The Fireworks Show Is Over: Why This Is An AI Rotation, Not a Bubble Unwind

Let's start out with a shout out, uh, to the Nick fans, of which I am a long time suffering Nick fan. Very excited about what's going on.

Um, stocks fall. Is this a bare market or a rotation? Uh, I will go through a bunch of papers that I, I wrote this week, which, um, talked about a lot of the things I think that happened, but also about the way I'm thinking about the things going forward. And then I will release a paper next week for subscribers on the new economic cycle. I'm also working on a paper on the, uh, 800 volt DC. Uh, that'll be a new theme. And I'm also working on the application part of the five layer cake, which fits in with more on Eli Lilly, which you're going to see more and more about. And I did release a very long, extensive piece on Eli Lilly, uh, for subscribers on Friday. Uh, I do believe this is a very big deal.

Uh, S&P down 2.6% for the week. Basically the worst week since Liberation Day week. Um, Qs down 4 and a half percent, easily the worst since then. And again, just like in the S&P, the S&P was up nine weeks in a row. Qs were up eight of nine weeks. Then you get into small caps. They were down, but again, a very strong rally. So, none of this should be a surprise.

Um, yes, the charts, you know, you're going to have people come out and say that was it. Looks like a bare market is starting. Doesn't at all to me. Um, looks like this is the beginning of, of what I believe is a necessary rotation for those of, for those people looking. Um, I do believe that like memory stocks, it's very possible, uh, that those have made a top. And the fact that I keep getting pushback whenever I'm negative on memory at this point. I'll go through the reasons why, uh, and at least give you the bare case, not bearish for memory, but I do not think it is the place to have your capital. If you get a move down in memory 40%, I think your risk reward has changed. So, you've already had kind of a move, but remember I bailed out of my Micron in the 600 and 700 area. We closed in the high 800s. So, the reason I was getting out there is I already thought the risk reward was, I couldn't get doubles and triples anymore in memory. I still think you can get doubles and triples in Marvell and in Eli Lilly and stuff like that, and I will go through.

So, in the Qs, you broke the 20-day. The reason I have this arrow here is to show you that after Liberation Day, we went through this long period, um, much longer than what we just did. But you can also see it wasn't as sharp. And in this case, we finally closed below the 20-day. We just finally closed below the 20-day. We hadn't even really ticked it. At least back here, we kind of hit it and never closed below it. But then once we did break below it, that ended the euphoric period. And it happened in the summertime. This is just before the summertime. But then you did go through some digestion, and that's kind of what I called the fireworks piece, is that the all you can eat, uh, memory and semiconductor and infrastructure buffet, where you could just buy any of them and they would gap higher 10 to 30% because the market was not on the agentic trade and they were not yet positioned for it. The catchup is over, and now that everyone is basically on the same thing, which is what the fireworks paper was about. Uh, I just think it's now a two-sided market, and you're going to have to pick your poison. But I would not be getting bearish. And I'll go through the, the foundational reasons why.

The S&P, same exact thing. Finally broke below the 20-day. In the case of the S&P, it continued to move higher. I think the same thing will probably play out this way. I'd be much more worried about the Qs than I am the S&P. And I will go through the reasons why.

Um, another reason to not be too bearish is how strong the market was. I do believe that sustainable tops or bad places for the market happen with divergences and RSIs. And in the case of S&P, we made the highest weekly RSI going back to '24. So, I think we'll make new highs and make divergences like we did here. And I think until that shows up, I'm just not going to be that worried about it. Here's the weekly chart for Qs. Same exact thing. Highest RSI in over a couple of years.

Now, the big thing that showed up was that beta. This is the Morgan Stanley, uh, volatility, um, factor, long short, uh, was down 10 and a half percent on, uh, on Friday. If you take this back all the way to 2000, I think you have to go all the way back to 2000 to find anything that was even similar. So, again, it was a big move on Friday in beta. Um, this is pure. So, this is sector neutralized. The Morgan Stanley one still has sector exposure. This is the pure beta or volatility versus profitability or quality. And it was the worst, um, day since going back to Liberation Day last year. So, we did see a cleanse.

Um, if I was going to pick a place where we already have divergences. So, RSI, this is a lower peak than what happened here. The volume here in memories and Micron in particular was very high, and this looks a little bit more like a reversal pattern. But again, I've bailed out in here. I can see where people can go. Maybe we get down here and we bounce and we go through it. But I think we need time to go through this. And I will go through the reasons why I am worried about memory when all of a sudden every single person around the planet is positive on memory. I spoke at the New York Stock Exchange this week. There were a lot of brokers from Asia and in particular from Korea. Um, they all confirmed what we've read. Uh, retail has basically put in an all-in bet on SKH, Samsung, and the Korean market. So, since we were already seeing divergences there that I'd been talking about with sectors like construction and machinery breaking down, and they are trading at kind of one-month lows still, um, and the breadth was horrible in Korea. I do think that this story that I put out, the fireworks show is over. Agentic AI moves from discovery to digestion. So, I published this on, uh, on Thursday in the morning, and basically the firework show is over, and the last couple of weeks have been the climax of an incredible display, but the agentic AI buildout has just begun. And I kind of go through this as finishing up as my father taught me, the odds on the tote board are now fair. So, let's sit out the race until we see better odds.

So, for everyone who's looking, the thematic portfolio that I put together is based on the agentic world. It has much higher highs to go. When I talk about memory, that is one specific channel that has had a parabolic move. You could have the entire index go higher and memory just kind of bounce from a lower level, and it won't matter. Um, Eli Lilly is in the basket, and it was up on Friday. Uh, it was up last week. Marvell was up big last week, and it was up after-market hours, as I'll show why.

