Bottlenecks, Momentum Breaks, and the Next Phase of AI

In Maine for the week. We'll be up here for at least a couple months soon. Holiday weekend.

I'm going to cover a lot of important stuff today, because I think we're getting close to an inflection point, and I want to make sure this is clear every single week.

From a trading perspective, we're in the middle of a secular trend in AI. You're not going to hear me say this isn't going to happen. I'm not worried that it lacks financing, that it's a bubble, or that once it collapses the whole economy goes down. This is a different time. It's being funded by companies that have the money, and the revenues are coming. They're just delayed, but they are contracted, and the momentum is unlike anything we've seen before.

That said, there are no free lunches. Markets are always about risk and reward. We've had a huge run, and I am seeing some global and sector correlation breaks that usually warn you to be careful. We had a very volatile week for momentum. Momentum shifts often happen around regime shifts.

I also think we have to start building in the fact that it's now been two months of talking about the Strait of Hormuz and how important it is, and there are still no ships going through in any meaningful way. Inventories are being drawn down. This is about the factual side, and now inventories are breaking through the lower end.

We also had Google I/O. That's important for two reasons: one, to show what happened last year, and two, to connect it back to crypto briefly.

For the RIAs, I am working with Morgan Stanley on index launches. I'm not planning to do an ETF on the AI thematic portfolio, but with the 100-name portfolio I am working on getting two things. First, an index that's tradable not just for mutual funds and RIAs but also for hedge funds, options, and everything along those lines. But most importantly, you have to understand that AI outperformance is driving the entire market. In my opinion, it's driving the entire economy. I'd debate any economist on that.

If it wasn't for AI and the buildout, if it wasn't for the exploding profit margins in these companies and the wealth effect that creates, I don't know what would be happening in the economy right now other than needing the government to step in more aggressively.

I've also worked on a concentrated portfolio that should go up this week. I've had a lot of RIAs looking for something that's not 100 names. So now you'll have the index for people who can trade ISDAs, and you'll also have concentrated versions, one with 10 names and one with 25. I spent a lot of time building a list that replicates the portfolio as closely as possible.

The S&P was up 88 basis points for the week. Rough start, good finish, overall a strong week. QQQs were strong again. But in the S&P we're extended. This is a lot of weeks in a row. We're going to get a correction at some point, and I think it will be more than a one-week correction when it comes. It wouldn't surprise me if it started next week, with mean reversion becoming more important after such a huge month, especially since momentum showed weakness last week.

QQQs were up 1.2%, IWM up 2.8%.

The hyperscalers, Meta, Google, Amazon, and Microsoft, as an equal-weight basket, are basically unchanged now. As much as the AI trade is driving the alpha, these names are the key to the market because they are the spenders. As long as they bounce, everything is okay. But they broke below the 20-day moving average. They've been basically unchanged over the past three weeks, and I would watch that because it tells me the market is getting tired in the places that matter most.

I ran the momentum work on Thursday, and by Friday I wanted to highlight that this historically isn't a great setup. When momentum gets that stretched, it either means we're at a top or we start building a top and then head lower, taking a while to recover the highs.

This week we did have a sharp move lower in momentum. The S&P didn't really follow it on a close basis, though we got much lower on Tuesday morning. That meant we gave up a significant portion of the last month's momentum rise.

I still think we're at the beginning of a secular trend, which means momentum will continue to work over time. I don't think the names will change that much, because we're at the beginning of a buildout for the agentic world. This is a once-in-a-lifetime opportunity over the next three to five years, something like the commodity bull market that happened when China was building out from 2002 to 2007.

Nothing could stop that train until the U.S. blew up. There was just a lot of money being spent. So what we're seeing now are continued unwinds and bigger two-day swings on both sides. I think that's what people have to get used to as the trend extends.

We've also got a historic breakdown in correlations between software and semis. If we're going to get a momentum-sustainable move in the other direction, it will include software that either doesn't go down or actually goes up while semis are coming down. That's what I'd be looking for.

