what it's driving the earnings of those AI right it would be so you know the vast majority of the private credit space where it's growing you know maybe it's stopping now but has been recently in the hyper-scaler spent but yeah you're absolutely right and I joke all the time we'd be in the Great Depression if it wasn't for Nvidia so thank you jensen for being longest Wang and you know so far we haven't had a recession we haven't had any problems but there's increasing risks on the periphery but if we look back you know talking about the appreciation of this massive massive colossal change industrial revolution times two probably right or at least condensed right in a much shorter time frame it's starting to be appreciated more right we've had and we came out with this memory piece on you know that outlined how it went from GPUs to you know ultimately you know there's kind of choices being made based on the constraints that there is and if you follow this intriguing index or semi-analysis or a rational analysis you see a tremendous amount of bottleneck finding right so what is the next one if the AI trade still goes you know potentially what if it's all over you know we've got increasing risks that are cooking interest rates have gone up you know continue to stay elevated that eats on a very delayed I think longer than people appreciate lag and we're looking forward to an election and potentially lame duck session so there's increasing risks and there's also at the on the other side a lot of bullish data coming out of these model makers now there's you know some things you can look at well is private capital driving up uh drying up for it you know every single time we have a deep seek or an open source scare it seems to be swatted away but before we get into all of that which will be the majority of the podcast I just want to talk about what is your strategy right you're coming into the markets I know extremely well read into a world dominated by pod shops even the single manager hedge funds are different than they used to be I mean Dan Loeb's a tech investor now right and you have the tiger cubs and the tiger cubs seem to be the same old lever beta except they don't make money on the upside the way that they used to and then the pod shops they're you know they're trading monkeys it's 30 day catalysts and you know that's sort of constraints and it's built around an intelligent model right and a lot of infrastructure which is how to optimize risk-adjusted returns but through all of that you found an opportunity different than my small cap lane but certainly an opportunity can you describe that yeah well you know obviously I think that there's never been more very intelligent guys on Wall Street trying to find trying to find where they can fit in and generate a big value ad for allocators and adding on top of that there's never been this much compute available I mean you have Jane Street with their own data centers a number of of market making hedge funds literally taking on compute responsibilities because they can just see that that's going to drive another exponential increase in speed and efficiency in the market you know I think that I kind of harken back to a little bit of an older model of you know if you're willing to take directional and some of a concentrated risk and basically I think that anybody who's running a strategy and I say this to investors any time you know they're going to come through the door and talk to me I say you know the really you want to make sure that your your manager understands what risks they're solving for because that actually exposes more truth about what their strategy is then them claiming you know this is the type of trade we're looking for this is how we make money so in our case we don't like a liquidity risk we don't like the risks that come with leverage and we don't like the risks that come with derivatives We also don't like single business-concentrated risk, but we're comfortable with the amatic concentration, and we're comfortable being directional. And really what that comes down to is that we think that it's certainly not easy, but relatively direct to find the most dynamic things that are happening in the economy. And when you find those things, and it kind of draws you towards a sector or a subsector, then you get to this level of predictability of earnings results that can create really kind of long-lasting and dependable trends in stock prices. And so when you marry up that type of fundamental work with a macro framework that helps you avoid hopefully, or even profit from some of the kind of major risk events that can come along, we do have a very highly levered financial system, there's a lot of leverage in assets. We have in many ways sort of a more brittle system, higher average valuations, less merchandise with more capital chasing that merchandise in public markets. And so when there is a major risk event, there's very sharp delivered jings. Just most recently you can remember July of this year in the AI trade, the tariff tantrum of March and April 2025, where the last, as we learned with the situational awareness episode, and as you could easily observe in the liberation day episode as well, the last 20 or 25% of that move to the downside was very, very clearly forced to action. And so if you're in a system that's more vulnerable to that, which I would argue the system is today, the gap that maybe people who are, you know, levered long only or who are very aggressive growth investors, I think the gap that they face is having some sort of macro and technical framework to help them avoid, you know, those episodes along the way to hopefully compounding growth at some of the greatest companies in the world. So we're really kind of coming at it from this position of why would I strictly be fundamental? Why would I strictly be technical or why would I strictly be macro? You know, macro doesn't always matter, but when it matters, it matters most. And then when macro is steady and there's not much macro volatility, that's really when you want to be diving in on the fundamental because that's when money is going to go to work, I think trying to find like the really interesting positive net revision and making sure that you're avoiding, you know, kind of being with management that's going to mess up a very clear opportunity. And so that's kind of the three, I'd say it's almost like a three-part vendiogram or I don't even know what I would call that. It's like that movie Captain Phillips where the Navy seals are sniping at the Somali pirates and they're like, ah, two green, one red, two red, one green and then when all three are green, execute. And so that's kind of how we try to do it. And so when we get macro fundamental and technical all lined up, that's when we really kind of feel very comfortable committing a bunch of capital. I wish the FINRA rules allowed me to just post scores on the doors, but they don't. I think that your strategy and the world of factor neutral, you know, underectional, hey, if we're going to get long a CPU stock, we got to get short a CPU stock. That's what everyone, that's what everyone's doing. And the reason is, you know, Jane Street, I just think it's bounty hunting, right? I don't think that there's any intrinsic value calculations there. That's just my personal opinion, who knows because it's invisible. But, you know, we're looking at the positioning of the market over time and trying to figure out what's underlying at all. And this is where the majority of our, you know, late night or long extending conversations go, I've tried to find an opportunity having an indication that the credit system broadly is way different than people think. And there's massive risks