Momentum Is Crashing, Bitcoin Is Bottoming, AI Agents Are Rising

Back in Maine. Uh, 90-plus degrees for one of the two times a year. Uh, holiday week. Happy 4th to everyone. Hope you got to enjoy it. I did this on Friday, so, uh, I'm seeing you guys Sunday.

Uh, midcycle slowdown continues within AI, and as expected, the narratives and the headwinds and what everyone's talking about. Uh, I spent a lot of my time this week, uh, talking to people about particular, uh, podcasts. Expats. Again, we are in a, uh, a point now where people are getting worried because it's not as easy to make money. So, I'll go through that. A big volatility shakeout. Momentum is just, um, killing people with the volatility, which I've highlighted. I'll go through podcasts. Greg Brockman had a great podcast this week. The consolidation playbook, uh, for subscribers. Kevin Worsh had an interview. I think, um, people need to start focusing on that. Some of what I'll be going through is going to be focused on Worsh.

Um, the application layer, insurance, and healthcare seem to be the two sectors right now benefiting the most as the consolidation is occurring within the infrastructure. Meta, another big, uh, week for them in terms of trying to get the market all enthused. Uh, open source continues to go. And do I say Bitcoin is actually giving me enough to talk about from a potential bottom? We're still below the 200-day moving average, but I'm spending a lot of time on this because I believe the application layer matters. I believe Worsh matters for Bitcoin. We'll go through it.

Uh, I just want to start with this equal-weight S&P. If you're listening to anyone saying that the market is a bubble or that a correction is coming, equal-weight S&P is breaking out and is strong despite all the volatility. There is a rotation going on. IWM made new all-time highs this week. The S&P is in a consolidation triangle at this point. So, for everyone who sends me stuff getting all beared up, got to short tech, got to short the AI trade. We're going to see capex cut. Everything is going on.

Um, the S&P spent the month of June consolidating. Uh, it's basically the same level it was since the beginning of May, which means we're taking some air out of the bubble. While it's an earnings-driven recovery and bull market cues, same triangle formation. The New York Stock Exchange cumulative breadth all-time high this week on Friday, Thursday.

Hyperscalers continue to stay in a range they've been in. They are unchanged, almost effectively, since last year. Again, this is Meta, Amazon, Microsoft, and, and Google. Uh, it's multiple compression. Their earnings are growing. Everyone keeps trying to pick the bottom of them. Everyone keeps trying. I, I think these again are just like software has become and the aggregate for SAS. This is just going through multiple compression. Every now and then, two of them will do well, two of them won't do well. I still like them as the funding side for this, but I don't, I, I think we've gone through enough at this point where the V so high. This is my thematic portfolio. We've been consolidating like the market has. Just again shows that the AI infrastructure trade has been doing everything.

Now, for any of the institutional clients that I've met with over the past couple weeks, when asked the question, well, what kind of consolidation? I keep saying the same thing. I think a 50-day moving average, uh, retracement either through the 50-day moving average moving up or prices coming down, or a combination of both, would be good. This is the Cosby, uh, two, uh, 200. So again, this is weaning out all of the people on margin. There's a bunch of ways to lose money when people are gambling. And one of the ways is just by the market stopping to go up, and then gradually they, they lose their attention. It doesn't have to collapse in a bubble.

Uh, this is SKH touching the 50-day and then rallying, uh, on the day off. Uh, this is the financial side. I'm going to bring this up as I go through in the US, but the financial side in the Cosby, uh, has been consolidating. If this is a GDP-driven market, if GDP around the globe is improving, as we're seeing with estimate revisions, you should start seeing the financials be part of the bull market. They were lagging this entire year, especially in the US, on the back of the Blue Owl situation, and this was impacted. Well, now we've consolidated. This is another consolidation chart, which looks like it's going to break higher. So, we'll see if financials become a space people want.

The megaphones. Um, we got one in the SOX rotation. So, this just again shows the massive increase in volatility. So, this is the same sort of the same thing as momentum, but it's just a rotation between the semi-software or the semiconductor ETF and the software ETF, and just how wide the ranges are going. This is a way to get hedge funds out of the trade. It's a way to get, uh, retail momentum traders out of the way. They don't want to be looking in something that keeps costing them money when they keep buying a breakout. For the hedge funds, they also get stopped out, or if VIX goes higher, the risk managers tapping them on. You can't be long this much VAR. This is the way to get rid of things. And these megaphones, this is the Morgan Stanley Momentum Index. And you can see this is all the way back to 2010. Just how wide it is, and just how many times we're trading outside of this 5% band. These things are just moving, uh, every day.

This is TMT. So, this is the tech side, and this goes back to 2010. You can just see how wide these have gotten. I mean, this is just dramatic. And I would say beginning from when the Opus, uh, Opus 4.6 kicked in, or 4.5, which is back here, we've seen more and more people going. This is a sign that it's just crowded. There's just too many people in the trade. And this is the volatility of that, uh, M Tech Morgan Stanley Mek. I mean, this is 5% a day. This is higher than the dot-com bubble. Um, this has just gotten to the point where it's flushing people out.