I wrote about this for 22V. I don't think this ends because of a bubble. I don't think it ends the way most people are writing about it.

When Amazon and the Mag 7 were going up, it never really ended. You just got periods where they corrected, but nothing stopped them from demolishing other companies and building moats. I think AI is going to be like that. But the risk is that at some point we hit bottlenecks. Hoarding happens. We don't have the physical world flexibility that software had, where nothing could stop growth. Here, growth can be stopped simply by not being able to build enough stuff anymore, or by all the inventory being tied up. That's the risk I think matters.

There are two parts to the AI trade. One is more on the software and semis side. The other is the physical industrial buildout side. If they're not getting what they need from natural gas, oil, or the chemical derivatives of that, they're going to have issues.

I believe the market usually shows where the pain will be before the headlines do. The world industrial index overlaid with the S&P shows that the S&P has gone to new highs while world industrials are still below where they were before the war started. We went up into the war, came down, bounced, but never took out the highs, and then failed there. So industrials are being impacted.

Then look at the machinery indexes in Korea and Japan. The industrial winners are under pressure. My thematic portfolio is up overall, largely because it holds many U.S. names and not many Asian industrial names. But within the physical side of the chain, you're getting a breakdown.

In Korea, construction and machinery have both peaked and moved lower. SK Hynix is a critical part of the AI market, and Korea matters much more than people appreciate.

Japan matters too, not just in one niche but across chemicals, semiconductors, and turbines. Names like Mitsubishi matter in multiple parts of this system. The TOPIX had rallied, but since mid-April we're seeing a divergence where my thematic portfolio kept ripping higher while Japan started to weaken. A lot of that has to do with what is going on in Korea, SK Hynix, Micron, power semis, and the whole supply chain.

If you go deeper into Japan, machinery looks the same as Korea: it went up and then back down. Construction looks even worse. Japan is dealing with higher oil and higher long-term rates at the same time, and that creates another pressure point.

Now let's go to sentiment. I don't use many sentiment indicators, but the Goldman Sachs risk appetite indicator is one I do watch, and it is at the highest readings since 2021. That's not a reason by itself to get bearish, but it is a useful warning that sentiment is extremely strong.

Micron is one reason everyone is so positive. Maybe we just pulled forward two quarters of orders and the next one comes later. But when net income is up at these levels, you're talking about something close to years of previous profits being compressed into one quarter.

We'll get another reading in around six weeks. The other issue is that profit margins in the S&P are almost entirely seven stocks. Margins are fine across other sectors, but most of the margin expansion that pushed the S&P to these levels comes from a very small group. Add Micron, SanDisk, and Western Digital, and you realize how concentrated the story is.

If margins came down, not because the cycle is bad but because there was hoarding in the first quarter, shortages, and then demand weakened because people overbought or costs got too high, quant strategies and risk systems would react badly. You could easily get unwinds and a very different risk backdrop.

Forward sales growth is projected at 18% over the next 24 months. I do think the orders are real, but I think you will have lumpy problems along the way because of bottlenecks. I don't know how long you can keep ordering if you don't take delivery. We already see delays in remaining performance obligations at Oracle and elsewhere. Nvidia used to have a big receivables issue and still does, though they've shifted things to make sure they get cash. Microsoft, Google, and Amazon cloud all face similar timing realities.

When you have multi-year delays in so many parts of the chain, the risk is that the buildout becomes lumpier, more inflationary, and more volatile than the market is modeling. That's what I believe is going to happen.

I can upload a chart and ask, what is the risk? And the answer is straightforward: bottlenecks create cost inflation. Cost inflation creates delays. Delays create revenue recognition risk. We are already seeing all of that.

The market is taking Q1, which came only a couple of months after people were still saying AI was a bubble, and extrapolating too much from it. I started hearing people accept memory only in October of last year, once memory prices had already gone higher. Before that everyone was still worried about tariffs.

Now you've had a huge move and I just don't think the market is fully accounting for what's going on under the hood.

