Running Hot Into Scarcity: Why Bottlenecks Are the Risk to the AI “Bubble”

All right.

Back from Florida and getting ready to head up to Maine for at least a week, maybe longer, and preparing for the move up there for the summer.

This will be a different video.

I really do believe that the combination of some of the things we've been talking about for the last three weeks is getting us ready for the next regime shift, which would be about inflation and the pressure it would put on the Fed, with yields possibly moving higher.

At the same time, everyone has now been forced in, and we clearly have the next phase of parabolas. I want to remind people, because I'm writing a paper on this, that last year the main parabolas of the year were Palantir, gold, and silver. Palantir had a similar move to what Micron has had over the last 18 months. Gold doubled in a year. Silver was a four- or five-bagger.

Those things have been consolidating, and in a world of bubbles, parabolas, and speed crashes, these things shift quickly.

There is absolutely no doubt that between the leverage ETFs and breadth breaking down, the market is now in a different position.

For me personally, I've been scaling out of things. I mentioned that I had taken two-thirds of Micron off. I am now, effectively, out of everything, definitely out of all of Micron.

That is not a negative view, and it is not me wanting to short this stuff. I will be looking to accumulate things, as I have been doing in silver and in Bitcoin. Those things are not at 52-week highs and do not have any bubble signs right now.

You're going to have rolling bubbles, but the agenda for today is hoarding, bottlenecks, and the next risk phase.

We're running hot into scarcity.

I wrote a paper this week on that exact subject. The government wanted to run the economy hot, but they are running it into scarcity. At the beginning of the year, my trade was short abundance and long scarcity. That has worked very well. It is not just the thematic portfolio that is up significantly this year. The other side was being short software. If you combine the two, you're talking about massive outperformance.

Now, five months into the year, the scarcity trade is becoming the dominant part, but scarcity brings bottlenecks, shortages, and changes in earnings. That's what I think is going to happen.

Let's go through the inflation data.

CPI came in in line, a 0.6 print, but it takes year-over-year CPI up to 3.8%, and it continues the trend. The important part is not just that it was 0.6. Outside of COVID, it is one of the largest prints since 2012.

PPI came in even hotter. This is not just an oil story. This is scarcity and hoarding tied to the dramatic shift in AI compute demand.

I've talked about shortages and model changes. I have moved almost completely to ChatGPT at this point. 5.5 is by far the best model. Codex has replaced Claude Code for me. I moved all of my OpenClaw-related work over to Codex 5.5 and saw amazing capability, especially while I was in Florida.

At this point, the compute shortage, the hoarding, the leapfrogging, and everything required to keep up with the first quarter's demand add up to a dramatic shift. OpenAI 4.5 and the move into the agentic world came much faster than anyone expected. It happened at the same time the one big beautiful bill and bonus capex depreciation were in place.

So hoarding is going on. Leapfrogging is going on. Revenues went higher. We have RPOs all over the place.

A lot of things are built into the market that can be delayed simply by not being able to build out data centers fast enough or get everything plugged in. That is where I think we are.

It is not a negative story, but you have to get used to AI cycles versus traditional consumer and manufacturing business cycles. AI cycles are part of a 90-trillion-dollar buildout that is going to happen, but we do not have enough supply for it. That means you're going to run into situations where something breaks.

Import price inflation is significantly higher. These numbers were already tracking high even before the war. That is important.

We are going in at levels that are much higher than what could have been expected. PPI, the input-cost side, overlaid with year-over-year CPI, makes the same point. I don't see how this avoids feeding into headline inflation unless it proves to be a temporary spike, and I don't think it is just temporary.

We are now back in a different regime.

Since ChatGPT was democratized and launched, rates had been above CPI. Now CPI is breaking above rates and three-month bills, so we have negative yields.

Around the globe, 30-year U.S. yields, 30-year JGBs, German yields, and gilts are all following the inflation situation. This is not just a U.S. thing.

There was no solution to Iran, so the Strait of Hormuz is still shut. That is the major issue. We are well past the point of ignoring the impact this is going to have.

It has obviously affected oil prices, but the longer it sits here, the more inflation pressure builds and the more pressure there is on central banks to respond and potentially raise rates.

Traditionally, that would be very negative for the market. But there have been two positives that have overwhelmed it. First, earnings growth dominated by the AI sectors has been far better than expected. Second, there has been a momentum chase, and I think we are close to the end of the biggest phase of that chase.

Supply-chain stress that peaked in COVID is moving higher again. The global supply-chain pressure index and the World Bank's supply-chain stress data are both reflecting new strain. You can see it in PPI, supplier delivery times, and everything else.