built up there over a very, very long bull market, right? You know, minor reset in 2018, minor reset in the grand scheme of things in 2020, minor reset in 2022, not the true leveraging of the bear market, right? So that's building up in the credit system. I have to give Liam props and he's in the discord and he's had plenty of, you know, voice calls about this, but the plumbing, the actual collateral system, repo markets, broadly, like he explained some of this stuff to me, understanding that allows you to be a better macro investor and figure out you're happy to be in cash and you're happy to be short and directionally quite short. That's something that the vast majority of the market doesn't have. For an allocator, it's like, you know, I can put in my money in private credit, I could put my money in public equity, I could put my money in private equity, you know, I could make a corporate bond portfolio, right? But if you're doing the alt side that's not private equity or private credit, all the hedge funds are generally the same and the vast majority of single managers have somehow you know, decided that, you know, they're going to be value investors or it's just significant underperformance quite frankly. So I do think that there's a very different type of allocation that you provide without leverage. And sometimes quite frankly, on the derivative side, I'm like, Liam, if you think that, I'm going to buy calls. You should buy calls and he's like, I don't buy calls. And sometimes they pay out pretty good, especially when I recognize that there's a lot of truth to it. Switching gears, okay. So we decided that AI is the macro, right, without Nvidia, we're all dead and unemployed. But we do have AI, it's unavoidable, right? And it's pretty incredible, the shift, and we came out with a memory war, right? Is it about valuation? Is that what's driving AI stocks? Well, look, I mean, I guess at a very meta and base level, yeah, it's always going to be about valuation. But what I think that the market is constantly trying to figure out is kind of a couple of quarters out, or even longer than that, depending on the kind of framework through which the market's viewing duration of an asset, which by the way, changes across the market. And I think people don't fully understand that, like clearly a lot of the optics part of the market right now is trading in 2029 or something. And that might be appropriate. And then there's other parts of the market that are trading maybe in a much shorter duration. A lot of that I think has to do with backlogs in the market's assessment of how dependable those backlogs are and maybe what they think is going to have to happen in terms of generations of technology and the evolution of technology and what that might say about whether or not those backlogs will be converted on. But that's been I think maybe one of the issues that people have had during this cycle is that they're really making comparisons that are not even really apples to oranges. It's like apples to pinball machine. Like it's they're completely different categories of company and they play very different roles in the ecosystem and you can't apply traditional valuation metrics to them because of how quickly the capital is flowing and moving and also how quickly the technology is moving. I mean, we after the agentic explosion of kind of mid 2025 through the economy and the boom in cloud code token usage, that led to such a meaningful change I believe in the way that that businesses were looking at CPU allocation and memory allocations that I think you really caught a lot of people off guard in the stock market. And so, you know, that's the type of thing that can in the near term break down these valuation paradigms and maybe it'll take a few quarters for the cell side estimates that those views are based on to reprice. Things are just moving much more quickly than really ever before or at least as far as I can remember in markets. The shifts that are taking place, the cell side is just not equipped to provide you with a reasonable estimate in my view of a lot of these things in real time as they're happening, which is why I think a lot of the market has tried to turn towards third party research, independent research like semianalysis and such. and irrational analysis and damn name and some of these guys, because they're literally saying they're saying, hey, the rumors from inside of semi-analysis about timelines on CPO or about whether or not AMD's chip is gonna be ready on time or any of these things, that rumor is actually in this speed of system more dependable than a firm cell side research number that is going through kind of traditional channels. And so kind of a long-winded answer to say, I think it is about valuations, but I think that because the valuations, because the numbers are so undependable because of the speed of the revisions and the speed of the changes and how the technology is diffusing, it's kind of making those numbers that you can check on your Bloomberg less and less reliable as kind of real tests of what valuations folks are giving to different merchandise in the market. And so that forces you to rely, I think, more on technicals and news flow and narrative because you need to be directionally correct on things more so than trying to really be precise about where an estimate is gonna land. I look for numbers on my Bloomberg that are quite frankly wrong. I don't think there's money to be made by reading numbers off of Bloomberg. And yeah, I mean, if you look at the cell side, they're getting absolutely frame mugs, like Benioff by sub-stackers and a lot of credits all the sub-stacks that we just named. If you do a reverse DCF and you try to think about the assumptions that go into some of these valuations, like you're either a singularity guy or you have to be extremely, you have to extend the runway 10 years, not a five-year DCF, a 10-year DCF. And then you're toggling, you know, ultimately the discount rate, right? The over five-year period, 4% of S&P companies grow revenue over 10%. Carrier. So you can't just put a terminal growth rate of 10% unless you're a singularity guy. So a lot of the valuations don't make sense on their face. What I do as an investor is I look, can I actually get behind the valuation here? And do I think that there's gonna be massive net revisions coming up in the next year or two? Not the next quarter, I'm not the next print guy. I'm, you know, two to eight quarters out and trying to figure out where the number on the Bloomberg is gonna go. It's not set in stone, you know, you have a team at Goldman Sachs and Morgan Stanley and, you know, for smaller caps, it might be Roth. That decide, here is the number. And it goes in with, you know, the other analyst at their desk and that's the number on the Bloomberg. It's not a source of truth. It's a source of expectations. And the difference between that and reality is where both of us are looking for opportunity. We met together on the memory war after, quite frankly, you called me and I was like, this is really good. And we've been batting around like semi-caps have to go, right? And it was over the past quarter that it seemed like everyone was super bold up on semi-caps. They kind of super spilled out. And then the prints were really, really good. And the stocks did not respond. If we're segmenting out semi-caps, right? Different semi-caps have different exposures, you know? We said, you could buy a micron. You could buy Asuka Honex, you could buy the memory, you could buy the GPU. AMD and Intel is ripping today on the CPU side. The networking, they had a great run. You know, if we're constrained by memory, you have to scale out and scale up. So there's more of a need