Now, this month has, over the last five years, has just been horrible for momentum. So, we're off to a horrible start. Uh, so far in 2026, momentum's gotten crushed. But you can see that this has been a bad month, just like December has been a bad month for momentum. So, the seasonals are in there.

Um, I happen to subscribe to this as well, not in terms of market meltdowns, but I think the more and more that individual investors, particularly out of Asia, are trading the markets using AI agents, AI agents are going to continue to push this stuff. Those charts are bullish. They're not bearish. The only people getting hurt are the people that have drawdown limits, like hedge fund managers that have month-to-month risk for liquidity purposes, which goes back to hedge funds. But retail traders and people that are there, they're trying to make money, and the VAR is not part of their job in this. So, they might get chopped up a little bit, but I think the people that are more likely to get chopped up are the ones that are worried about the mark-to-market performance, where they're borrowing money from people. So, I think the better way to put this is we're going to see more vol-vol, and I've been saying that for a couple years. But I think as the agents get involved, their number one goal is to make money, not to meet a certain Sharpe ratio.

Uh, I just want to bring this one up because I've, I've repeatedly seen people show this chart, which is the margin debt, and say that this is a bubble. And I'm just sick of watching it. It's just such lame analysis to sit here and say that debt is getting bigger without looking at the size here. Just go look at the debt of the housing market. If the, I mean, we get into the point that if the value of the houses are going up, well, then the debt's going to go up. This is it relative to the market cap of the US, which is a better way to look at it. Now, it is rising right now, but again, we're below the midpoint of the range over the since the Great Financial Crisis. So, I don't see debt as being a big story.

So, let's get back to what I showed last week, the different types of consolidations. And like I said, this would be great if the AI infrastructure trade went up, you know, 20%, and then you consolidated in a tight range. The consolidation is going to be somewhat representative of two things, in my opinion. One is how much of it was fueled, uh, or how much, what were the earnings growth during the time period, and is this something that, uh, in price movements was big? This is basically something where I would expect a very good Sharpe ratio with something that's a little more boring. This is when it's more speculation-driven. I don't believe the AI trade is speculation-driven. I believe much, most of it is earnings. There is some speculation, and that leads to this action, and that's where I think we are. We are resetting now the thought process on this. And these are all real, and I've talked about all of them every week. It's the reason why I go through the headwinds, not because they're going to stop the AI trade. Headwinds don't stop things. Planes don't hit a headwind and then go straight down. You get turbulence, and that's what we're going through at this point. So whether it's the data centers, the power and cooling side, which is causing some of the data center side, the memory bottlenecks, the regulation, all of these things in here, of which there's many more, too. They're all part of the headwinds that are now becoming the bigger story. The reason they matter more now is because we've gone through the earnings. This is no longer as big a surprise, and people are crowded into the trade. There's no way to refute the fact that they're crowded into it.

Um, I thought this week's "All-In with Gavin Baker" covered a lot of the headwinds and some of the topics, uh, fairly well. Um, they also went through the memory side. Uh, they talked a lot about GLM 5.2. I do think you should listen to this one on it. Um, Gavin Baker says the future is composable models. This fits in very importantly with GLM 5.2. You need to have your mindset on this. It is silly to think that it's all or none, meaning, okay, I'm going to use open source, I'm not going to use these. But that's never going to happen. The best models still have been in the frontier side. They've still been in the US. Think of it as those are the IV Leagues or the highest IQs out there. Um, the companies that have the most money, which spend the most money, which drive the economy. And I'll say it this way, in the Russell 2000, the total market cap is about three and a quarter to three and a half trillion dollars. Well, that's one company at the top of the S&P. Pick your pick. So, one company equals 2,000. So, if 2,000 people use open source that are on the smaller side, but one company uses. Okay. Well, if all the big enterprises do, the S&P is a massive side. So, I do think over time they'll be there. But don't fall into the trap of people saying that no one's going to use the higher models and Anthropic's revenues are going to stop growing. I think that is just again, silly talk. Um, and I'll get into the composable models later. And then they go through the AI regulation. And then the final section, they cover the modular compute and a whole bunch of other things, which again get into where we're going on this.

Uh, on the GLM 5.2, two, they put it as a serious open-weight challenge to US frontiers. And for any small businesses, anyone who's sophisticated enough to be able to use open source, it's going to happen. I've been using an open-source Chinese model on one of my setups since, uh, April. So again, this is going to happen. I don't have enough hardware right now to deal with GLM 5.2, two, in terms of on my own computer. But go, I thought this was the most important line. Gavin Baker says the model's performance on GLM 5.2, the open source, forced him to rethink part of his framework. I do think how good GLM 5.2 has challenged some of my beliefs. He talks about it. Those beliefs have to do, number one, with could open source keep up? I talked about that with Leopold Ash, uh, Leopold Ashbrunner, um, who also got that wrong in his, um, his situational awareness. Uh, but this is the other part. The future is composable models. And I think we're all starting to realize that to keep the cost down, because no one expected the cost of these things to go up so dramatically, that you're going to have to make sure that the smarter models are making the decisions. The smarter models are being used on the most important decisions in a company, and certainly the most important things like drug discovery, science, energy, all of that stuff. So, you're still going to need the models. You're still going to have people paying a lot of money for them. We still have scarcity on them. Uh, but I think the open source is going to take up most of the, let's say, workflow jobs. So, think anything that you're using Copilot for. Copilot is not this. So, think of open source as being behind Copilot. And don't remember, don't forget I highlighted that Microsoft was talking about DeepSe possibly for for that.