There was also a very important unanimous Supreme Court ruling this week that may not have reached most people's news feeds. The ruling shields freight brokers less than before and makes them potentially liable if they hire unsafe trucking companies. This could be an extinction event for 30% to 50% of freight brokers.

The issue is that it could create major supply inefficiencies and destroy hundreds of thousands of transport companies. The fact that it was unanimous means it is not likely to be reversed.

We also had the worst U.S. spring drought on record since 1895.

We got the crop report. Conditions are poor. These are all bad situations. El Niño is another thing you are going to hear more about this year. You can try to solve Hormuz. You cannot solve weather. This can become an issue for many countries around the globe, usually through dryness and fires.

I bring it up because when you combine Hormuz staying shut with weather-driven stress, inflation pressure broadens even further.

The bonus depreciation and front-loaded incentives allow companies to elevate earnings for now. The question is what happens when that rate of change shifts.

I put together a framework around peak Q1 earnings because people need to understand what happened. We had insatiable demand. The rise of agentic AI was not expected. The speed of adoption was not expected. Once agentic AI showed up, all of a sudden demand turned parabolic.

Parabolic demand means more chips, more cooling, more tubing, more optical fiber — everything. That sets off the strategic AI race. It coincides with the one big beautiful bill. Massive amounts of dollars start flowing through. RPOs build up. Companies are forced to compete with each other. That leads to all sorts of components being ordered.

The problem is there has almost certainly been over-ordering and hoarding. Even if it did not start that way, every week that goes by with the message that if you want it you need to buy it now pushes pricing pressure higher.

Then you add the Strait of Hormuz being shut down — an April event that is still real.

This is a big mistake for people to ignore now. There is risk here, and this is not a bearish thing in the normal sense. It is more that you have to be ready for speed crashes when you're in parabolas.

I did a special video this week to go through a growing set of warning signals suggesting the AI leaders may be entering their next phase of rising volatility or consolidation. What stands out is that weakness is no longer isolated. Signs of contagion are showing up in the hottest market in the world, South Korea.

I walked through the possibility of a momentum unwind. Last week I also highlighted the exhaustion model I upload every week for subscribers. It breaks things down into extreme exhaustion, elevated exhaustion, somewhat exhausted, and low exhaustion.

To test how it worked on Tuesday, I had OpenClaw calculate the average and median moves for each category. For the extreme exhaustion bucket, the average move was about negative 4.1. That told me the exhaustion model is working.

We've also seen oil-sensitive things that had been very correlated to SK Hynix start to break. SK Hynix is one of the key lines I watch. The KOSPI machinery index peaked and started to break down. Construction has also weakened. When I see that, it tells me not everything is benefiting anymore, and oil is part of the reason.

Another chart that bothered me was power semis.

I highlighted power semis back in late March. This is a basket of six names I recommended. In less than a month, they had enormous moves, and they have not even fully seen the demand yet. A lot of that move was simply because they were cheap.

The power-semi side is going to be driven partly by edge devices and partly by the DC conversion side, especially around Nvidia's 800V DC architecture. A lot of those things are still on the come.

When prices start moving that fast ahead of reality, that becomes a classic momentum trend. For people who made money on it, that's fantastic. But when you're trading parabolas, you have to be ready for the turn, and I think the probability of that turn is increasing.

Goldman Sachs' desk also bothered me. They said they had seen sovereign wealth funds and asset managers being stopped into the tape as buyers since the calendar flipped to May. A clear momentum chase, highly concentrated in AI.

This is the rotation I wrote about in Institutional Investor in my article, Your CapEx Is My Opportunity: The Benchmark Arbitrage of the AI Buildout. Institutions were underweight the things that were working and overweight the things that were not, so they had to chase. That benchmark arbitrage is what I have been trying to get RIAs and advisors to focus on. It will be a permanent part of the next decade, even though there will be temporary disruptions.

The thematic portfolio finished the week down 46 basis points. I don't show that because I think it was a good week. I show it because it had just had an enormous run. Everything related to capital goods is through the roof. The earnings are justified in terms of the rotation.

When I look at semis versus the Nasdaq, especially SOX over NDX, a big part of that is justified. The semis are winning relative to software and the hyperscaler spenders. If anyone says this is a bubble, the bubble is in the spending and the race to catch up.

There are two ways bubbles end. One is the dot-com version, which I do not think is the case. The other is that we run into bottlenecks. That is what I think this is.

I overlaid DRAM prices with SOX relative to NDX and found that going back to 2017, they track pretty well in z-score terms. Historically, when DRAM prices start to roll over in second derivative terms, SOX over NDX also tends to roll over.

Now, DRAM pricing momentum has turned negative. I would be very wary of that in semiconductors.