There was also a must-read interview with a Siemens employee explaining how big the demand is for energy equipment right now. We are only about 12% of the way into this process, and the key question is whether decisions are speeding up or slowing down. Decisions get made quickly, but production and delivery still get delayed.

One line stood out: imagine getting orders that are double what your entire factory can produce in a year.

The Dodge Momentum Index slowed during winter but is starting to pick up again. Still, labor shortages, high material costs, and supply-chain disruptions are weighing on confidence.

Year over year, the Dodge Momentum Index was up 14.1% versus April 2025. The commercial segment was up 37.2%, but only 5.8% if you strip out data centers. That tells you everything. Data centers are the story.

This is also why I think people should pay more attention to oil. I spent a lot of time listening to Art Berman this week. I haven't tried to go too deep on oil beyond the main point, which is that we still haven't gone anywhere, sentiment doesn't care, and now it should care because we are starting to see draws in inventories.

The shortage is now hitting U.S. inventories. We've been releasing from the SPR, and that is part of the draw, but total U.S. inventories are still tightening. If the energy side gets tighter while this AI buildout continues, that matters a lot.

I recommended some podcasts and discussions because they do a good job of going under the hood. Adam Parker's idea that investors need to unlearn the past is something I strongly agree with. I wrote about that in a paper called The Art of Unlearning back in 2016.

You're going to have to start putting math on everything differently when dealing with AI, debt, the size of the country, and the printing press. Exponential innovation has changed the setup.

On software versus semis, that divergence chart I showed came from Adam Parker's framing.

On bottlenecks, the grid problem is also a real issue. The data-center buildout probabilities by 2030 are not comforting. We have a lot of high-risk and acute issues if we want to get this buildout done on time. To assume it all happens cleanly is unrealistic.

Again, investors are using mental models built for the last regime. Capex-heavy businesses may deserve a rerating. Capex is not wasteful overhead anymore. It is the toll to participate in the new growth cycle.

That is why “your capex is my opportunity” matters so much. If someone else's spending is the revenue source for the right suppliers, you want to own the suppliers.

Now let me connect this to crypto. I think crypto is in a similar place. People still have not fully connected that AI infrastructure is the main trade right now, but at some point the handoff becomes the redesign of the global financial system.

If you watched the webinar I did with the Kraken team and the other groups involved, I think this is coming very soon. People should be doing their homework on the space.

The Google I/O side matters because of agents. You don't even need to read every detail; just look at what we're talking about across the ecosystem. Combined with search, the agentic world is going to be a huge jump.

For Bloomberg users, I highly recommend watching Sundar Pichai and at least learning what Google is going to enable, because we're getting to a point where agents will let you have 20, 30, 40, 50, 60, 70, 80, even 90 agents all working on a problem for you, each with a different specialty. That's what makes token demand go exponential.

Then there's the Coinbase side. Whether it's tokenization, stablecoins, or the onchain economy, I think people are missing the parabola that's forming.

One group many people ask me about is why it isn't participating, and I want to highlight Vistra. It's basically unchanged since the end of 2024. Its multiples were too high back then, and at that point there were fewer names to invest in on the AI trade.

By October, the broadening out happened. Relative to Nvidia, names like Vistra look to me like the defensive side of AI, especially if we're close to some kind of momentum unwind where the hottest names are running out of steam.

This is almost the inverse of Micron. It doesn't have the same parabolic downside, but people were looking for beta. Within AI, these are the defensive areas.

If I wanted to play something over the next months, I would be increasing size in names like Vistra. I also created an index of 10 IPP names, the independent power producers. That group looks consolidated, and if defense starts working inside AI, I think Vistra and some of these IPPs are where money rotates. That is where I would move some capital to reduce beta while staying in the theme.

If you subscribe to the free Substack and still cannot see the paywalled material, that is because the subscriber wall is separate. I'm just bringing it up because people keep asking where to find the details.

The main message today is this: the secular AI trend is intact, but the next phase is not going to be smooth. It will be shaped by bottlenecks, momentum breaks, energy constraints, and the transition from a narrow AI trade toward a broader system-level rebuild of both infrastructure and finance.