for CPU. Let's highlight, right? Because we have a massive memory shortage. We have 90% gross margins from these guys. That's never happened, right? So clearly there's a shortage. It's indicative in the pricing power. You could argue that in video GPUs could have charged in order to get 90% gross margins when they were the ultimate choke point. But they ramped up, they built out CapEx with TSM. And then quite frankly, they're probably subsidized their chips a little bit, based on what the going market rate would be. Now we have a dynamic, it's really all about memory. Can you explain the choices when dealing with that memory constraint? - Yeah, well, I think that it's interesting that you frame it that way, because I think a lot of the underperformance in the semi-cap space, it kind of makes a lot of sense. And I think the first reason is that there's a fear around kind of a zero sum way of solving this issue. I think part of it is the concern that perhaps you could get later, the cycle could be choked off by extreme memory pricing. But then there's also concerns about de-specking HBM from systems moving forward. And that might reduce the long-term need for a lot of green field expansion and new way for capacity. I think also, and this is a little bit more of a, maybe a constituency investment view. I think that final big surge that happened in semiconductor capital equipment, kind of from the beginning of this year into May and June, I think that there was a lot of generalists who kind of figured, okay, I've missed on memory, I've missed on some of the ways to play optics in this space. But these companies have fantastic kind of compound or type of investment profiles and I know it's oligopolistic and it just seems like a one-way bet that we're gonna need more chips and so I'm willing to pay up for this stuff. And so I think that there's some guys who kind of, now are gonna wash out on the other side of that who kind of got poor entries. But look, I mean, I think that what really attracted me to the view was that basically the capital intensity ratio of the memory industry has completely collapsed. And I can't decide whether or not I think that that is a new structural phenomenon or if that's gonna be something that fades as the cycle goes along. So I think we put it in the piece, but the past couple of kind of major memory booms, going back to the mid 2010s into kind of 2018, which was really big memory upcycle. And then in the early 2000s, you were looking at kind of these memory IDMs. This was a mix of data sources that we were trying to estimate from spending kind of anywhere between kind of, I think, 30 and 40% of their revenue on CapEx. And there was even a downcycle in there that we looked at kind of right after their financial crisis in 2009 to 2013 that I think that they were spending in kind of the low 20% of revenue on CapEx. And last year and this year, they're kind of at, I think, 17%. So they're in the biggest boom they've ever had. They have these long-term agreements and these kind of special customer agreements that they've worked out to kind of offload inventory for the next few years at, I think, Micron's price floor is above the highest pricing that they had in any previous cycle. So these are really, really advantage kind of setups for these companies. And then maybe it's just because of this extraordinary surge of revenue, but their CapEx plans aren't really lagging that in terms of as a percentage of that revenue. Historically, they have looked at it and said, hey, if we spend super heavily in a few years, we might regret it. In this case, I think that there's some reasons to believe that they may eventually break down in that belief. They're starting to see memory as being, I think, less commoditized going to high bandwidth memory or architectures, they're seeing that there's kind of this platform shift type of dynamic taking place that theoretically should continue as a priority at these large-hyper scalers, even if there was a recession. And so these may be longer lasting assets for them if they were to build new greenfield sites. And they also, at this point, they actually have to be a little bit careful about not spending, because if they don't spend, it can increase pressure to despec and to build around in terms of the topology of networking in a data center, maybe using a shift in the memory hierarchy. Every company outside of memory has noticed that it's a rewarding prospect to raise the idea that they're going to be a company that helps solve this shortage. And so it's either going to be that micron and escalonics and samsung solve the shortage by paying a lot more money for a lot more etch and deposition tools from LAM and for, you know, high NA, EUV from ASML and even, you know, just building fresh greenfield wafer capacity so that they don't have to cannibalize as much of the legacy wafer count to be stacking these, to be stuck in DRAM or somebody else is going to solve this problem and so they need to kind of figure out what's the right amount for them to begin to, I think, break this discipline that they developed maybe as a result of the really, really nasty down cycle of 2022 and 2023 in the memory market. So you have two choices. Everything can be unwound by the macro, you know, diesel prices. We'll talk about the downside risk at the end of the podcast, but, you know, again, diesel you get the consumer, you got, you know, just generally, you know, capital, how available is it towards all of these, you know, buildouts that could unwind everything. So excluding everything downside, you know, if demand stays high, you either need more memory or you need to figure out how to use less memory. There's an engineering approach and then there's just a pure, you know, increased supply approach. So people don't seem to understand about semiconductors is how hard they are to make at a yield that is competitive in the market. You know, you see a lot of, I see a lot of small caps that, you know, say they're going to scale. It's like, you haven't done anything, Aluma, you haven't done anything, you know? So when we're talking about memory, memory is particularly capital intensive. If you're going to build an optics, FAB, it's not going to, it might cost you a billion, five billion, maybe two billion. A memory FAB is 50 billion, you know, just even remotely have the scale to compete. So that's kind of why that maintains a choke. Otherwise there would be a new entrance, simple economics. If price is high enough, new entrants are going to come into the market. The problem is it's hard to enter this market and it's just simply hard on a scale and know how of the workforce that you would need to do it. So as of now, maybe in the future, there's a Sarah Bruss, but there are three companies, and there are two that are in South Korea, and there's one, Micron, an American champion. Can you describe, first, very quickly, what are the engineering workarounds? And second, excluding that, because maybe these memory guys decide that they will be worked around, it's not a question of, we're going to bust, but it's a question of, we're going to bust ourselves, because if we have prices this high, it's encouraging competition, and it might not be another memory maker, it might be a new solution. So could you first talk about the engineering workarounds and then how you see the competitive dynamic right now? Yeah, totally. Also, I would throw in that there is a very important fourth, which is CXMT in China, which, you know, it has been working on its HBM solution. It's a lot of potential wafer capacity on DDR4 and DDR5, kind of a more traditional DRAM, they are making breakthroughs on HBM, not a scaled supplier at this time, but definitely