So, when I talk about the fireworks show is over, that just means the easy, let me just put my money on any of these, I think is over. And I think now you have to do your homework. It's a long short side, but I believe the agentic side will continue to outperform, but maybe for the next three to six months, it's, it's a much choppier, um, perspective. Uh, I will take you through what I started to buy this week, but get a chance, read the paper. I spent a lot of time thinking about it.

And the reason for all of this was really broken down by three events that I wrote papers after each one. The January CES, that was when Jensen, uh, introduced the Vera Rubin. It was a huge surprise to the industry and for the AI world and for investors. It had only been about six weeks before where they had been saying it was a bubble. So, at the end of October, the consensus was that AI was a bubble, that the capex wouldn't work. Jim Chanos was doing interview after interview, or at least, uh, posting. And then he did do one podcast, which I referenced, which was, we were questioning everything, and that's when Oracle started to peak, and everything started to go down. OpenAI was questioned, and there were fears that it would be under. Sam Altman did his interview with Brad Gerstner. All of these things happened just before CES. So, when CES came out and he basically talked about tokens per watt, the need of memory, opticals, inference, and he basically said, this is it. That's when the five layer cake was born. He mentioned it, but he really explained it in detail during the Larry Fink interview in Davos, which was a couple of weeks later. So, at this point, traveling all over, talking to mutual funds, talking to hedge funds, people didn't buy into the AI situation. By the time we went through the March Morgan Stanley Tech Conference, that's when I started to hear a shift from people. That's when they were reaching out saying, "What do I buy? What's, what's not too late?" And in there, it was because companies like AMD and Intel and Nvidia and everyone, Dell was talking about how big their backlog was becoming. And by the time we got to Computex last week, it was the same story. It happened. I saw nothing in there that was new. Nothing.

Um, and in the interim, Broadcom beats numbers, great results, stock trades down. Sarah Bloom's IPO comes out, trades down 50% from there. Google announces an $85 billion financing, as I'll show. Meta is now talking about it. You've got all of these IPOs coming out. And all of this is related to the exact same thing, guys. It's to fund this buildout. So, this was the surprise. This kind of stuff looks like more of a top, meaning they're coming out to raise money. It's not just IPOs like, "We have a great business, go buy our stuff." It's, "We need the money, not for monetization for the people that have been invested. We need the money to go buy more chips and more everything." But I think people are getting crazy and believing that the bottlenecks and the shortages and the stuff won't prevent them. Everyone believes we need lots of money. As I go through this, as someone who's been on this, I really do believe people are missing a part of this market that they're going to have to think about. And I will say it here, there will be a lot of the capex spend which is not needed. I've been writing about this since I wanted to just highlight. So, before CES or right at the same week, I wrote this thing on the inflection point when AI crossed the threshold in December. I wrote why 2026 is the revenue inflection year. AI agents are here.

So, remember again, as you're looking at me, whatever you think about what I've created, what's gone on. I'm not just trying to be contrarian. The odds have shifted. My father taught me, the top board has shifted. Everyone is in. They're now worried more about, "Did they buy something at the peak?" And it's because the certainty is there. And I don't believe anyone should ever have certainty over a three-year lookout, ever. And I will go through why.

You had speculation. There's no way to get around it. In the semiconductors, the gross option premium, the Cosby over the last, last six sessions, the Cosby's up 12%, yet breadth was negative each day. And not by a little bit. I mean, come on, guys. Like, at some point, it just gets completely ridiculous. The skew, the single stock put one-month put call has now collapsed to the slowest level in Goldman Sachs' entire database. So, it was easy to make money. Retail was involved, mutual funds were involved, hedge funds were involved. Everyone was playing catch-up. And remember, because of the Iran situation and how many strategists came in, I really don't think people jumped in, at least with one foot, until after the Morgan Stanley TMT, uh, side, which was after the beginning of Iran and when people started to see a bounce in Iran. So, that's why I think it was locked up.

Um, I, I've never mentioned John Hussman on here. Um, and it's because almost everything I've ever read by the man has been bearish. But I think his work is quite good, actually. Um, and I like anything that looks like this where I don't have to get people's opinions on things. And the only reason I bring this up is because for people who want to be super bearish, you can look at this. And I think this is the kind of stuff that I would build. I use the word cluster at the same time. I'm looking for warning flags. This is an element of what I like to do, which is I don't want to pick and choose. I just want to take as much data as possible. And again, he did have a lot of things that went up. So, if you want to be bearish, if you want to read this and ignore what I'm saying, all good. You got a lot of signals in here that occurred during bull markets. So, I wouldn't sit there and go crazy on it, but I thought it was interesting.

So, here's where we get into the reality of stuff. These are all facts. You're going to have to expect this to change. S&P earnings. This is what has happened this year, guys. And again, normally what goes on is this. Normally, every year we go into the year and then we start revising the numbers lower. It is like clockwork. Well, this year, up. So, normally everything goes down by about 2%, and this time it just rocketed higher. This is overlaid with nominal GDP, which highlights the profit margins. It highlights the productivity gains. I think you're making a huge mistake getting in on the train of AI.