So, the frontier, I left that up. Frontier tokens are capturing 90% of the economic value, and open-source tokens are probably 80% plus of tokens processed. That's the thing that matters. Everyone is starting to look at this and going, this is going to kill Anthropic tokens are capturing 90% of the economic value. So, I don't agree with the people that are getting bearish on this side and trying to make some kind of cracks in the AI, which I heard repeatedly in podcasts over the last three days.

Um, I, I didn't highlight this last week. It came out the, the week last week when I did the video, but I think Dean Ball, who was hired, a former Trump official on the AI side, was hired by OpenAI. I think the paper that he wrote is kind of important, what should be done. I, I'm not going to go through a lot of the details on it, but he's just saying that the government regulation, uh, it creates a dangerous limbo where these models are not being used, and it's allowing the open-source models to catch up. It's creating other issues. It's allowing countries around the world to think that maybe we shouldn't be using this. So, he's just going through this as a risk. Again, a headwind that I think is important. The administration and the states have not figured out how to deal with this. We still have a midterm election coming up. Again, I think these are headwinds. I don't think these are stoppages.

Same thing with the data center, the backlash, front page, an economist article last week. JP Morgan put out that more likely to have a cyber attack causing a banking crisis than a credit loss. I've talked about this before. Agent swarms are bigger risk to the fiat system than quantum is to Bitcoin. Book it. Remember it. Stay with it.

Um, this is a memory risk that came out, and if you didn't see this, it definitely got pushed around. It's starting to feel a little like the crypto market, always in the equity market, where there's this pump and dump thing where people are just posting stuff. Now, Andrew Curran is a very respectable, um, person within the tech and AI community, and I highlighted a paper he wrote last week or the week before, which was excellent, but he put this out, which is probably true. There has been a significant breakthrough in architecture, specifically around memory efficiency, not by one of the big labs, but a team that was spun out. 2.1 million views. So, it was floating around. I think. Let's assume it's true. Um, don't go sell your memory stocks because of this. But as I've said since the day that I got out of my Micron and said, "Hey, I just don't think these things can continue." If you're lucky enough ever in life to make a 7 to 10 bagger, don't expect there to be that much more going forward. Maybe there's a double over the next three years. Maybe there's, uh, 200% over the next three years. It isn't going to be like what it was. And I just thought there were better places where I still could get triples on the same theme. So, for people that are long memory, I wrote a paper on it this week. I still believe memory is, is a scarce issue. And even if they've come up with something, the question is, until we get details on how fast they can get it up, how fast this can be brought through, whether it works for everything. We don't have any details. So, I would not spend a lot of time on it.

Okay, those were the headwinds. Now, let's get into the podcasts for you guys to listen to on where we are in AI and stop listening to the noise.

Um, "The Future of Software" interview with Scott Woo, who's the co-founder and CEO of Cognition AI, believes most people underestimate AI because they pattern match from history instead of reasoning from first principles. He highlights that AI has gone from handling seconds of work to hours of work and asks why that cannot become days, weeks, months, or eventually a year of uninterrupted work. The most interesting future agents are not just task executors, but mission agents. I, I, I'm bringing that up because that's getting back into something I talked about last week, which I'll show in the next slide. Uh, uh, AI may be one of those rare periods where pattern matching fails because the underlying variable is improving exponentially, which I completely agree with. Um, the key constraint he focuses on is how long an AI can work without human interruption. We are already moving into a world where agents can perform hours of work. And I just want you to remember this, hours of work. That is a huge change. Once an AI can take a well-described task, work through multiple steps, debug errors, search for context, test its output, and come back with a useful result, it starts to look less like a tool and more like a junior coworker.

Combine what Scott Woo said in this part of the interview with what trying to give you guys more. Oh, with the Boris, uh, interview I showed last week. Boris Churnney, uh, I showed an interview on Loops last week. So, the story is the same in both interviews. AI is moving up the abstraction stack from writing code to completing tasks and running loops, and eventually owning missions. Scott Woo describes the philosophical arc. Boris Churnney describes the operating reality inside Claude Code. The industry has already moved from humans writing source code to agents writing code to agents prompting other agents. His concept of loops is the operational version. The unit of work is expanding.