something to continue watching. Yeah, I mean, look on the engineering workarounds. I don't wear away lab coat. And so I'll give all the credit to plugging, you know, substacts like a rational analysis and like Damning, who I think does the very best full explainers of all the routes that are being pursued in the market, you know, he has written up, I think, literally almost all of them from CXL, which is the express lane memory, basically, kind of using software solutions to use or to create more efficient, you know, memory processing. Obviously, optics, you know, very important from a bandwidth standpoint, you know, high bandwidth memory is very memory intensive in terms of manufacturing. You can rely more on kind of easier to make and easier to procure memory and then solve for bandwidth using better scale up and scale in solutions. So, you know, co-package optics could end up making a major difference in our regard. And then I do think that there's ways to de-spec within the actual, within the models themselves. And that's been something that I think has not been fully exposed, but, you know, we're going to get to a point, I think, where, you know, the engineers at OpenAI and Anthropic, they'll continue to ask the models that they make and the more and more advanced internal models how to make themselves more immune to the memory shortage. And so, look, I mean, working around the memory shortage is going to happen regardless. It is happening, but the, I think the point of our piece was to say that if you imagine a truly maximalist view and you take the absolute ball whip of all ball whips of this entire thing, you know, really the ball whip is usually found in the, in the long duration, long cycle asset. And so, you know, when you look at iCore systems or when you look at ultra clean technology and these subsystems providers that are really critical for Greenfield fabs, you know, lane research or apply materials go out, they get an order from, from micron or from, or from Samsung for, you know, a fab that is going to be under construction and won't be finished until 2020 or 2029. And then they're turning around and they're ordering sub, you know, sub component things from, you know, the sub component providers, the subsystem providers, like some of them that we named in the piece. Most things are going to have the kind of the most violent, I think, whipping effect in the AI space and we're seeing that kind of right now the way that they've kind of unwound. But if you were really going to sit back and say, okay, let me imagine a world where the AI investment cycle actually makes it into the end of the decade. He get to the $4 trillion cap ex per year number that I think, in videos, CFO socialized earlier this year as being, oh, maybe that's where we'll be by the end of the decade. If we were to get there, I think that the kind of the best risk or war that you can get at this point would be something that has kind of really had a meaningful multiple contraction and is still pricing without tremendous faith in the Greenfield story. You know, when you look at a lamb or applied, I think that there's this thought that part of the reason why Micron and SK Heinix and Samsung might not really go for it on wafer starts is because, you know, they have to have some view of where the market's going to be in three years' time. And so, if you imagine a world where that dynamic breaks down and maybe that'll, you know, require more certainty on, you know, cap ex from, you know, the key players and maybe you need successful IPOs from Open AI and Anthropic and, you know, maybe you need to start to see really tremendous breakthroughs in terms of operating income or operating results at companies that are adopting AI, you know, kind of give a new life to all this and maybe it can't happen when the feds hiking interest rates. But, you know, I think that at some point, the thing that's underpriced in the market now because the engineering work rounds, those things to me are very well understood. They're priced. The thing that I think is underpriced is this concept of if we were going to go back to the same level of capital intensity of this industry and if there was going to be a market share battle between these kind of three major players, you know, that capital intensity number would have to roughly double to get back to being kind of in line with historical capital intensity, that's an enormous, enormous amount of wafer fab equipment spend, you know, into the kind of half a trillion dollar range, you know, I think in the next couple of years. And I don't think that there's a single semiconductor analyst that has that as even their bull case for, you know, the semi-cap equipment companies. So it's definitely something that needs unlocks and it's definitely something that could not play out. But I think that when you look at history as an investor, especially in technology, you know, a lot of people like to say this time is different or the most dangerous thing to say is that this time is different. That's not just during a bubble when you're looking for the top. You know, sometimes that actually helps you find opportunities. too. And on the long side. And so can the memory industry actually continue to persist in this level of diminished capital intensity? I don't know. And if they can't, there's going to be a huge opportunity, I think. One way you can get to that four trillion cat-backed number is you just 10X the memory prices again and then keep the same quantity. So, you know, you could do it that way. As far as, you know, the engineering solution, let's go dovetails right into our piece. It's inevitable. It's like Thanos. It's I am inevitable. There will be technological solutions. What memory is worried about and what the real equation is for someone who's sitting there with this potential question is, can memory innovate fast enough, kind of like Nvidia has, right? Where their HBM6 beats CPO or offsets it enough for them to maintain the share of spend going towards the hyperscalers, right? So it will happen. It's just a question of a race. And if we're going to describe the game theory here, you think about it. There's three companies, right? They're functionally acting like a cartel. I'm a little bit more cynical on it where I'm listening to the micron in December 2025 earnings reports. And I'm seeing the beat. I'm seeing the cap X number and I'm like, what is this? And then you see the call and they're like, well, we don't want to get burned again. It's like, you just guided for, you know, I think at the time it was $20 of EPS and the next month. I'll say, I think that a rational analysis actually put out a piece just recently and he pointed out, I actually forgot that this had happened, but like they, they, Sanjay, the CEO, Sanjay Muraf try to read a micron. He actually was pretty publicly like pleading with Samsung to like cut wafer starts back towards the end of last down cycle. So like, there's definitely, I mean, look, I don't want to get sued. So I do not think there's an illegal cartel here, but I think that there is a, you know, like we pointed down the piece, there actually was a price fix in cartel in the in the late 1990s into the early 2000s that came apart because micron defected. But there's clearly, you know, this obsession with, with discipline. And I think it has something to do with the fact that like, you know, I think that the financiers are in charge of the company now, not the technologists. And it's the same thing that happened in the energy industry in oil, where like, you know, it's not landmen that are in charge of the country, the company anymore. It's, you know, it's guys sitting in boardrooms in Manhattan who are saying