Uh, forward earnings, nominal versus real, and what happens in a recession. So, I'm only showing this because I'm going to start showing you guys this more and more. Before a recession occurs, before some kind of big problem occurs, you start getting leaning lower. This one is really important, and I'll, I'll let you know why. This is the dot-com bubble. So, I've shown before that things didn't get bad in the dot-com bubble until late in 2000. The NASDAQ went down, and then the rest of the market went down. So, I'm looking for a rotation, not a bare market. If the rotation fails, it's going to be because the economic data fails. And I'll show you what that'll look like.

Here are semiconductor sales. These are not PEs. These are sales. Profit margins. All these yellow lines before we get recessions, guys, even in the dot-com bubble, here's the dot-com bubble. Profit margins go down. Profit margins make all-time highs. And not only that, there is no mean reversion. For 50 years, your, you, corporate profit margins always reverted. You have to think about '22, '21, '22, '23, '24, '25, '26, guys. This is when ChatGPT was launched in late '22. Sorry.

Forward earnings change versus is the market expensive? Here's the S&P on a forward PE ratio right now. This is not a bubble, guys. I could say it again. And if the S&P were to fall 20% and earnings don't go down, here's what we'll be. So, again, the reality is when you have these corrections and what goes on, I just want you to think about them. COVID, that was a decision by the government. Rate hikes, that was a decision by the Fed. Tariffs, that was a decision by the government. Unless you believe that Donald Trump is going to do something. So, maybe we get a correction. And the reason is because oil prices go to $300. I got no problem with that. It's on the distribution, but if it does, some other things are going to show up in the market first. Credit spreads never lie. They always lead. Very seldom have you not seen in history. Credit starts to go because that's the way the capital system was worked on. It's all changed. And this is what has happened while profit margins have been exploding, as credit spreads have just continued to go down. Jobless claims, same exact thing. Look how cyclical they are. Every time before we get a recession, we get jobless claims going higher. Even in the dot-com bubble, we haven't had a budge yet. Nobody's getting fired and and getting, uh, claims. PMIs, another thing, and this is overlaid with capital goods, just showing the massive expenditures going on in AI. Do you really believe that the expenditures are going to stop in AI? You're nuts.

So, credit spreads are near all-time tights. Jobless claims show no growth in insurance claims. Earnings are growing rapidly, and revisions are growing. PMI manufacturing just came out. Blah, blah, blah. Are these conditions normal for a sustained bare market? No. What you're describing is close to the photographic negative of a sustained bare market. Guys, don't listen to people in their bearish things. Don't pick an economist who is perma-bear and read what they're writing. Throw it out. If you want, copy what they do. Take the snapshot and put it into all of the LLMs and ask them all the same questions. I wrote something last week that I use LLMs to do this because they don't have a bias. And someone said they do have a bias. They don't have a bias when it comes to calling recessions, guys. Just put them into all five then and take the average of all five for the answers. It's an easy way to deal with it. The honest caveat, the one variable missing from your list is valuation, which I didn't include in there for a reason. And that's because 2000 was one. But even in 2000, it only became a sustained bare market once the macro broke. Claims rose, ISM fell, earnings declined, profit margins came down, the valuation crack started after that. Handicapping it. This is a low base rate environment for a sustained bare market. You'd have to have a lot of stuff go on. If those things are going to happen like they did in 2007, the market will still make new all-time highs, and you'll be able to change the risk going forward. The probability of an '87 style move at this point. You can make the argument there. That is not where we are. Bearish sentiment is everywhere. We do have retail heavily involved, but they've been making money for a long time now. And if they're using call options, guess what? Call options have a defined loss. Stocks, owning stocks, your defined loss is a lot bigger than putting money into a call option. So, if they're using call options and having replaced stock, but they also changed their minds and they have no trouble getting bearish.

So, if I was going to pick where the S&P is likely to go, I showed it before. Here's the 50-day moving average. Here's the prior consolidation on what I would argue was a wave four. And then we've got the 62% retracement. So, that's where I think it could go. If it went below that, I think it would be quick. And just so you get a sense, on Friday, when it felt like the world was ending and the S&P was down close to 3%, five sectors were up. This was a rotation, guys. And not surprisingly, the ones that were down were the ones directly related to AI.

Okay, Jeff Degraff, very, very good at this particular stuff, looking for areas to invest in. He's been talking a lot about healthcare lately. Adam Parker also been talking about healthcare from another perspective. The capital that exited the chip complex didn't leave the market. It rotated into healthcare and financials. The day's two strongest sectors. I like me some healthcare. Mark Newton, another one. I see good likelihood of rotation, rotation, rotation, as healthcare, fins come back to life. Okay, these are two guys, don't work together, compete with each other. And that rotation out of here, some into there, in my opinion, and some into here, some into there. It's this that I'm worried about. You have to be more selective on the chip front.

The application side, that is what the Eli Lilly paper is about. I think you should read it, if for no other reason, just to understand what the application side is and why we are entering one of the most important times in our lifetime with regards to pharma and the ability for people to deal with longevity. This is a PT, the PT title of the paper as of now, that's going to go out Monday or Tuesday. From labor versus capital to compute versus energy. I'm just going to read this part. Yuang's other formation is even simpler. The input is electrons. The output is tokens. That is the new production function. The factory consumes energy and produces intelligence. The irony is that the great software winners were already showed us that the labor versus capital relationship was breaking. This again is what AI is about.

