Industrials Became the AI Trade — and Most 13F Analysis Is Still Looking at Chips

For two years, "AI exposure" in a 13F meant a short list of tickers: Nvidia, Broadcom, Micron, the semi-cap complex, a hyperscaler or two. That shorthand worked until it didn't. Through 2026 the semiconductor names have been the source of the year's sharpest institutional pain, while a sector that reads as deeply unglamorous on a holdings screen has quietly delivered the returns the AI thesis promised.

Industrials are up roughly 31.5% over the trailing twelve months — the largest gain of any S&P sector over that stretch — with the sector ETF posting something close to a 20% calendar-year gain. That isn't a defensive rotation story. It's the same trade, bought one layer down the supply chain.

The two names doing the work

The concentration inside the industrials rally is worth stating plainly, because it changes how you read a fund's sector weight.

Caterpillar is the largest holding in the sector ETF at roughly 8.4% of the basket, and it has returned something in the neighborhood of 84% — driven by data center construction and by global miners upgrading equipment into higher commodity prices. GE Vernova, at roughly 5.2% of the basket, is up close to 78% on the electrical hardware that data centers physically cannot be built without: turbines, switchgear, transformers, grid equipment.

Between them, two positions account for a disproportionate share of what looks, on a sector line, like broad industrial strength. A manager who owns industrials but doesn't own those two has not had the industrials year the sector chart implies. This is exactly the kind of distinction that gets flattened when analysis stops at sector allocation, and why it's worth working through individual investor pages rather than reading sector totals.

Aerospace is a separate trade wearing the same label

GE Aerospace belongs in the same sector bucket and almost none of the same reasoning. Its case has nothing to do with AI capex and everything to do with a supply constraint that has now persisted long enough to be structural: Boeing and Airbus production remains below demand, which keeps older fleets flying, which drives engine flight hours, shop visits and spare parts consumption.

The numbers underneath that are unusually durable for an industrial — roughly 50,000 engines installed, engines on something like three of every four commercial flights, and a backlog in the region of $190 billion. As of the Q1 filings, 119 hedge fund portfolios held the name, up from 117 the prior quarter.

That ownership number is the interesting part. A two-fund net increase on a stock this size is not conviction — it's drift. Institutions were already there and didn't meaningfully add.

What the filings can and cannot settle

The most recent visible book is Q2, filed August 14 against June 30 positions. Q3 filings don't land until November 16. So the entire September repricing — the 10-year touching 5.01%, the accompanying damage to long-duration growth — sits in a window no 13F has reported on yet.

That matters here more than usual, because industrials sit on both sides of the rate trade. The data center capex leg is a growth story funded by hyperscaler balance sheets and largely indifferent to the ten-year. The aerospace leg is a services annuity. Neither behaves like the semiconductor complex did in Q3, which is part of why the sector held up while chips did not.

What you can check now is whether the Q2 books showed institutions already rotating into industrials before the third-quarter move, or whether they will appear in the November filings having arrived after the fact. The sector flow data distinguishes those two cases; a fund's Q3 sector weight on its own does not.

The reading that actually helps

Three things are worth holding onto.

First, treat "AI exposure" as a supply chain rather than a ticker list. Power generation, grid equipment, cooling, heavy equipment and construction materials are all downstream of the same capex, with different cyclicality and very different valuations. A portfolio can be heavily AI-exposed without holding a single semiconductor, and the reverse is also true.

Second, check whether multiple independent managers converged on the same industrial names or whether one large holder is doing the work in an aggregate number. The overlap tool is built for that question, and the answer is frequently less impressive than the headline.

Third, remember that industrials are the sector where 13F blind spots bite hardest. Much of the real AI-infrastructure exposure lives in private construction firms, non-US equipment makers, and the debt financing the buildout — none of it reportable. And the macro-sensitive funds that position around rates rather than narratives will have moved first; the macro consensus view aggregates that cohort specifically.

The AI trade did not end in 2026. It changed address, and the filings are a slow way to find the new one.

---

Data sourced from public SEC 13F filings. Educational research only — not investment advice.