like, hey, let's keep cap X low and like, you know, let's not spend into a hole here. But I think you start to see signs of that breaking down because of the new type of memory that is being created, because of the platform shift. And it puts a lot of pressure, I think, you know, we were going to say on micron as the number three, you know, kind of to be, you know, looking at it and saying, are we comfortable being just like the persistent also ran number three, you know, really number two, you know, with 25% or less of market share across DRAM products, 75% plus in, in Korea. We're the natural partner for American hyperscalers. We are the company that should be benefiting the most from all of this. But meanwhile, you know, Nvidia is getting referred pricing from SK Inex. And, and Samsung is now taking market share from micron after getting qualified into into Nvidia's new systems. And so that's kind of the where we see the opening that that micron might be willing to take the outside money that SK Inex has to date apparently refused to take. There is this rumor of Intel and a hyperscaler JV. But we think that micron would be, you know, much more certain to consider that type of price. Yeah, a McKinsey consultant might say, oh, no, you just keep supply tight. Prices go up forever. There's no problems. But we're outlining the engineering work around the all three of them are probably thinking about. They're like, how high is too high on prices just for that matter? And then, functionally, there's entrants in the market in China. I write it off because I think that the Trump administration will just ban it. But it's worth a consideration. And then it's one of them defective. And we've written a piece where we believe that micron could be the defector. It could be Samsung. It could be SK Inex, although least likely. That would lead to the same outcome. You know, we might get the defector wrong and SK Inex with Intel. There's some stories out about it to understand this for everyone. Hyperscalers are offering the money to build out these fobs. Like, hey, we'll give you the money. We'll guarantee that we'll purchase the memory. Just build it, man. And they're still like, ah, I don't know. That's how much this cartel quotes alleged cartel has been holding. And just a quick little story, you know, the price fixing stuff. My first 20-bagger in the small cap world was capacitor company that had a $25 million suit that I knew. I was working with PJ Solid out of Potomac Capital in college for an internship. And we knew it was going to get settled. But like this stuff, I say that because the price fixing aspect, it always comes to a head, right? It's never been different. If you look at the prices, eventually something is going to be worked around. You just need to go to, you know, intro to microeconomics in college to figure out how this works. So let's talk about the Thanos versus the Avengers. Because Nvidia, Nvidia, Jensen and Nvidia, they are Thanos, right? And on Thanos' side are SK Inex and now Samsung, since they got the certification. So South Korea and Nvidia, Nvidia doesn't want more memory supply. If you think about it, right? They have $200 billion or something crazy, commitments, guaranteed. Google a year ago went to SK Inex and was like, can we have some memory? We need more memory. And they said, ah, walk, walk, walk back out. Google fired their whole procurement team, okay? On the other side, right? On the Avengers, you have Jalapana. I like to call it Jalapana. You have made by Broadcom and, you know, not just chat GPT models work on it. It has actually great specs for multiple models, which is different than what you would say about an Amazon ship or a Google chip. But there is also Amazon and Google. They all need memory. So it seems like on the other side, AMD as well, right? Intel. Potentially. I don't know where Intel sits quite frankly. But there's this incentive for more memory and they all want more memory. And Google would literally give micron $50 $100 billion if they needed it in order to guarantee the supply on a long-term agreement with the price floor. That's above the last cycle's peak, right? So that's the game theory. If micron decides to defect or if it's S.K. Onyx, what does that mean for flows of this CapEx and ultimately security price changes potentially, you know? No guarantees. Yeah, I mean, I think that there's a couple of things that happen downstream from there. I think that the first effecter that is willing to take outside capital in a really meaningful way to build the Greenfield Fabs. Because by the way, we've already seen micron raise its own outlook for how much HBM capacity they want to build. So they're pressing forward on this. But I think really what the market's looking for is Wafers Greenfield. And I think that the initial thought is that if one of them defects they all do in order to maintain market share. So either everybody has discipline or nobody does. And then I think that the next thing from there is that, you know, there would be a high level of competition for relatively few tools looking out. Because really, you have a lot of other fab projects that need similar tools and similar subcomponents. So, you know, Elon's TerraFab, which actually is real or the I thought it was when he first announced it. I want some I conductor and they're pushed for catbacks which, you know, they're kind of moving along on New Greenfield projects. And then, you know, SK Intel, which is building fabs in the United States, Samsung, Microt. So it would just be a rush for a backlog of tools that at this time really is pretty limited. And so the operating leverage at the subcomponent players I think would be very meaningful, you know, for every new tool that LAM and applied and KLA corporation and such would need to provide. There would be a lot of competition for new supply from the subcomponent makers and the subsystem makers. And then there would be a strong pricing power at ASML and at LAM and at some of these companies because there's no matter what there would be, you know, relatively fewer slots. And so they would basically capture just all the upside and pricing. So it's expensive stuff. It's long cycle stuff. It is lumpy. It drives big operating leverage. And so it could lead to, in our view, very, very meaningful revisions in the in the earnings outlooks and the long-term earnings outlooks for these for these companies. That is kind of why in the framing of these companies just coming off 50 to 60 percent, you know, some of them and they're them having multiple contractions of 30 to 40 percent during that time, it sets up a very interesting, very interesting kind of setup that could make for one of the more explosive investment setups in technology in a very, very long time. It kind of just goes back to this place where in 10 years time, if you look at the middle part of the of the 2020s. And you say, okay, like what was the really obvious thing? And you remember that there was just miraculous shortages of memory of CPUs, of GPUs, of advanced packaging. Of all of these things, advanced packaging, obviously also needing inspection and metrology, which are also critical tools that the semi-cap equipment makers make. If you just looked at that and you said, okay, the cycle survived, what do they need? Oh, they needed more tools to build more chips. And you wake up in a world in 2030 or 2032 and there's a huge chip glut. And you ask yourself, did that chip glut come to be because AI was a worthless technology that ceased to grow? Was it because there was a massive depression? Or was it because they built a ton of chips, like a ton of chips? And then there was some sort of recession or as there always is some sort of down