In July, Bank of America's global fund manager survey found 82 percent of managers naming long global semiconductors as the most crowded trade. That was a record for the survey. In the same month, a fund named Situational Awareness fell from about 45 billion dollars in assets to roughly 10 billion, losing around two thirds of its value and selling its leveraged public positions to Citadel at a discount.

By August the survey read 53 percent.

The thesis was not wrong. Data centres are being built, chips are being bought, power is being contracted. The fund was destroyed by the number of other people who agreed with it, and that is a distinction worth a great deal to anyone running a business rather than a portfolio.

Being right is not a position

Reporting on the fund describes a strategy of owning the suppliers of the AI build-out, chips, data centres, power, while betting against software companies expected to be disrupted by it. Filings as of 31 March showed stakes in Nebius, Bloom Energy, SanDisk, CoreWeave, SharonAI and IREN. Leverage was reported as high as 400 percent.

Every part of that reasoning is defensible and most of it has been vindicated by events. What it lacked was any allowance for the possibility that the reasoning was widely available. When 82 percent of surveyed managers hold a version of your position, you no longer own an insight. You own an exit queue, and the price is set by whoever needs to leave first.

The leverage then converts a drawdown into a liquidation. You are not sold out because you were wrong. You are sold out because the person next to you was margin-called and their selling moved your marks.

Where operators carry the same shape without noticing

Nobody reading this runs 400 percent leverage. The structure still shows up, because concentration is not a financial-markets phenomenon, it is what happens when a lot of independent actors respond correctly to the same information.

Count your own. Traffic: what share of your leads comes from one platform, and what happens to your month if its algorithm changes. Vendors: how many of your systems would be affected by one supplier's outage, price rise or acquisition, and could you name them without opening a spreadsheet. Tactic: if every competitor in your market has adopted the same offer, the same channel and the same script, the return on it has already been competed away and you are paying the new higher price for the old lower yield. People: how many processes have exactly one person who knows how they work.

Each of those was, at some point, a correct decision. That is precisely the problem. Concentration is the residue of good decisions repeated, which is why it accumulates quietly and why nobody's dashboard flags it.

The specific error to avoid

The lesson is not to be contrarian. Contrarianism as a discipline is just being wrong on a different schedule.

The lesson is that a position's crowding is a separate variable from its correctness, and it has to be tracked separately. Two questions do most of the work.

How many people are doing this, and how would I know? For an operator this is genuinely observable. If every dealer in your market is running the same August event, if your entire competitive set is bidding the same keywords, if the vendor's case studies all describe your exact configuration, you have the answer. The crowding is visible in a way it never is inside a portfolio.

What does my downside look like if I am right? Not if you are wrong. If you are right, and so is everyone else, at the same time. That is the scenario the fund did not price, and it is the scenario in which a correct decision produces a bad outcome, which is the only kind of bad outcome that surprises people.

The August number is the useful one

The crowding reading fell from 82 to 53 in a single month. Nothing about the underlying industry changed in four weeks. Data centre construction did not pause. What changed was positioning, and it changed violently, which tells you that a very large share of that 82 percent was not conviction at all. It was momentum wearing conviction's clothes.

Apply that to your own market. When a tactic is universal, some of the people running it believe in it and some are running it because everyone else is, and you cannot tell the two apart until costs rise. When they do, the second group leaves quickly and all at once, which is either your opportunity or your problem depending on which group you were in.

What this does not say

It does not say the AI build-out is a bubble, and the evidence here does not support that claim. A crowded trade and a mistaken one are different things, and the whole point is that they were conflated in one direction on the way up and are about to be conflated in the other direction on the way down.

Situational Awareness has reportedly returned with about 400 million dollars. Somebody still believes the thesis, and they may well be right. The fund that died was not refuted. It was outnumbered, at 400 percent leverage, in the month the survey hit a record.

Being early is indistinguishable from being wrong. Being crowded is indistinguishable from being right, until it is not.

Sources

Related Reading