That leads into the Greg Brockman interview. If you haven't listened to Greg Brockman, he was interviewed with Alex Caner-Witz on his podcast. Uh, I think it's called the, yeah, it's "The Big Technology Podcast." Definitely worth the 40-minute listen. Um, it's evolving from a chatbot into something closer to a unified super app. Rapid adoption inside OpenAI. Tools like Codex are seeing explosive internal usage. Um, so the broader ecosystem likely to embrace AI agents rather than resist them. Now, I've said this before, I use Codex more than Claude Code. I use Codex more now than ChatGPT than I had in the past. So, every day Codex is taking over more of my, more and more of my work. So, if you guys have been, uh, stuck with Claude Code and using it, it's great and you're used to it. I highly recommend using both of them. Um, I use, still use all of them, but my, the amount of time, as I've mentioned, I use on Gemini and on, uh, Grok has gone down significantly. Same thing with Perplexity.

Compute is the bottleneck. Demand for AI will exceed supply. Access to compute infrastructure will be decisive. He basically went into, whoever has the most compute wins. So, you can sit there and doubt whether the compute is necessary. You can somehow say the capex is going to be canceled, whatever you want. The reality is, every single person who's at the heart of AI and is the ones seeing the models get better, Greg Brockman and, uh, anyone at Claude, but Dario in particular, it never changes.

Uh, AI and healthcare is already impactful and expected to become mainstream. I highly recommend listening to the final seven minutes of the interview. For those of you who do not agree with what I have said, or have family members with cancer, they go through more stories in there of people that have used LLMs to diagnose what's going on. And there are just unbelievable stories. And when he was asked, is this something that's going to become the standard? And he said, absolutely, it will be the standard.

Here are the key insights. Enterprise adoption, we are at the inflection point. This is critical, guys, to the application layer. This is critical to the profit margins. This is actually critical to the AI infrastructure trade. And in my opinion, this is also critical to Bitcoin. The phase shift is happening now. Now, it didn't happen before. So, phase one to, to early '25, we need AI. We're falling behind. Now, it's, show me the ROI. So, now they're bringing in these companies to do it. The constraint is supply, not demand. Can we supply enough compute and energy? We don't have it. And companies are not adopting AI tools. They are adopting AI workers. This shifts the model to labor substitution, productivity leverage, organizational redesign. It's a redefinition of how we work. Productivity leverage. So, I want you guys to think about when I get into the Kevin Worsh part as well. Just remember what Greg Brockman started. These are the quotes. There's not just not going to be enough compute in the world to satisfy all the demand. Don't know how he can make it any more clear. And he is at the front of the best models in the world. There's not, there's just not going to be enough compute in the world to satisfy all the demand. Every provider sells out all their compute. That's the world we're heading towards. Even at tens of millions of agent users, just look at that number. Tens of millions. We haven't brought this to the planet yet, and that's for agents. Not enough compute, guys. It can't happen. Regardless of what you listen to and hear, what you read in X, there will never be enough compute. Never, ever, ever, ever. Just like there's never enough intelligence necessary to solve the world's problems right now. So, you can sit there, you can argue, you can try to go through it, but I think it is a mistake to get into it. There will be capex air pockets. There will be time where bottlenecks are happening, and I believe we're in one of those now, or where the narrative gets too far, or the prices get too far, and we're going to get a pullback. But those pullbacks, those consolidations, those are opportunities for you to get involved. Customers are now asking, how do we control spend? How do we observe what the system is doing? That's the enterprise adoption. We're investing heavily in enterprise readiness and the tools for customers. This is where it's going to start to see productivity, the size of the market. None of us were anticipating how steep that curve is going to be. This is a vast transformation, far bigger than I think most people realize today.

Okay. So, we're in a consolidation. We're in a narrative-driven point. I think it's going to continue. You, these things are going to break out eventually during the course of this year. All of them, in my opinion. Um, or at least the majority of them. I've talked about, I've shown you where you can use the stuff that I'm giving you every weekend to figure out where the trends are. Over 85% of the names are above the 200-day. Over 80% are above the 50-day. The 50-day are pointed, pointed upward. Sorry, the slopes are pointing up. You get all of these healthy things. So, within consolidation, that's why I created this AI consolidation playbook. So, had a lot of people reach out. Where was the consolidation? I put it back in again this week. Um, remember for the pullbacks and everything there. This gives you, I gave you a prompt for the pullback hunter. It'll come up with a list of the names that are the best ones based on the sheets that you have. Here's part two of the AI skill architecture. Went up on the site this weekend. That way, for the hedge fund analyst skill, you can go build your own. So, for all of you that have done the knowledge brain, that are feeling more comfortable on either Claude or Codex, you're going to be able to just upload those and do them.

Now, for the sign of just whether you should be worried, again, I'm going to say I'll show this again and again, uh, not every week, but at least every month. When revisions start turning negative, when half of the companies all of a sudden are getting negative revisions and nobody's getting upward revisions, or if no one gets an upward revision and 20 companies get a downward revision, revision, you're going to start seeing some red bars. Right now, we are still great. This is the US estimate revisions. Again, S&P peaks here, revisions start going down, red starts showing up. That's what needs to happen. Here's Europe. Europe just posted its biggest one in the last two years. Um, Europe was suffering. They didn't have it. But look at this. Going higher. Starting to see the benefit in Europe. And on the global basis, again, sitting near the highs in just every week is just strong.