cycle. But it feels like it's more likely that would come on the backside of much, much, much higher wafer fab equipment spend. I don't think that we've ever seen during a platform shift, a bust, or a glut that happened without first having a boom in supply of that thing. And so the fact that we're not getting that same level of capital intensity kind of makes me imagine, okay, work backwards from the glut. How do you get there? First, we're going to need to have a boom in terms of supply. Yeah, that's a really interesting reverse engineering. What was a Gavin Baker say that we might be saved from a bubble by shortages? Yeah, I think that the discipline of TSMC and their unwillingness to search advanced packaging capacity at some remarkable level. And that has been the gating factor of how many Nvidia GPU systems are out there. And I think that it's a similar dynamic here, even though it's complex and even though it's expensive, it's not that complex to bring online a ton of memory. It's expensive. It's long lead time. You know, you have to, it's a ballsy move to do. But that's how Game Theory works. Prisoner's dilemmas break. And even though the move is risky because there's no other move. And so in this case, I think that that's, it's more something to have your eyes on amid all the risks to the downside. You know, I think that as an investor, you know, you like to think about opportunity using optimism and you like to think about risk using pessimism. But what really I think makes it easier as an investor for me to judge these things and not be colored by optimism and pessimism is view it all as risk. You know, where is the enormous upside risk? Where's the downside risk? And this strikes me as one of those where there's a severe upside risk that is under price that doesn't require a tremendous amount of optimism. But I would be very paranoid about being short, some I conduct a capital equipment or being short, you know, kind of the companies that are going to benefit the most from bringing on a ton of new supply of memory against this level of capital intensity, which like we have said is unprecedented in terms of how low it is versus historical booms. So to bring this elementary for a second, it takes a long time to build memory fabs. I mean, we're talking two years. But you have to order the equipment that goes into the memory fabs. Those also take a very long time to increase supply, like an extremely probably, maybe even longer for ASMR, maybe not as much for for a mat or lamb and then probably lower down the chain is probably less. But because of that, that's where you could get pricing power there. Leon mentioned it. And the great thing about pricing power is you don't have to increase quantity. You just raise prices and it flows through the bottom line. So what would break this cartel right now is $100 billion from Google guaranteed. What is my current going to do? And once they get 50 or $100 billion from Google, what are they going to do? They're going to try to lock down supply. And as soon as they get that first deal, the first deal is always the biggest bounty, right. Then Samsung and SK high next are going to be immediately called up and they don't have the money, but they're they're supposed to be buying back shares. They do have a lot of money, right? They have a lot of cash flow. So they're going to be competing for the same semiconductor equipment, right? And that will lead to a step change. In the growth rate, but the growth rate is not just you know, quantity. Quantity might be going up 20% per year, but all of a sudden prices have just been jacked up 40, 50, you know, memories up 500% or more, right? So it's a gross margin story in a big way. It's volumes pricing and gross margins. It's very similar to what happened in the fourth quarter into the second quarter of this year in the memory stocks and the way that they benefited not only from selling an increasing number of memory chips and high bandwidth memory, but also from mostly from, you know, incredible price increases. And yields are incredibly important, especially when it's a green field. You know, there's always a risk that it's not going to have the same yield as, you know, you're already up and running for, but for the average listener, sometimes they just don't think about the fact that the vast majority of growth is just coming from price increases, right? In a video, in a video's case, they're actually producing not many more chips about, you know, they did just raise prices, but the price times quantity plus efficiency equals the gross margins, and then it becomes the gross margin story. Yeah, you know, the one, the one caveat I'll say too is that I think that, you know, for somebody who might push back, they'd say, okay, so, you know, that much money coming from Google, don't these hyper-skillers seem like they're a little tapped out, already on their CAPX plans? And that's why I think that what would be more likely is that you'd see a financing partnership. And you'd see KKR and Apollo and, you know, Blackstone get involved. Because this is right up their alley in terms of, it's a, you know, it's a multi-year build on an infrastructure, on an infrastructure project. They have a lot of experience in that. They love investing in that stuff, and it's a very long-life asset once it's made. So you can imagine a world where they provide the financing partnered with a Google or with a Broadcom, like they did with Broadcom on Compu just earlier this year, and then Micron is going to operate the FAB, and Broadcom or Google is going to guarantee the take or pay arrangement for the supply. And so that kind of, I think, would be the way that you would see it, you know, all come to pass. A balance sheet is going to be leveraged, right? And, you know, it's looking like increasingly, the hyper-skillers are not going to have non-recourse access to capital anymore. And just, you know, maybe for this, it's a little different. It's like kind of like a, like a Vernover turbine, where it's still easier to underwrite versus, well, we've underwritten so many GPUs. What if we tweak our assumptions on the liquidation value of this? Like that's offsetting our portfolio. Well, this is much easier to underwrite, much easier because you're literally, you can turn to your board and your shareholders and say, we're getting host hundreds of basis points a quarter on memory prices. We're trying to solve the problem. We have an exclusive arrangement here on this supply. Like, you know, who cares what it costs up front? 50, 100 billion. million dollars, it's going to save us hundreds of billions over the long term. So it's easier to underwrite, but I would still say that it seems that, given the backdrop right now, the hyper-scalers have gotten a little more ginger about going out and kind of putting forward long-duration capital in the tens of billions of dollars, it feels like an odd time in the cycle to be doing that. Right, and so just a twinge, I think we've given people plenty to think about, but and we'll have part two on the downsides, more macro stuff in the economy. But how would you overview the risks in this system? Obviously everyone's heard my private credit and insurance views. Aside from that, you can briefly mention private credit and where they get the funding from. But how do you see the overview of the macro and what would end this from your perspective? Yeah, I mean, gosh, I think that and I should come into this, you know, just being very, very transparent in my bearishness, which is when it comes to the memory war and when it comes to that thesis, that's understanding an upside risk that can lead to a major repricing. But there's a bunch of reasons to be very suspicious