So, in the old world, it was labor versus capital. You had credit. When profit margins went down, you started to see asset prices come down, you had problems with debt, and it would lead to people being fired. Well, now it's just compute versus energy, and human beings are not a part of the equation. Credit's not a part of the equation. And I know that's funny to see, but the reason I'm bringing this down, Microsoft, Alphabet, Meta, Amazon, and other platform companies scaled revenue, margins, and market value without scaling labor or brick-and-mortar locations. They didn't have any debt. To win the battle, it was all based on revenue per employee. They didn't have to hire as many people. Their profit margins are above the S&P 500, and they are the ones now funding the next leg in an extreme competition to avoid obsolescence, and most of them, in my opinion, will lose. The business cycle has changed.

So, this is the old one. Credit expands. You borrow and hire to build capacity. You hire people. Demand slows. I mean, everyone has done this who's managed people. Morgan Stanley opened an office in Brazil. You hire more people. They gave me more headcount. I refused it because I didn't believe in the future of Brazil at the time, coming out of the emerging market crisis. Demand slows, margins compress, layoffs transmit. That's not going to happen ever again because the hires don't consume. They don't have children. They eat tokens. We are building compute and memory. The cycle of AI is about bottlenecks and shortages. It's about the power grid, about chips, about networking. It's the physical world. The AI cycle breaks when digital demand outruns physical supply. Now, instead of breaks when you're going through it, this is when you have an AI, quote unquote, recession, or when the AI trades aren't working. I think we're in that mode for some of the things now. Tokens are the food source of these digital workers. A human employee consumes wages, benefits, office space, management, time, and trading. Then they have families, they buy cars, have children, and leverage up through a home. A digital employee consumes compute, memory, electricity, and data. That changes the bottleneck, and it also changes the business cycle. I'm building a business. I have a company. I've managed hundreds of people over the years. I'm not hiring anyone. There is one person who works with me to help grow the business, aside from all of my digital employees. This is a completely new world with high-margin businesses. Established companies are going to have a harder time adopting, which is why I am negative on all Fortune 500 companies by 2030. But for the time being, we are here to make money, guys. And the way to do that is to follow this.

So, what happened? And what did SemiAnalysis, I've directed you guys to their fantastic agentic traffic has surpassed human traffic across the worldwide internet for HTML web pages, guys. It's already started. Transactions, same thing are going on this way. Whatever. Raul Pal and I have been involved in macro for a long time. Raul still does GMI. He's obviously jumped into crypto, and for some people, they've stopped listening to him. Go listen to this 40-minute podcast and see if you understand what he talks about. He breaks down how AI agents will create a vast invisible economy operating at machine speed, and that crypto rails, especially major layer level ones, are the infrastructure needed for agents to transact, settle, and coordinate value. This is the reason why I am buying Bitcoin. This is the reason why I have been, this is the reason why I believe in building out for you guys a crypto YouTube to basically create the ecosystem.

Now, Bitcoin, as I said last week, is in a bare market, and until it crosses the 200-day moving average, there's no need to be involved in it. I'm buying it because I believe we're going to get there, and I'm buying small bits and pieces on the way down, but especially because we're here and because of this. And again, we've got divergences here on the downside for now. So, we'll see what happens. If I'm right about the equity market being fine, if I'm right about three-month bills remaining under inflation, I think we've now done a correction. And I'm not going to spend a lot of time on this, but I want you to think about here was the ETF in for Bitcoin. Here was Trump taking over. This whole period of time for me, now that we look back and we're going through it, was kind of a sell the news event. If we get back above the 200-day, I think it's an important thing. But we're at the 200 weeks. Let's go back to Charlie Munger. If all you ever did was buy high-quality stocks at the 200-day moving average, you would beat the S&P 500 by a large margin over time. The problem is very few human beings have the kind of discipline.

So, here's what I'm going to say. Bitcoin is obviously not a high-quality stock. Charlie Munger didn't believe in Bitcoin. I do. I believe it is the only high-quality stock that will exist for certainty in a decade. Regardless of that reason, the 200-day moving average is where we are in Bitcoin. So, this is the part that matters. Few human beings have that kind of discipline. So, just when everyone is getting bearish, just when everyone is throwing in the towel, this week I had Patricia, I had Jessica, I had Rick, I had Tad, I had a bunch of people inside the community along with Jordy, all nibbling down here and buying a little bit. It's a very different world than talking to people on Wall Street like during the bottom of Liberation Day with Bitcoin. The people that have been involved, they've been paid off handsomely for buying during times like this. I just thought it was ironic that all on the same day, whether it was yoga, whether it was a friend, whether it was a trainer, it didn't really matter. Everyone was nibbling. My thematic portfolio was down 4%. And again, one, two, three, four, five, at least greater than 3.89% with all those weeks in a row. Again, this just creates opportunity.

Here is the Morgan Stanley one that was created equal weight as of last week or two weeks ago. And this is it versus the hyperscalers. This is my favorite trade. This was down, but it was up for the, uh, it's up for the month. But more importantly, the hyperscalers were down big on, on Friday. And there's a reason that's important. For those of you with the subscriber list, put all of the sheets. I ran this and just said, "Hey, go through and find your top 10 names that you think I should buy based on the PEG ratio and based on the pullback that's happening from a technical basis." And again, it combined the technical side. It went through it. It combined everything and it said, here are the three. There's two buy on weakness names, meaning these names have been weak. Their technical scores are fairly low. And then you've got one where their technical scores are high, where because of the PEG ratio and because of their earnings growth rate, because of the, I would stick long and you can look to buy them. That's the way I'd be using this stuff.