Um, payrolls this week. Uh, again, I, I, I read the payroll. I go through this. I'm so bored of people just not taking this. Especially with AI, guys, the aggregate weekly payroll is the most important thing. Weekly hours are not turning higher. They are sitting below the levels that they were from the period after the Great Financial Crisis into COVID. Hours are not expanding. We're having a nominal GDP acceleration. I listen to a lot of macro people, which I respect, that are saying wage inflation is going higher, the jobs market is tight, blah, blah, blah. None of that's true yet. Could it get there? I guess. I don't think it will. But one thing is certain. If we're at the point of AI agents, I wouldn't be getting too optimistic. Particularly when again, these jobs are being created in healthcare. This is healthcare jobs, healthcare jobs, healthcare jobs. It is dominating the monthly numbers. This is the aggregate, aggregate, uh, payrolls, which takes the weekly hours, the wages, and the number of people hired. And we are still sitting at very, very boring levels. The same levels that were going on past the Great Financial Crisis. So, there's no acceleration here as the AI boom is happening. Here it is on a rolling three-month basis of the aggregate. So, again, we're just around the same level we've been during okay times. There's nothing great going on. There's nothing horrible going on.

All right. To Worsh, if you can get a chance, you can go into YouTube. You can see him interviewed by Sarah Isen from CNBC. Uh, Kevin Worsh spoke at the ECB forum. So, let's go through what he said. We've seen this rejected forward guidance. You're just going to see a lot less information from the Fed. It's just funny that the first press conferences he does, he's hawkish, and now we're stuck in this limbo where we have no idea. But he did speak. The Fed should not spoon-feed markets. If people thought the Fed would tolerate inflation above 2%, they would be disappointed. Of course, he has to say this. This isn't a, he doesn't want inflation higher. I don't know why this was a, a big story in some hawkish thing or people that talk about it. The Fed independence. That's obviously important with what went on the last couple years.

This is an important part on AI and productivity. He said the improvement in AI models is moving at an exponential level and called it hyper-Moore's Law stuff. He suggested that current business surveys may be underestimating how quickly AI adoption will show up in the economy. Again, this means he is in, he's on top of this. He follows AI. He gets AI. I, I really would be listening that if he believes AI is about to do this, how quickly AI adoption will show up in the economy, why would he raise rates? And I'm going to go through this whole thing because you had banks saying three times, as of, I think Bank of America said three times. I could not disagree more, and I don't even understand where it came from. And I'll go through again why this makes no sense to me with, uh, inflation and wages. But regardless of what you believe will happen, if he's a pro-AI person, why would he raise rates? I just don't even get the sign. Why would he raise rates to get ahead of something when he believes deflationary pressures and job impacts? Those are the two things that AI brings. Deflationary pressures and at the same time job disruption. That's what it brings. There's no question about that. If he believes we're entering into that, this gets back to the Alan Greenspan thing. We are in the first or second inning of this revolution. We haven't even started, guys. He expects jobs to be greater and prosperity to be stronger over time, though the timing matters for the Fed's dual mandate. So, he's acknowledging the fact that, and again, what they have to focus on is inflation and wages and what's going to go if we have higher inflation, but the jobs market is weaker because the displacement is happening before the rehiring gets. He's got a dual mandate. He's got to focus on it. So, both of them have to be moving in the same direction. Sees AI as a major paradigm shift for central banking itself. Okay, again, I'm favorite economic indicator. He said his hope is that in 9 to 12 months, the Fed will be using new technologies to understand the real economy in a much more contemporaneous, real-time way. It's very important stuff. Uh, Fed is going back to first principles. Nothing new there, except for the fact that central banks got used to suppressing volatility, uh, higher VIX, higher, letting things, you know, if the market corrects 10%, not stepping in and doing something as long as it's not something big. Um, I mean, honestly, they haven't really had to do that, uh, aside from SVB, but we'll go see what's going on. Um, task forces are not meant to prejudge outcomes. This is on the task force side, but it's meant to bring the best minds in. He wants interest rates to be the monet, he wants monetary policy to be the tool because he wants the balance sheet to be reduced used. He made the point that it took 18 years to get into the current large current state of the large balance sheet, and that it would take more than 18 weeks to get out of it. So, he's just telling you he's going to do it when he can. Um, he was upbeat on US growth. Worsh came across as a hawk, hawkish reformer, but not in a simplistic rates now way. It needs to stop acting like it's still in the post-2008 crisis and overguiding markets. So, that's his thing.