about where we are in the cycle. Diesel over six a gallon and, you know, over 200 a barrel and, you know, that kind of being the biggest commodity issue of the current environment, driving a lot of input inflation, you know, 75% of freight in the United States moves by trucks, all of them use diesel. So, you know, that's part of why the Fed, I think, feels some pressure to make sure that these inflationary pressures that are coming up in kind of idiosyncratic ways don't spread into a more general inflation. But, you know, outside of that, I think that a lot of investors, and I think, you know, a lot of investors that are not professional investors, and even professional investors, by the way, because this is what makes a market. I don't think that they have really caught their bearings in terms of if you've been around markets and have studied economic and market history, just how violent this earnings boom has been. And I think a lot of investors have kind of become somewhat accustomed to the pace of this earnings boom, and they kind of just assume that, that hose that we talk about is going to, it's going to go somewhere new, and oh, what stocks am I supposed to be at next? And maybe it's going to be these stocks and that's, but historically, this violence of an earnings boom tends to have, you know, a bust on the other side of it. And if you look back at other historical booms around, you know, new technologies, they all have different colors to them in different stories and narratives and characters and companies. But if you were watching them from a million feet up and you're just looking down at them, it's an earnings boom that an earnings bust. And so there's a bunch of theories that run around today about where that could come from, from, you know, these labs are never going to be able to outrun their debt. And eventually that'll come calling to, we're not going to build the actual data center capacity in time. It's going to lead to a glut in chips because there's going to be a gap between infrastructure delivery timelines and chip delivery timelines. There's people who are suspicious of scaling laws and saying, oh, you know, the technology is not going to continue to advance and become more useful. There's a lot of different, you know, kind of bare cases for why this earnings boom might unravel. And I think the one that I've become most concerned about is one, I do think that this is a long duration market that says, okay, there's an earnings boom now. But in order to have any level of valuation tied to these things, even, because it's such a clear level of over-earning. And, you know, people can say, no, it's not. We're going to continue to go. But the market at least, I think, understands that there's over-earning happening, or at least a high probability of over-earning, which is why they have Nvidia at 16 times four. Because they're just like, how can we give them the benefit of the debt on anything past that? And you see the memory and you see it across the market. I think that if you're looking further out at 2027, 2028, 2029, just how much can the economy get pinched by geopolitics, by very high diesel prices, by very high real interest rates, a 10-year real rate that's now at a multi-decade high, a Fed that is clearly not in the business of providing to put any more, and is trying to regain credibility on fighting inflation. So, you know, if there's any beta to the economy of this boom, I think that can start to be called into question. And then the other really big risk to this earnings boom that I think people are waking up to, but I think is under price in the market is politics. And right now, if you're in DC, and I talk to people in DC, all they talk about in Congress is AI. And it's become the topic to sure. And, you know, immigration's not the thing, tariffs aren't the thing, even Iran's not the thing. You know, that gets talked about, and it gets hit on. And obviously, energy prices aren't easy, an easy pitch for Democrats to hit at the president right now. But AI is really something that is becoming much more discussed because of the safety concerns. And also because I think the Democrats have identified that this is a good wedge issue with the president. And they're going to force him into a position. He loves to talk about 80-20 issues, whether it's transgender people in sports or immigration, illegal immigration. He loves these issues where he feels like 80% of the country is with them. AI pulls at 20% approval. And the AI industry is even more unpopular than the tech. And so, you come into November, you know, 90-plus percent chance that the Democrats get the house. Polymarket has them now over 60% chance they get the Senate. They get in there with an opportunity to work with the labs who want this regulatory capture, push for industry slowing regulation, put a bill on the desk of a lame duck president, and force him to veto a popular anti-AI bill. And for JD events or for Marco Rubio to have to stand next to that in 2028. And so, I just think that that's a risk that if I was a Democrat running for the Senate, I would be like, oh, this is like a huge slam dunk. Let me force the president to basically nuke the economy or veto the bill that would nuke the economy. But it's an extremely popular bill. And so, it's a really difficult issue for the president to get away from. I can see that kind of forming. We'll see if the market grows more concerned as we approach the midterms. But I think that that could be the first source of real friction in the market around the AI trade. Call them to question the earnings boom. And it's longevity, you know, because Washington chooses perhaps to get in the way. The other 80-20 issue is the Iran War that he's on the wrong side of. So you have the political thing. And one thing I would just put out there is it feels like all the power people are behind Trump and Johnson Wong, you know, hey, Mr. President, but in the lame duck term, it always changes. I think that people don't anticipate that. Everyone's playing for the next selection, right? All of a sudden, you see Mike Mike C. Mark Cuban instead of trying to get, you know, cost plus drugs through all the sudden he's playing for the next one, right? So politics. And then the other side is the real economy, right? Because AI cannot just go up. There's a $36 trillion economy. I can show you how I would see AI providing economic value in order to justify $2 trillion a year per side, right? But what that needs is other companies to make more money. Otherwise, it just doesn't work, right? The thing is that picture looked to me, honestly, like a window you could get through. And you could get through the whole despite all the leverage in the system. Then you have the Iran War. So you have a lot of macro forces that are going to cause under-earning. And otherwise, if they started earning and all the earnings are up, it's not just the AI companies, but it's basically all due to AI, right? But if that starts getting suppressed and all of a sudden the hyperscaler is going up here, and then you see a plateau inside of that plateau, there might be a lot of AI efficiency. But there also is diesel at $6 a gallon, right? And the consumer is getting crushed. spending $500 per week, basically, depending on how much they drive on gases, right? And then you have the housing market. So, if you were to see that part of the economy lag and inside of the, you know, not just the GDP numbers, but really like corporate earnings, then you can't justify the hyper-scaler spend, right? So, yeah. And that's simply, but you've seen it, you've seen it in the market immediately after the rate hike, you know, the market returned really kind of to this AI theme and to the