Divergence Within the AI Trade: Industrials Underperform Chip Stocks by Over 55 Percentage Points
nashnova research
Industrial ETF XLI and semiconductor ETF SMH started 2026 in lockstep as twin AI beneficiaries — by September 25 their year-to-date returns diverged by more than 55 percentage points, as valuation, defense weakness, and data-center headwinds weigh on industrials while Q3 earnings will determine whether the gap narrows or widens.
How did a 55-point gap open up?
At the start of 2026, XLI and SMH tracked almost identically — both seen as core AI plays.
The break came around April 14: XLI has since fallen nearly 3%, while SMH surged 33%.
This means → the market did not abandon the AI thesis — it ran a selection *within* the AI trade, rotating capital from industrials into chips.
Why are industrials lagging? What are the three headwinds?
Stretched valuation: industrials trade at a 23× forward P/E — above the S&P 500's 19× and their own ten-year average of roughly 20×. At the single-stock level, Caterpillar sits at 28×, while Nvidia — the AI bellwether — trades at just 18×.
Defense pullback: the Aerospace & Defense ETF (ITA) dropped 10% over three months as geopolitical uncertainty — including Trump's suggestion of a post-midterm Iran ceasefire — kept traders cautious.
Data-center construction stalls: energy costs, power demand, and community impact face tighter scrutiny. Caterpillar fell 23% in three months; GE Vernova and GE Aerospace lost 12% and 13.7% respectively.
In plain terms = industrials are expensive *and* their two main growth pillars — defense and data centers — are wobbling at the same time.
What is the money flow saying?
Bank of America flagged "the largest outflows" from industrials in an early-September note; four-week average flows sank to historic lows.
The sector was also tagged as one of the most "crowded" and "expensive."
This means → it is not just a stall — institutions are actively cutting positions, and positioning is shifting from overweight to washout.
How much does the AI narrative really matter to industrials?
According to Melius Research, industrials doubled in market cap since late 2022, adding roughly $2.6 trillion.
57% of that gain came from companies tied to AI and aerospace — precisely the two areas under the most pressure now.
This reflects a real dynamic: the "AI premium" in industrials is genuine, but it also means that when AI sub-themes cool, the drawdown concentrates in exactly those names.
What comes next? What is the verification point?
DataTrek co-founder Nicholas Colas told CNBC that industrials' risk-reward ratio is "quite good" and expects the sector to match or beat the broader market ahead.
His logic: AI-linked industrials have reached a scale that can generate steadier long-term earnings → they deserve "a sustainably higher multiple."
DataTrek also expects data-center headwinds to fade after the midterm elections; long-term demand logic is intact and the current oversold level is "worth watching."
In plain terms = the key verification point is next month's Q3 earnings: are backlogs thick enough and earnings streams stable enough? Deliver, and a valuation recovery begins. Miss, and the 55-point gap keeps widening.
市场有风险，内容仅供研究参考，不构成投资建议。