Now, the reason I want to bring this up is because one of the stories, and I think one of the, the, the important stories, has been real two-year rates. Now, this is real two-year rates inverted, and I highlighted here because this is the end of March. Now, on here are Bitcoin, gold, uh, silver, the dollar, and then this blue line here is inflation expectations inverted. So, at some point, this white line started to impact the debasement trade. And what started happening is, oh, okay, the Fed's going to raise rates. And this is basically when, because inflation has come down and rates stayed the same and actually went higher in terms of expectations what the Fed would do, all of these trades started to sell off together. Um, the dollar was rallying. So, that's inverted. But between the dollar, gold, silver, and Bitcoin. And yet inflation expectations on the swap for two years are now at the lowest level of the year. Yeah, I'm going to guess that this is all based on belief that the Fed is going to do something.

Now, this is the chart of rate cuts out to the end of this year. So, we've gone to where the December, uh, end of this year, what has happened into the expectations. Well, now we're into the hiking side. So, I wanted to just show where we broke this, and especially here. This is exactly where it was. This is the May period where, up, the Fed's going to hike. That had an impact on things. And so, if people are right and the Fed's going to hike, I don't think that's going to happen. But if we get three hikes, I don't think you want to be near gold. I don't think you want to be near silver. And I don't think you want to be near Bitcoin. I don't think the Fed is going to raise rates. So, if that message is going to be delivered somehow, or if the market is now positioned for it, which is what we're seeing because we've got crowded positions and all of those trades the opposite way now, meaning they're all playing for the Fed to hike rates. It took a while, but if you go through all the coot positioning in dollar positions, uh, you go look at what happened in dollar-yen, people are just very crowded now into long dollar trades. So, if we get an unwind the other direction, I think it's good for gold, good for silver, good for Bitcoin. You guys know that that's the majority. That's where I've rotated the majority of my AI infrastructure money, particularly on the semi side. Uh, I still have Marvell and I still have Enterprise and I still have Fluence and some other ones that I've, I'm in. They're not as large as the memory side was, except for Marvell. Uh, but I've moved into Eli Lilly. I've moved into, uh, a bunch of silver and Bitcoin names.

So, here is the overlay of the 10-year swap rate, inflation swap overlaid with 10-year rates. So, the question is, when people say, and I listen to people say this, that the reason rates were going down is because the market is sensing peak growth. The market is following inflation. The only thing that's not right now is it's not playing exactly with it. So, if we get inflation data that starts to suggest that things are different, maybe that'll change.

Here's crude sitting down here, and the white line here is gas futures. So, the crack spreads have widened because the gas futures here. Gas at the pump is down to $3.80-something from up near $4.50. So, you've seen a 17, 18% decline. But if crude stays down around this level, around $70 bucks, you're going to see gas futures come down. The reason that's important is because we're already going to get the negative inflation side, most likely in the next print for headline CPI. Currently for the June number, we have, uh, slightly below zero, and currently early at this point for July, we've got minus .2 from the Cleveland Fed. Well, if we get those numbers, we're moving the CPI, which was at 4.20, down to three and a half year-over-year. So, I just find it very hard with that going on for people to get all excited when they can basically jump on one thing, which is PCE core, which, oh, that's right, that's the one the Fed prefers. Well, let's go remind ourselves. Back in May, Kevin Worsh spoke in front of the Senate Banking Committee. And during his confirmation hearing, that the Fed's current preferred gauge, the core personal consumption easy. So, hearing that the Fed's current preferred gauge, the core PCE, offered only a rough swag on inflation, being short for scientific wild guess. Okay, this is Worsh mocking the PCE core. He has stated for a preference for trimmed gauges, especially the Dallas Fed's trimmed mean and the Cleveland Fed's median PCE. Well, here is the Cleveland Fed median PCE sitting down here. Not too much higher than it was in the period after the Great Financial Crisis. So, yes. Is it about, say, 50 basis points higher? Yeah, that's a lot different. But here's the other thing. This is where the Fed funds rate was during this time. This is where it is now. We have the Fed funds rate way above this at this point. It's 80 basis points above. We've got restrictive monetary policy if he's serious about this, compared especially to the prior decades. So, why would he hike?

Let's continue wages. The white line here is the Atlanta median wage. Back to the median again, 3.5%. Here it was when the Fed started hiking, up at six and change. Here's where we are here. Basically, we are exactly the same on the wage side as we were from 2010 to 2020. I put this over the quits rate because the quits rate lines up with it. And the quits rate is, is important. And the reason it's important is because it's a good sign of whether people feel it's okay to jump to another job. The market, the labor market is not strong. It is not strong. So, using the backward way of just saying the number of people hired or using the claims, whatever it is, everyone has a job, but the ability of going out and getting higher wages, AI is having an impact. If you strip out healthcare, there is no jobs market. So, again, look at where you have the quits rate. We're in the middle. I mean, I, the wage right here, the orange line, quits rate here. Like, we're in the middle, guys. There's nothing there. If you go look over here, the effective Fed funds rate over the Atlanta Fed median wage. Again, here's where the median wage was from 2016 to 2021. This is when we saw an uptick. This is when they raised rates. Okay, so here's the Fed. They raised rates. We were here at the peak here. So, we're not too far from where they are. I just don't see a need at this point based on the data that is there. And oh, by the way, how we have forgotten the narrative that was in play for the prior two years that we ignore now. The federal debt path is worse than reported. And here is the CBO. How do people forget this? This is why you're long gold. This is why you're long silver. This is why you're long Bitcoin. The situation on this. This is the math. The math just gets worse. Now, as an AI guy, do I think this number is overstated? Yes, because I think we will be curing diseases and the mandatory spending won't be as much. But we're not there yet. And I don't know if everyone's going to get it, even if we are able to do that. So, the discretionary spending is very little. So, when you get back to it, the mandatory spending for 2036, $7 trillion in net interest, and this is assuming that the Fed funds rate is here. If Kevin Worsh raises rates three times and fights in the, the dreaded inflation, which is nowhere, I, I, people are, this is PTSD on what happened before.