Now let's go to my DRAM situation. So, it really bothers me when people are referencing this Goldman Sachs on memory DRAM to remain in under supply until at least '28, 2028, blah, blah, blah, blah, blah. Okay. Now, again, here are the facts behind Goldman Sachs and Micron. Hold, hold, hold. I don't give any credibility to anyone talking about memory because, as I repeat, I was abused most of last year, throughout the year, even into September. The only thing that time this changed was October, and that's because DRAM prices were not only up, but they were still continuing every week. That's when the most sophisticated tech people started to at least nibble, but it had already gotten away from people. And then by the time people got on board, my issue is this chart. All of the stuff that Goldman talked about is based on this chart. And I'm telling you guys, although I agree with this, you cannot translate token consumption into memory. There are many things that can go wrong along the way for memory, many in terms of not needing as much, and that's just a reality. So, when you start extrapolating something all the way out here and say, "I've done the numbers to go through it," you're really going the wrong direction. You're really trying to do something that most people, uh, just really, uh, make a mistake with on the bearish side too.

So, this is another one that comes out in terms of the capex right here at this point, again, before it got to this, because this includes the final quarter. You have to remember that people didn't believe this was going to happen. So, now they're believing it up here, and now they're believing this is a guarantee. The reasons that people were negative on this, some of it still is true. The bottlenecks are here. If we don't do seven, if we do 785 this year, but we don't do this number until this year, or we do only two-thirds of this because we don't have the data centers to build it, the problem is going to be is we've already built in this whole future. We've already built it in, meaning we're, we're ahead of the game. We have no cyclical shelf. And for those people who say, "No, Micron is still cheap. SK Hynix is still cheap." I get it. But the reason they're still cyclical is because there can be solutions. They can be algorithmic. They can be efficiency gains. The problem is, if this bottleneck delays the ability to build it out, there will be solutions. The fact that memory is so expensive will lead to solutions. That's to a degree what Sarah was. And if you don't think there's risk of this, Meta is building dozens of massive tents at campuses across the US, sticking billions of dollars of chips inside and powering them off off-grid turbines. The IRS has officially entered the Mad Max phase. I've heard this so many times now. I've heard it from people this week, not in X, but people saying the hyperscalers are in shock at what they have to do now to get the power. JP Morgan on 27 data center buildout. The latest analysis based on satellite images shows that over 60% of data capacity planned for completion in 2027 has not begun.

This is the key thing. Anthropic urges global pause in AI development, flag self-improvement risk. Now, you can read this and you could go, "Oh my gosh, a pause in AI development. That's what caused the market to sell down." That's not the important thing in there. Two bearish implications for memory makers from the Anthropic self-improvement article. AI developers could slow because of safety and government concerns. So, that's one side. You've also got Trump posting, maybe Bernie Sanders is right. David Sax is posting it. Maybe we should give people part of the AI companies. I don't know what that would mean to them being as aggressive and building out if the government is taking a stake. I don't know. It's one of the risks that I've talked about in here with the hyperscalers is should they be trading at the multiples they are when it is obvious that the government, either this administration, the next one, or someone, will be involved with memory names. It is impossible for me to not believe that somewhere in South Korea there won't be a politician that realizes we should be taxing these. There are a lot of negatives which should lead to multiple compression, but this is the one that is the biggest negative for memory. Recursive self-improvement could allow AI to solve its own memory bottleneck. So, I'm going to go back. I'm going to just make sure you guys read this flag self-improvement risk. The easiest way to solve the memory problem is for RSI. The bearish memory risk from RSI is two-sided. Safety concerns could slow the buildout, while recursive self-improvement itself could accelerate the discovery of memory-efficient architectures that reduce the amount of HBM needed for pure intelligence. When this starts happening, you won't be able to get out fast enough. So, that is my, uh, logic. That is my, not even guess, that is the reality that you have to build into the risk reward, which is why for me, and Micron, that risk is far greater than investing in Marvell.

The optical side is in the very early stage, and that is why Marvell, which Jensen Huang said will be a trillion-dollar company. It's $250 billion. Give me that one all day long, guys, over Micron at this point. That is the reason why more firms turn to DeepSeek. This is the other issue for memory. More US firms turn to China's DeepSeek over pricey Silicon Valley. In fact, I didn't read the bottom part there for you. Chinese artificial intelligence took the top spot on a major US business spending index in June, as more companies swap out expensive American options like Open, Anthropic, in favor of more. If we get any sign on those parabolic charts for Anthropic of a slight peak, a slight peak, if all of a sudden in July, you look and it's the same ARR, what's going to happen to memory? What's going to happen to the entire AI thought process? I'm telling you right now, do not take these and extrapolate them. There are a lot of issues involved. Do I think S&P earnings are going to be good? Yes. Do I think they're going to continue at around where they are? Yeah, we just blew out numbers. We're probably likely to beat numbers. But do I think that that is a place to have your money? No. I want to move to the application layer.

Now, the application layer is the benefits that are coming from recursive self-improvement. For those of you who don't follow my Substack and my, um, subscriber at this point, I'm trying to help you navigate through exactly this stuff. For all of you on the institutional side that have now jumped in, trust me, you've jumped in at a much more dangerous time than it was in October. And that is the reason why I'm trying to talk like this. Edge open-source substitution risk. Users migrate towards cheaper DeepSeek like open-source specialized or on-device models, reducing dependence on centralized frontier models. Everyone that I talk to that has a brain is figuring out ways to do this stuff. I already have a Chinese model on mine. The only reason I don't have DeepSeek yet is because the hardware isn't there for me to have it. But the hardware will be there. That's why I want to be on the hardware side. But it's not the memory side, guys. It is just the hardware.

