This Earnings Season Confirmed the AI Power Bottleneck Is Real. Here Are the Industrial Winners Hiding in Plain Sight.

Eaton, Vertiv, and Cummins are highlighted as “picks-and-shovels” beneficiaries of AI-driven data center buildouts, with Eaton announcing an Aug. 17 partnership with Trane to develop Nvidia DSX AI-factory reference designs and higher-power systems for AI data centers. Eaton posted Q2 sales up 21% to $8.53B, with record adjusted EPS of $3.15 (+12% YoY), and guided full-year organic revenue growth of 11%–13% and adjusted EPS of $13.40–$13.60; however, $9.55B Boyd Thermal and $1.53B Ultra PCS acquisitions increased long-term debt 112% to $18.5B. Vertiv reported Q2 revenue of $3.27B (+24% YoY), EPS up 53% to $1.27 (adjusted diluted EPS +60% to $1.52), and raised full-year net sales to $14B (+31% at the midpoint); Cummins reported Q2 revenue $9.5B (+9.4%) and EPS $6.73 (+4.6%) while projecting FY sales growth of 10%–13%.

Analysis

This is less an AI-equity call than a bottleneck call: the incremental economics of the buildout are migrating from compute to power density, thermal management, and redundancy. That favors ETN and VRT first because their content scales with every new rack and every higher-watt deployment, while TT sits in the adjacent integrated-cooling lane. CMI is a secondary beneficiary, but its demand is more lumpy and procurement-driven; it participates when customers overbuild backup rather than when they expand capacity.

Near term, the risk is crowding. The market can bid these names ahead of true margin realization because backlog is visible sooner than install cycles, so any pause in hyperscaler capex will hit VRT first and ETN second. A softer signal would be improved utility availability or faster-than-expected server efficiency, which reduces the urgency for premium cooling and on-site generation; that would leave CMI most vulnerable to relative underperformance.

Over 6-18 months, the thesis is that data-center infrastructure starts trading more like regulated utility-adjacent assets than cyclical industrials, supporting persistently better pricing and mix. The contrarian miss is valuation dispersion: VRT has already absorbed a lot of the story, while ETN still has the cleanest combination of backlog leverage and balance-sheet optionality. If credit spreads widen or hyperscalers trim capex guidance, the market will de-rate the whole basket quickly, but ETN should hold up better than the more momentum-owned names.

Key Decisions for Investors

• Go long ETN on a 5-8% pullback or via a 6-12 month call spread; best risk/reward in the group because it has the broadest exposure to grid interconnect and data-center electrification, with a cleaner path to converting backlog into earnings.
• Do not chase VRT after its run; add only on weakness or through limited-risk upside structures, since it is the most crowded expression of the theme and most sensitive to any slowdown in hyperscaler capex commentary.
• Express the relative-value view as long ETN / short CMI for 1-3 months if you want to hedge market beta; ETN should outperform if the market keeps rewarding higher-power AI deployment, while CMI is more exposed to lumpy backup-power orders.
• Use TT as a watchlist long rather than a full-size buy; if integrated cooling wins share in upcoming data-center design wins, TT becomes the lower-volatility way to participate in the same bottleneck trade.
• Set a thesis breaker on any renewed capex moderation from AMZN/MSFT/NVDA or a widening in industrial credit spreads; those are the first indicators that the infrastructure trade is getting ahead of itself.