All right, here we go. Application layer. Remember this stuff, everyone's long. Nobody's in this stuff. This is where Greg Brockman's talking about. This is where Scott Woo is talking about. He's talking about applications. Remember, companies are not adopting AI tools. They're adopting AI workers. Productivity, productivity, productivity. You start seeing profit margins go higher. Scott Woo again, we are already moving into a world where agents can perform hours of work.

So, the KIX, the KIX, the insurance ETF or index. Yeah, this is not the ETF, or maybe it is. I can't remember. Um, regardless, we're breaking out here on the insurance side after a long consolidation. Remember what I showed you on KRE? KRE, regional banks. I think you're going to have a lot of consolidation within the regional banks. I think they're going to be buying up the smaller banks. Uh, again, I think the bigger banks will be the AI ones. I think companies that can do this on their own books are going to buy up smaller banks. Put the same thing in. They're going to be able to immediately turn them. You have to think about this. If you're big enough to put AI into your business, and you're able to cut workers, you're going to be able to do the same thing with smaller banks that don't have the budget, don't have it. You're going to see a lot of consolidation within these medium-sized banks into smaller banks, in my opinion. And if not, it'll come from the bigger banks. But you're going to see consolidation, M&A.

Uh, I wanted to show this on the breadth side. So, remember all the Hindenburg stuff that I was highlighting about the market? Well, now that we've done enough shakeouts, we've done enough kind of consolidation here, we're back to 8%. This was on Wednesday. 8% of 52-week new highs in the S&P. So, about 40 names out of the 500 with only one 52-week new low. Now, 13 of those 40 were in financials. Nine of those are in healthcare. You're looking for breakouts in them. You're looking for this. You see a lot of things in here. Let's go into the next one. Uh, inside the insurance side, CBTRV, you've got some of these on there. You guys can go look at the charts yourself. Within the PNC insurance side, AI dominates agendas. You're starting to see the same thing that I talked about within the infrastructure. They're starting to see it showing up on many, many levels. It's worth reading about it. In particular, I would go read on Travelers and go read on on Allstate, which are both two names that showed up on the 52-week new high list. So, the KBW insurance index, worth looking at. I don't think you're going to see massive moves in these, but you go back and look at McKesson. If you can improve just a little bit on these bloated companies and get rid of some of the paperwork and make them more efficient, get rid of some of the call centers, you can bring a lot to the bottom line very quickly.

All right, let's get into Meta. Building a business to resell AI compute had everyone excited this week. Oh, they're going to, that was this is this is not something to pay much attention to other than the fact that Meta looked at SpaceX valuation and recent deals to sell exit capacity and came to the conclusion that investors will look favorably. I happen to agree with this. But remember, in March, Meta signed a $27 billion cloud service deal. In April, they signed a $21 billion expansion of their cloud services. Uh, two weeks ago, Bloomberg reported that Meta signed up rent 1.0. Now, all of a sudden, they have excess capacity. I mean, it's really stupid to to to leap to that side and not see that maybe it's because their company's under pressure because the stock is acting so badly. They've got a company backlash going on. So, let's just go through what they said in their earnings. We are still compute constrained. This is from their Q1 earnings, most recent one. And even as we aggressively add capacity, demand keeps outrunning our own expectations. Compute is now strategic, not optional. They are aggressively trying to secure capacity. The biggest need is inference capacity, not just training. Again, I don't, I don't know. Continued to underestimate our compute needs, even as we have been ramping capacity significantly. She tied that to faster AI progress, new product ideas, and growing internal use. They are building with flexibility in case they overshoot. That didn't make investors happy. So, that is the reason why that I'm sure they're saying, hey, if SpaceX got away with it, maybe that'll allow us, as we're doing this, to say that we have that option so people at least put a put underneath.