First podcast to listen to. Norris Bank, Nikolai interviewed the head of IBM. He says that some of the AI infrastructure buildout is a bit ahead of what the world can tolerate the next few years. He estimates that one gigawatt of AI data center power requires roughly $60 to $80 billion of semiconductors to populate it, just of the semiconductors, okay? Remember, it was only $50 billion for the buildout out a couple of months ago. He does not think the revenue pool is large enough yet to support all the capex. He's doing all this based on a very simple equation of this is how much it costs. This is how many gigawatts we need. This is how much infrastructure would need to be built, and do we have enough revenue? Now, a lot of these companies have to build the revenue out of for no other reason for the cloud side to be able to get the clients, assuming they can get them. Some will disappoint, many will thrive. In other words, the first phase was about securing chips, power, land, and capital. The next phase will be about proving utilization, pricing power, customer ROIC, and payback. Eli Lilly, other places in there that are benefiting from the ROIC. It's going to be a very different world coming out. Frontier models may become more commodity-like than people expect. If customers can move between models, which I do every single month, it seems like I am off of Claude as my dominant one, and I've moved to GPT 5.5. And everyone who argues with me, I think at this point, you have just become too connected to Claude, and many people were too connected to ChatGPT. I like to shift based on what's going on, and I use all of them every single day. The capital cycle of the adoption cycle, he says, is still early, no longer the first inning, but maybe the second inning. The bears are wrong if they think AI is just hype. But the bulls may also be early if they assume enterprise adoption will happen instantly. I could not agree more. I think everyone is starting to get disappointed with the cost. There's no doubt that this is becoming more of an issue. AI might compress the cycle again, but he still thinks it will take years, not months. That is the key tension. The capex cycle is behaving as if demand is immediate, while the enterprise adoption cycle still has to pass through the trust, integration, measurement, and it's trying to do it with costs rising rapidly. The biggest investment implication is that the market may move from a capacity land grab phase to an ROIC discrimination phase. So, that's where I'm going to leave that at this point.

Jensen Huang, if you didn't believe, uh, Krishna's, $60 to $80 for these chips, Jensen says the cost per gigawatt is going higher, $80 to $100 billion per gigawatt into AI factory from $50. Each one of these was a $20 to $30, then it was $50 to $60, and soon it will be this. All of this while the adoption is way behind the cost, way behind. Sam Altman says AI budgeting has recently become a huge issue for some companies. Never came up last, last year. Company's AI bills are bigger than ever and coming due.

So, I wanted to bring this up not because I agree with Michael Bur. I want to make sure that all of you that are now bullish because you've had the Goldman Sachs research on what DRAM is going to look like going forward and what the capex numbers are going to be, in the same way that he's using math in exponential time to argue for what's going on. You cannot go the other direction as well. Extrapolating five years in when AGI is now being said by Deisabis, who as of a year ago was saying it's not going to happen before 2030. I'm leaning towards the early 2030s. Now he's publicly saying 2029. You've got Anthropic saying recursive self-improvement. This is what I talk about in Eli Lilly. This is what you want to be long now. You don't want to be playing the math game on capex, either on the bearish side or the bullish side, guys.

Alphabet's record-breaking $85 billion raise for Google AI business is a hell of a good signal. Hell of a good signal. The good signal is there's enough cash to do this. So, anyone worried about what's happening, they're buying this for the cap, they're doing this for the capex. The fact that they're front-running the IPOs that are coming out, which are also in the game of getting the money for capex, is unbelievable. The fact that it was $85 billion and barely budged anything, and that is bigger than the, I, I think all the companies in the S&P except about the top 200, 160, the number 200 market cap is less than the raise that they did, and it had zero impact. And then on Friday, Meta's big equity raise after Blockbuster, Google, and if this is going to happen, guess what? You know it's going to happen before the IPO does. Uh, Goldman Sachs analysts are predicting this is what the IPO numbers is going to look like. We've already seen what the debt side is. The wall of supply, Paul Tudor Jones talking about it. I'm not going to go through all the numbers, but let's just say there's a lot of stuff going on.

At the same time that that's happening, this is the reality of what the World Semiconductor Trade, uh, group is saying. The WSTS forecasts that we're going to go from 90% year-on-year growth to 27%. And this is assuming no real bottlenecks. So, the question is, if that's the case, what's going to happen to these names? If the second derivative goes, how much of the market understands this? Now, again, if you're a mutual fund and you're buying these with the expectation that three years from now, you're going to get money. They're in my thematic portfolios, and they're going to remain there. Uh, I think in the near term, they're going to need to be priced properly for the risk that I'm measuring out. I don't know how long that's going to take. It's either going to be through time or price. If Micron goes to $500, I will be involved in starting to buy a little bit back. If it goes to $400, I'm absolutely going to be buying it. I don't think it's going to do that. I think more likely we're going to look back four months from now, we're still going to be around the same price. I was wrong getting out at $700-ish. It went up to a thousand. That's a, you know, a big move, but Marvell went up the same amount and is now outperforming with the move that we had there. The whole point was I didn't want to be in those things.

