medium convictionactive · updated 2026-06-06T00:00:00.000Z
AI capex order books → non-tech industrials trade like semis (SMH-correlation screen)
The AI-capex boom (~$700B/yr, ~7% of GDP) has turned non-tech industrials' order books into AI-capex order books — so ~15 non-tech S&P names (~$2T mcap) now correlate ≥0.5 with the SMH semis ETF while GICS still files them as "industrials." The correlation is both a *screen* for hidden AI-capex beneficiaries (VRT, ETN, CAT, CMI, GEV, GNRC) and a shared *downside* if capex rolls over.
The chain
1
AI capex is now a macro-scale forcing function: ~$700B/yr hyperscaler capex, ~7% of GDP — past the 2005 housing-investment peak and the 1990s dot-com/telecom peaks.
From 2026-06-05-podcast-the-compound-and-friends-neil-dutta-skanda-on-why-the-recession-signals: Skanda Amarnath (Employ America): "we've ripped past 7%... past the peak for housing investment in the 2000s... [past] the peak of the 90s tech boom. The dot com boom."
From 2026-06-05-podcast-the-compound-and-friends-neil-dutta-skanda-on-why-the-recession-signals: Josh Brown: "these hyperscalers and others joining the party are spending $700 billion of capex."
2
That capex flows into physical data-center inputs — power (gas turbines, gensets, backup power), cooling, and electrical gear — so non-tech industrials' order books become AI-capex order books (CAT engines/gensets, VRT cooling+power, ETN electrical, GEV turbines, GNRC backup power, CMI engines).
From 2026-06-05-podcast-the-compound-and-friends-neil-dutta-skanda-on-why-the-recession-signals (research by Neil Dutta / RenMac, read aloud by Josh Brown): "Caterpillar sells backup generator sets and engines into hyperscaler data centers. Vertiv sells cooling and power management. [Eaton] sells electrical components and GE Vernova sells gas turbines for data center power."
From 2026-06-05-podcast-the-compound-and-friends-neil-dutta-skanda-on-why-the-recession-signals: "Generac is doing better now because of the data center build out to build backup power for these facilities."
3
Empirically, a 1-yr daily-return correlation vs the SMH semis ETF flags 15 non-tech S&P 500 names (~$2T mcap) at ≥0.5; 12 are industrials. They "trade like semis" because the same AI-capex demand drives their order books.
From 2026-06-05-podcast-the-compound-and-friends-neil-dutta-skanda-on-why-the-recession-signals (research by Neil Dutta / RenMac, read aloud by Josh Brown): "we ran a one year daily return correlation between every S&P 500 name and the Semiconductor ETF SMH. 15 non-tech S&P 500 companies collectively worth $2 trillion in market cap now move with semis at correlations of 0.5 or higher. Twelve of these 15 are industrials... They trade like semis because their order books have become AI CapEx order books."
4
GICS sector classification hasn't caught up with the economic exposure — the AI-capex picks-and-shovels hide in the "industrials" bucket — which also surfaces as equal-weight breadth (S&P 493 +12.6% YTD vs Mag7 +7.3%).
From 2026-06-05-podcast-the-compound-and-friends-neil-dutta-skanda-on-why-the-recession-signals: "The gics sector classification has not caught up with the economic exposure."
From 2026-06-05-podcast-the-compound-and-friends-neil-dutta-skanda-on-why-the-recession-signals: "the Mag 7 are up 7.3% year to date... but the 493 are up 12.6%"; Batnick: "this is going to support a lot of the equal weighted indexes because it's such a profound effect on the broader economy."
What would falsify this
- Step 3: the named industrials' SMH-correlation falls below 0.5 in a risk-off tape without their order books changing — i.e., the correlation was risk-on beta, not AI-capex exposure.
- Step 2: a named beneficiary (CAT, VRT, GEV) discloses data-center/AI-capex is a small or shrinking share of its order book / backlog.
- Whole thesis: AI capex decelerates — the signal to watch (per Amarnath) is non-tech companies' willingness to keep funding AI as an *expense* needing ROI (vs hyperscalers funding moat); when that ROI question bites (à la Y2K front-loading into 2000), the basket de-rates with semis. Watch credit spreads.
Contradictions / tensions
- Single-source articulation (one podcast, RenMac's framing), though the underlying capex scale is corroborated across the wiki's AI-capex cluster.
- Correlation cuts both ways: From 2026-06-05-podcast-the-compound-and-friends-neil-dutta-skanda-on-why-the-recession-signals: Skanda Amarnath: "correlations do go to one in both directions" — in a risk-on tape, beta inflates cross-correlations, so part of the ≥0.5 may be risk-on beta rather than fundamental order-book exposure.
- "GICS lag → mispricing" is interpretation, not demonstrated mispricing — these names may already be priced for AI-capex exposure.
Implications
- Tradeable basket of correlation-identified AI-capex industrials: **VRT** (Vertiv), **ETN** (Eaton), **CAT** (Caterpillar), **CMI** (Cummins), **HUBB** (Hubbell), **FIX** (Comfort Systems), **GEV** (GE Vernova), **GNRC** (Generac). Diversified picks-and-shovels exposure to the buildout without paying semis multiples.
- The SMH-correlation screen is itself a repeatable tool to surface hidden AI-capex beneficiaries before GICS/consensus re-tag them.
- Supports equal-weight (RSP) vs cap-weight as breadth broadens beyond the Mag7.
- Overlaps existing power/cooling chains: ai-power-gap-to-genset-bridge-power (gensets), dc-rack-density-to-800vdc-white-space-tam (ETN/power), ai-capex-to-power-and-materials-cascade.
Companies
Concepts
noneOpen questions