Um, Meta's potential mega equity raise puts AI ambitions in dilution focus. Remember after Google raised money, they talked about going out to raise more money. Why would they go raise more money if they already have excess compute? Why would any of these companies, why would Google do it if we're at the beginning of the end of needing, uh, compute? Uh, this is more likely having to do with it. I think this is a good way to look at it is just for all of these companies, the three, the three bad boys, Microsoft, Meta, and Oracle. The free cash flow, the price to free cash flow, negative over here, negative over here. I mean, at some point, if they're not getting it in quick enough, it's hurt. And here are their stock.

prices for Oracle, Meta, Microsoft. So you get this bounce in Meta. It came all the way right back down to here. These companies bounced slightly with it. All the companies that were hurt with it, the Neoclouds, they're still down. This was just a story again that got taken out of proportion and you'll just uh I'm sure not have to deal with it. And it's possible Meta starts showing uh revenue gains. I I doubt that's going to show up, but don't be surprised if we don't get some bounces in these names. They've been hurt so much.

Um, this is the more likely thing is that the physical capacity could limit hyperscaler capex surprises. If they don't announce big bigger numbers, it's more likely to be because of the bottlenecks. And that's basically what I've been saying now for a couple months is the rate of change is going to go down. You're going to see disappointments or you're just not going to see the amount of surprises that you saw before. The worst case scenario for the US is becoming increasingly realistic and it's basically that we have few too few data centers. In particular, semi analysis put out a chart just basically showing we don't have enough of this. I'll get into that. I'll show you that chart in a little bit. The BIS put out something just showing the risk involved in it. And here is the semi analysis side just saying um how much gigawatts we need based on the growth in AI and chip supply. And we're just down here. We're not anywhere where we need to be.

Now, just a quick thing on GLM 5.2. Um, again, American and European enterprises will ditch open AI and anthropic and adopt Chinese models. I don't agree with that, but again, uh, case concerning the worst case scenario for USA, Chinese open source keep getting this is from Mark Andre. So, again, you don't want to ignore Mark. Many smart people are saying GLM 5.2 is the first Chinese model to match and often beat the American big watch model. Incredible timing given our current events. This is the clawed moment. This is from the head of data bricks. So a company that would know the demand we're seeing at data bricks is astonishing. The world is going to see massive adoption of open-source LLMs. More companies will shift towards post-training their own models on top of the OSS models and owning the weights. Um, reminder, Gavin Baker said that we're going to go to more composable markets and said he had to challenge his thought process because of GLM 5.2. So, in my opinion, this does create a potential issue for the slowing down of the ARR for Enthropic. I believe that is a possibility. Now, remember, they have been a 10 timeser in terms of the revenue. You heard many many places say, "Well, if I multiply it again by 10 and then they'll be at a trillion dollars at the end of next year." Anything that starts showing that they're not going to hit 10 times, I think this will become a major story, which I think is realistic. I I do think that open- source models are going to be used by a greater portion of the country uh than people realize particularly as these expenses go and particularly when we get the hardware out there to be able to run it.

All right, now to finish up the last few slides for Bitcoin. Reminder, here is the AI tech part V momentum. This gets really important. As this goes higher, it's getting harder and harder for people to speculate with inside the AI infrastructure trade. All of the headwinds I showed, I think you're going to make this not as easy. So, if going forward we're getting 40% a year, that's still a great number. It's better than the S&P will do, but it might disappoint people that are looking for eight and 10 baggers.

So, now we go over to Bitcoin. Bitcoin sitting on its 60-day V at basically the lowest V it's been. um and it's been sitting down here now for quite some time. So, we've had a restructuring involved. You can thank Michael Sailor's um job for that. As people, as I said last week, when people are calling me up saying Michael Sailor is going to blow up Bitcoin, he's horrible for it. That's usually the point that the bottom comes in. So, I start looking a little more closely. Here is the rate cut chart overlaid with Bitcoin. So regardless of whether you like it or not, as we've taken out the rate cuts and actually gone to minus cuts, meaning hikes, Bitcoin has followed it lower. So if we get any kind of change right now or stabilization where this trend doesn't get as worse, maybe we'll see something. I just want to let you guys know that over the last 100 days, we are back where we were. So we're at minus 5.8% as of Thursday. That's pretty impressive. Meaning for all this bare market, the rate of change has been negative the entire time, but now we're getting less bad. And I think that's important if we start to break higher. You can see that the 20-day moving averages here. We need to see some positive technicals. And here we go. So, for the first time, we have divergences. And what I mean by that is we actually have lower lows and higher highs. And this is going back to July of last year. It's been a long wait. I've been looking at this. I have this up on my Bloomberg. I've been waiting for divergences and then I wanted to see a week where it happened on bad news. Clarity Act passing falls to 39% and President Trump discloses 1.4 billion in income.

All right, um I do want to leave you guys with one more thing on the Bitcoin side. again, now that we have hikes built in, if they just stay status quo, and Kevin Worsh keeps saying that, you've taken the pressure off. Trump hired Kevin Worsh, and even though he wants him to quote unquote fight inflation, fighting inflation with talk is very important, but the midterms are coming up. So, I find it highly unlikely for him to be too hawkish ahead of the midterms other than just say that inflation shouldn't be here. And if we get negative CPI prints, I do believe it gives him the ability to be more patient. That might be the side that ends up happening. So, I'll leave it there, guys. Again, I hope you had a great fourth. I'll see you next week.