Now, at the same time, your boy did buy silver and bought Bitcoin. That's gone down. So, not all of the things that I did worked out. Um, I'm, I'm going to give these guys another shout out. Compounding Friends. This is without, was without Josh Brown, but I think this is a good one to listen to on the topic of bubbles. Um, Michael Batnick. I, my appreciation for these guys just grows every week, just because they get on there and they are not, they're, they're trying to do what I'm doing, which is provide facts to push back on extreme things. Not real things. It's extreme things like saying, "This is a bare market. The market's going to collapse. This is a bubble." That's an extreme statement. Agreed. Retail's over their skis a little bit here. There's too much gambling going on. What they talked about here is Michael argues that experience can actually hurt investors during a major regime change, because people often become experts in an earlier version of the world. I completely agree with that statement. The deeper point Michael is making is that experience is not only w, is not always wisdom. In markets, experience can form, can become a form of anchoring. Investors live through a traumatic period, learn a set of lessons, and then carry on those lessons forward as if they were universal laws. But markets are not static. I've talked about this. They're behavioral. They are biological. They adapt. The world changes, the dominant companies change, the source of earnings changes, and the cost of capital changes, and the economy reorganizes around new technologies. The investor who has seen this movie before may actually be watching the wrong movie. Pattern recognition can become pattern imprisonment. I think that's a good line.

All right, another podcast, another keynote speech. Go listen to Jensen Huang. Again, there's nothing new in it, but I think you need to listen whenever he speaks. Um, Frames AI is the new industrial system. AI factories that turn electricity into tokens. Vera Rubin has announced, uh, is launched. Uh, Nvidia and Microsoft are reinventing the PC era. So, again, you're going to start getting more hardware out there, which should make it easier to run these bigger models, which means more on the edge side. So, again, we start moving to the edge. Go back to what Gavin Baker said as the really big negative for the capex trade is if all of a sudden everything moves to the edge. Well, if the Chinese models keep staying up with the US models, and trust me, they're right there, everything's going to, everything's going to shift.

All right, Vera Rubin, you guys should make sure you're spending time on it. That's why I'm going to do the 800 volt DC. Uh, that'll be a thematic, uh, index as well. I'm also going to try to do one on the application side. It hopefully will.

be diversified, but it may be specifically towards certain um by sector. In this case, healthcare is the one I'm focused on. Uh Dell connected to Ver Rubin and the first shipment was confirmed. Dell AI factories in the industrialization of intelligence. If you guys didn't read this, I would highly recommend reading it. I think Dell's earnings were the and and the company as a whole are one of the most important companies to pay attention to because of this part.

So the chip part, the building of the data center part is not as interesting to me as the agentic side, which is about turning tokens into intelligence. And the intelligence is done through the agentic world. I think you have to again spend time on that. AI agents are moving closer to the edge. Dell gave a presentation on this. If you didn't believe in Dell last week, well then we got ULIP Packard Enterprise Company this week blowing out numbers and here's what happened to their stock. So if you thought Dell was a bubble, then you clearly thought that HPE was uh as well. And the what I wanted to show with this chart is okay. So here's the last decade now of Hulip Packard. This is not a bubble for one reason. Here's the total annualized equity return of 15%. It is still underperform the S&P even with this bubble-like move. The hardware side eventually this will be massively outperforming. I'm sure you'll see this at I would expect this to be at some point over 30%. eventually.

Um, if you want to go through what they said, elevated prices to persist well in 27 for DRAM, uh, memory bottleneck is a multi-year issue. Uh, and these are the reasons why that people are involved. So, this whole setup for the DRAM thing is where this is going to be. But he also talks about they're explicitly saying they can push multiple rounds of price hikes into a tightening memory environment. So they're passing along the memory side and I think that's what makes them very very interesting to look at from what's happening. Um meeting European customers uh did not want the product because not one of them said they did not want the product because it was too expensive. So there's HP just again to look at it in the bubble side after not moving.

Uh Marll Matt Murphy was on stage with Jensen Yuang. They talked about the partnership. They talked about all these things. Customers do not have to buy everything from Nvidia, but Nvidia still wants to be inside the data centers when custom when uh customers build custom asex. Um this whole thing of listening to it is very important because it goes through as the copper wall moves inside the rack. Murphy argues this will create an explosion in optical demand. Now the key thing is here Murphy argues this will create an explosion in optical demand. The next wave is co-ackage optics or CPO. This is why I keep saying this is the early stage of something that does not have as much competition. This is not just the fiber side. This is the actual movement side, the chip side. I wrote a paper again on March 30th on this. Again, the price was around $90 at the time. a powerful thematic trend that remains in the early index. This is two months ago. Here's the chart. This is before Jensen Yuang spoke. Then he said it was going to be a trillion dollar company. We finished the week at 260ish. Uh so up again. It was up almost 30% for the week. And then after market hours, it was trading up another 7%. Marll will become the next trillion dollar company. It's about 250 billion. Marll stock was just added to the S&P 500 energy.

I was buying Chevron and Exxon this week. Uh partly because I think the stocks are completely mispriced relative to what I'm about to show you in terms of where we are with Kouchy and everything on the prediction markets. Um I think these stocks are going to be necessary. I don't envision these being doubles and triples in a year, but I think energy at times you want to have it in there, especially now because if we don't get the straight solved in some way and it's still closed, these are going to be a hedge in the portfolio. Even if we get something that looks like some ships are going through, the reality is we're not going to get back to normal for a while. My guess is after looking at surveys on oil and how scared people are to be long, we're probably going to see oil prices either migrate higher or stay around the same level. And I think energy is going to become more attractive because the chip names are going to be more volatile. These are defensive safe names. I bought Exxon. I bought Chevron.