The risk is that once DRAM momentum rolls over, the semi trade becomes more sensitive to bad news because the market can no longer rely on accelerating memory prices to drive the story.

That is why the bottleneck matters so much. At this point, any kind of bad news becomes more dangerous because everyone is already in the trade. We are going to get episodic moves.

There was also an AI windfall-tax debate in Korea. Whether it happens or not, it shows how politically concentrated and visible this trade has become.

Momentum itself has become isolated to the AI trade. People have been trying to call the top of SMH versus software for some time, but it kept going higher. Now, after a two-and-a-half-bagger in a short amount of time, you need to respect the possibility of speed crashes.

The reason I am out of Micron is simple. I bought lower. I had a huge gain. If you get a five- to eight-bagger in your life, you should respect it. I may miss more upside. It may keep going. I may never get another chance to buy it. That would not surprise me. But I am still fine being out here.

When I compare Micron with the Mag 7 ex-Nvidia, Micron went from being the worst performer as of Q1 2025 to by far the best. That tells you how violent this rotation has been.

If you look at the winners over the last five years, nearly every top performer is an AI name. If you avoided this "bubble" for five years, you've missed a huge move. Some names are up 37 times. The move is real.

The question now is what it looks like a year from now.

I also showed data on the number of days with plus or minus 1.5% moves in the U.S. momentum index. Since the end of 2020, the trend is clearly higher. In the old QE world, this was not the norm. Now we are living in a period with much more frequent violent momentum shifts both up and down. This is what I mean by speed crashes and unwinds.

Estimate revisions are still massively positive. That is why it is hard to have a sustainable down move in the market without the red bars suddenly appearing or peaking. What I worry about more is a peak in revisions that then starts to move lower.

Because if the market had a parabolic move and revisions start to roll over, that is what causes the fall.

I also think Hormuz could create supply disruptions that feed into semiconductors and affect the ability to produce. Remember, when you are selling something, it is not only about high prices and strong demand. It is whether you can actually make the necessary volume. At some point, you may have sold everything off the shelf, and then there is nothing left to sell. If production cannot keep up, that is the issue.

The whole data-center stack is interconnected. One piece not getting built makes it difficult for the rest to get done. That creates real supply-chain correlation risk.

And again, I am talking about changing risk-reward because prices moved higher, earnings moved higher, and supply disruption remains unresolved. Hormuz has not been fixed. That means there could be a shift.

The estimate revision factor for the S&P still justifies the index moving higher. But if revisions start to move lower, that is where the issue begins.

Another reason not to listen too much to people screaming bubble is this: since the start of 2026, the S&P is up 8%, but consensus forward estimates are up 13%, which means the multiple has actually gone down. In bubbles, multiples go higher.

On the turbulent side, nothing has shown up yet, because this has all been one trade. Rates had not moved until Friday. Oil had moved higher, and that mattered, but the macro framework had not yet broken the trade.

Where we are seeing warning signs is in the divergence between the S&P and the VIX. The S&P is far above prior highs, but the VIX is also much higher than it was back then. That is a warning.

If you look at household durables, only four of twenty-six names are up on the year. In restaurants and retail, there are huge drawdowns. Breadth can be bad for a long time.

I still think we are in something like the 1970s, meaning up, down, up, down, up, down. Over the next decade, because of AI disruption, I think it will be very difficult for the capex spenders to avoid multiple compression.

Code2's reports make this point well. The fight right now is between the sellers of AI inputs and the hyperscaler buyers. The sellers are the new leaders because they have cash flow today, not sometime in the future. The hyperscalers are spending heavily now.

Their earnings are not being hurt yet because they get to depreciate capex, so they still get an earnings boom. They still have significant pricing power and all-time high margins. But if you look at free cash flow relative to capex, they are being penalized. Their multiples are coming down gradually.

That is the tradeoff. Some of these supplier names have doubled, and they are doing it at the expense of the hyperscalers.

Microsoft now has capex equal to 37% of revenue. Yes, there is a huge RPO backlog and future revenue visibility, but if bottlenecks last too long, they will not be able to realize those revenues on schedule.

The market would not have gone higher without the hyperscalers bouncing. The semis would not have worked without that bounce. Retail would not have joined to this degree without that bounce. But even so, the hyperscalers remain the chart I am watching, because year to date they are still down relative to the S&P.

So the practical takeaway is this:

The secular AI trend is real.
The scarcity trade is real.
The earnings have justified much of the move.
But the next phase is about bottlenecks, hoarding, logistics, weather, freight, energy, and speed crashes.

That is the risk phase we are entering, and it is one investors need to understand without confusing it for the end of the AI story itself.