semiconductors. And you've got a couple of days of squeezes that we've now been through. And I think that that's really just because it, in theory, is the only investible part of the market. But inthropic and open AI are the engine of this, you know, of this spending boom. They now make up, I don't want to misrepresent it, but I think that they make up over half of the cloud backlogs, cloud revenue backlogs for the hyper-scalers. And so, they need to continue to make money and to raise money either, you know, through private sources or through public, you know, going public, which has now been delayed for both of them, at least to November for inthropic, perhaps even into next year, you know, depending on market conditions around the midterms. But they need to continue to raise money and they need to continue to make money. And it's harder for them to make revenues to support raising money if the enterprises that they go to and they say, we want for you to spend a bunch of money on tokens and adopt our agents in house. If those companies are getting squeezed from other places, you know, they may be less eager to try to innovate right now with their IT spend. And so, it's just a reminder that early on in this cycle, when the spending was hundreds of billions of dollars or, you know, 150 billion dollars of CapEx or 300 billion dollars of CapEx annually, that was, you know, able to stay contained in a vacuum in one part of the economy that as you grow into bigger and bigger numbers, you know, it goes to anthropic and open AI. There's there's many people now who suspect that anthropic is really pushing to go public because they can't access any more private cash at these valuations. You know, maybe the sovereign wealth funds are all tapped out and basically they know, okay, if I'm willing to write these guys a check at, you know, over a trillion dollar valuation, my first mark might be down 30 in the public markets. And so, you're getting to this place where, as they become more central to the system, the rest of the system needs to stand up and needs to be healthy in order to support the ecosystem. It's push and pull, it's synergistic, you can't have AI pulling forward the economy, you need the economy to be in a place where it can also pull forward AI. So, that's why I think it's now worthwhile to think about the broader cycle. And to be open-minded to the fact that, hey, this may just end up having been a fragile earnings boom, by the way, that would not be permanent fud or permanent bear case for the AI industry or AI capabilities. There were earning booms and busts in the railroad industry and it recreated the entire country. And so, that can happen. It happened, obviously, in the internet. And so, we could have an earnings bust in AI and we could still get robotics and orbital compute and, you know, AGI and ASI and recursive self-improvement. And all types of incredible, we could cure cancer. But first, we could have an earnings bust. And I think that if we were going to have one, the macro thought would be we're getting towards the numbers that would suggest, okay, there's more vulnerability, which naturally will put you closer to the later innings of the game rather than the very early innings of this specific capital cycle. It can't just be AI cannibalizes the world at least for asset prices, right? If that happens, it will teach everybody a completely new way of things working. And then there'll be a permanent underclass. And it'll really just be Silicon Valley will have all the money. And it'll be like Elysium, like that Matt Damon movie. But it's just that is historically not how these things work. And I think that it's a very, it's a fanciful and self-important view to believe that's how this one would work. It can still be the most revolutionary technology ever while still living and dying by the rules of capitalism. And I would argue that it's more likely to than almost anything ever because it consumes more capital and is more intertwined with the capitalist system than, you know, past booms. And so that's a, you know, I think that investors should be at this time a little bit more antsy about, hey, you know, maybe I'm not just supposed to be looking for the next bottleneck. Maybe I should be considering the potential that, you know, the entire, you know, hoes is about to just give up, you know, spewing. Yeah, it would there would be a tremendous amount of capital destruction if it was that way. This energetic, we get a breather in AI, the real economy stocks start taking off, it's a rotation. That's like, it kind of happened. What, by the way, that kind of happened this summer. You know, you got this broadening out. But that broadening out, it really, it kind of surprised me when it was happening. And I don't think that it was a, it was an appropriate from a fundamental standpoint. I think that there was just money that was leaving AI because of the correction that was happening there in the momentum correction. Perhaps the movement to healthcare was justifiable. But moving into some of these real economy parts of the market, you know, felt surprising to me. There was this whole theory that we were going to get, you know, a big steepener with worse coming in and the financials would be investible. And we have continued to flatten. And so, look, I mean, from here, I think that there's, like I said at the very top, there's nothing as dynamic happening in the economy or in the markets as AI. The entire kind of direction of travel for me is as a kind of a macro manager who's focusing on mid to large caps. I think Nick, you do a huge service by providing people with a reliable stream of alpha in names like Groupon and Tenable and of course there and these stories because those things I think will increase in value as an information source as this kind of binary question of earnings boom and bust or, you know, I think that there's a bull case around AI or at least a favorable case where the earnings boom, but then the capex instead of it correcting, it stabilizes at a high annual level. And then it just becomes, you know, kind of a recurring, you know, revenue that is shifting around with market players inside of the AI infrastructure space. I would really challenge somebody perhaps they could like email me or something or get in touch with you, but I challenge somebody to find me a single earnings boom with this level of violence that has not been followed by some sort of earnings bust with some level of violence and instead has seen a consolidating level of spend. That would be it would be a real accomplishment I think of capitalism for that to happen. There's nothing new on Wall Street except for this belief that economic cycles do not exist anymore. In the past, it was just it didn't matter. Now people actually do not believe that they exist. So that's a new one. New one. The Kobe and Shaq handoff, that's something to pay up for, just in some, watch the politics, watch the earnings. There's so many variables. It's quite fun to be in this market quite frankly. On my side, I love the stocks I look at, but I would rather teach somebody how to do financial analysis, be differentiated, make money in the market, or even just get excited to potentially over 30, 40 years, just put money towards equities, even if it's an index fund. Liam and I, we love what we do. I love speaking with you. It's so great to have you on. You'll be back on. I'm looking forward to it and for those who don't know, Nick and I basically do one of these podcasts three days a week, but it's just us. And so it's fun to let people into kind of what one of our conversations looks like. We'll do more of them. Awesome. Thank you, Liam. All right, Nick. Appreciate it. [Music]