Fluence had a big week up 44%. For those of you I've done calls with lately, I've talked extensively about batteries and the fact that this is going to be a place that two years from now you're going to wish you spent a lot of time on. Um because of all the problems we have, because Meta is using tents, batteries are a necessity. They are a necessity. I wrote a piece on this uh three weeks ago and the reason that this was such a big issue is because of the deal. Fluence energy will integrate its battery energy storage products into the data center designs for flexible power. It did a deal with Seammens and Nvidia. So they already mentioned their earnings which I highlighted that they have two hyperscalers uh involved. We signed master supply agreements with two major hyperscalers. The first one expected during the third quarter and then the seaman's relationship adds another layer validation. This is before the announcement. So this was me picking out what was already going on. I've had a high hit ratio with Nvidia announcing things like with Marll the day after I wrote the paper. I write the paper on Fluence and Nvidia does a deal with them in two and a half weeks. This is just me going through the earnings and paying attention to everything going on.

I just wanted to show this because we've got CNI loans going up. We've got demand for loans going higher. We have the ISM PMIs going higher. I just want to make sure you guys realize that we are in the very early stage of this whole thing. I hear it from all kinds of places relentlessly. The nominal GDP side is there. And for anyone who writes negative things about what's happening in the economy, they are using the old cycle of capital and labor. Remember what I said, AI agents do not buy houses. They do not buy autos. Autos and housing and all the things that go in those things are very important to the way the LIE works. All of the people that were trained on housing matters. It doesn't matter anymore. This isn't a bare market in housing. This is called a stuck market. And I think it'll be stuck forever. Real estate stuck, private credit stuck, stuck, private equity stuck, VC stuck. I think all of them are stuck. Which is why tokenization will unleash twothirds of dormant assets and lead to another phase of what Raul Pel talked about. If you want to learn about the software side of the future, read my extensive multi multi-page paper on Eli Liy and start listening to everything I'm talking about in crypto. If you want to stick with the hardware trade forever, good luck because we are moving into the application side right now. Rio is the story.

So, here's the paper I put out on Friday. I have spent about 10 months working on this. It is very hard, as all of you know, to get up to speed on what's happening in healthcare. But GLP1s, I wrote papers about them at the hedge fund. So, I'm going to just give you the extreme hot takes that are in this paper. First, I believe by the end of this decade, Eli Liy could be the largest company in the US. Larger than Nvidia, larger than all of them. Eli Liy is becoming the most important AI company in the world. Peptides are this decade's API keys. So, the Mag 7 needed API keys to get there where we are. This is human software, guys. Eli Lily made new all-time highs on Friday. Lilyod is an early sign that the enterprise AI cycle is moving from generic models trained on the internet to specialized models. Read that line as many times as you need to to understand the difference between this and a general LLM. Eli Liy has a lily pod with a thousand plus Blackwell GPUs trained on their data, not on the internet. And that may become the defining mega trends of the return on investment phase of AI. This is the AI factories. This is it. and Eli Lily, as I write in the paper, go read all of the things that they've done, all the partnerships they've done, all the money they're spending right now, and where that money is coming from. They have the cash. Their revenues are growing 55% year-over-year.

Hardware is now more of a risk. S&P 500 is up nine days in a row, nine weeks in a row. The reason I want to show this, this is not some ab some you uh some thing that you should fade. Historically, when you go through what happens in returns when you get this kind of a strong market, nine weeks and nine days on the back of earnings means this is something explosive. I don't know how explosive. I just know not to fade it.

Crude, I'm going to finish up here. We cannot ignore this. It just continues. And unlike the tariff situation, I do not believe that the White House meant to be in this situation. And I do not think it is their choice to get the straight of Hormoose open. And everyone I think now gets that. So if Iran wants to keep playing a game and keep this closed, I don't know why that isn't a risk for the market that's greater than the way people think about it. So again, the reason I'm buying energy stocks along with Eli Liy, I like that Lily was up. I like that Exxon and Chevron were barely down on Friday. I think there's a risk here that as the trend and the charts keep going this direction. If you don't have energy stocks in there, you are playing a big risk with the other side of the equation because I think once you get a downtrend in a lot of the AI names, as I said before, if oil is going higher and rates are going higher, we have a new Fed chair coming in, he's going to be sitting in a difficult position. I don't know what's going to happen. Japanese crude oil reserves. So this is not just a US problem. Here are the probabilities on the straight of hormuz. When will it return to normal? Meaning before it was and again before October it's now down below 50%. So again normal is a different thing. We've taken off some of the supply.

So I'm not bearish on the market. I don't think the straight of hormones matters nearly as much. And for the smarter people out there that I've heard speak, the pipelines and all the things that are being built will be ready in 3 to 5 years. So the straight of Hormu's importance will go down. So for those of you who are extrapolating out, I think that means that Iran does have a reason to negotiate because if you can go around them, then they're really losing everything in including their oil. So I don't want to get bearish on this. But I do believe from an inflation perspective and the fact that it probably doesn't go back to normal quickly, you have to build that into what you're doing in your portfolio. And I think energy stocks and the fact that we're going to have a bid underneath is going to be there. If you want a good um balanced look at the oil market where they talk both about the negatives and the reality of what's happened and the positives, I think it should be here.

I'm going to end on this note. In the same way I talked about recursive self-improvement, I am not worried at all about energy 5 years from now. I'm not worried about nuclear. I'm not worried about any of those things because recursive self-improvement will have the ability of solving problems at a level which will allow us to not need as much energy at the same time and make it easier on the supply side. That's where we are. That's my story for this week, guys. Uh thanks to everyone who helped this week. Thanks to all the subscribers. Go to ai.22vresearch.com. I also put in the description every week where you can go to it. Let's go make some money.