One key part of the AI trade has decoupled from the rest. Why that is, and whether it can come back

The Industrials Select Sector SPDR Fund (XLI) is up 8.8% year-to-date but has fallen nearly 3% since April 14, sharply lagging the VanEck Semiconductor ETF's (SMH) 64% YTD gain and 33% rise over the same period. Industrials trade at 23x forward earnings versus 19x for the S&P 500, while defense weakness, data-center project scrutiny and record fund outflows have pressured key AI-exposed names including Caterpillar (-23% over three months), GE Vernova (-12%) and GE Aerospace (-13.7%). DataTrek and Citi remain constructive on a potential recovery, contingent on Q3 backlog and guidance strength, with analyst targets implying more than 36% upside for Boeing and C.H. Robinson and more than 62% for Axon.

Analysis

The opportunity is not a broad industrials beta call; it is a dispersion setup. XLI still embeds a premium valuation despite a growing split between short-cycle, backlog-backed operators and businesses whose earnings depend on incremental data-center power construction or defense-budget/geopolitical assumptions. A Q3 guide that merely confirms demand should be insufficient for multiple expansion; the market needs evidence of order conversion, pricing durability, and cash-flow delivery rather than backlog growth alone.

CAT and GEV have the greatest near-term sensitivity to a pause in hyperscaler project starts, permitting delays, and higher financing costs, making them vulnerable if management tempers 2027 revenue or margin assumptions. By contrast, HWM's commercial-aerospace aftermarket exposure offers a more durable mix of replacement demand, constrained capacity and pricing, while AXON is driven more by municipal procurement cycles and recurring software attach than capex. BA remains an event-driven recovery trade, but its upside requires operational normalization and balance-sheet repair, not simply a sector rebound.

The key catalyst window is the next 1-3 months: earnings commentary can force a sharp unwind of crowded underweights if book-to-bill and free-cash-flow guidance hold. The 6-18 month bull case depends on electrical-grid and power-equipment bottlenecks persisting after political scrutiny fades; the contrarian risk is that data-center developers solve power constraints through project deferrals rather than higher equipment spend. A sustained rise in long-end yields, downward revisions to hyperscaler capex, or weaker orders at GEV/CAT would falsify an AI-industrials re-rating thesis.

Key Decisions for Investors

• Favor a 3-6 month pair trade: long HWM / short CAT in equal dollar amounts. HWM has superior aftermarket and pricing insulation, while CAT carries greater downside to delayed power-generation and construction orders; reassess if CAT raises next-year power-systems guidance or HWM commercial-aerospace aftermarket growth decelerates materially.
• Do not add broad XLI exposure before Q3 reports. Establish only after aggregate industrial order commentary and cash-flow guidance demonstrate that backlog is converting; use XLI as the liquid vehicle if earnings revisions stabilize, with a 6-12 month horizon rather than a pre-earnings momentum entry.
• Treat GEV as an earnings-confirmation watch item rather than a dip-buy. A long position is justified only if management sustains margin and order expectations while disclosing credible power-project timing; a cut in backlog conversion or working-capital outlook would likely create outsized multiple compression.
• Avoid using BA as a core industrials proxy despite analyst-target upside. Any long should be sized as a 6-18 month idiosyncratic recovery position and paired with strict monitoring of delivery cadence, free cash flow and net-debt reduction; missed operational milestones outweigh a favorable sector tape.
• For defensive AI-linked exposure, prefer selective AXON over capex-sensitive industrial suppliers on a 6-12 month horizon, but require continued recurring-revenue growth and government-contract conversion. Procurement delays or decelerating software attach would invalidate the relative-defensive thesis.