Buy AI-Led EME, PWR and CAT for Long-Term to Gain From Recent Softness

Zacks identifies EMCOR, Quanta Services and Caterpillar as AI data-center infrastructure beneficiaries, with all three rated Zacks Rank #1 and trading 21.8%-25.6% below their 2026 highs despite year-to-date gains of 19.2%-46.1%. EMCOR's RPO reached a record $17.14B (+43.9% YoY), while Quanta's backlog rose 49% YoY to $53.4B, underscoring demand for power, grid and data-center construction. Caterpillar expects power-generation growth from AI and cloud buildouts, plans to nearly triple reciprocating-engine capacity versus 2024 by 2030, and targets more than tripling Power Generation sales.

Analysis

The investable distinction is where the AI capex bottleneck sits. EME monetizes the final, mission-critical fit-out phase and should have the highest near-term operating leverage when projects move from shell completion to energization; PWR captures the upstream transmission, substation and interconnection spend, where utility rate-base support can make demand more durable but revenue recognition is longer-cycle. CAT is less a direct data-center contractor than a beneficiary of constrained grid access: onsite gas generation becomes the bridge solution when utility interconnection queues delay campus commissioning.

The key second-order risk is that hyperscaler capex can remain strong while these stocks disappoint because projects slip at the power-permit, turbine, transformer, or utility-interconnection stage. That would defer EME's high-margin deployment work and raise working-capital intensity at PWR, while CAT could benefit initially from temporary generation demand but face reversal if grid connection timing improves or gas-generator economics weaken. Over the next 1-3 months, the relevant catalyst is not another broad AI spending announcement but evidence of conversion from awarded backlog into revenue, margin, and cash flow; over 6-18 months, labor availability and fixed-price contract discipline matter more than headline backlog.

Consensus likely treats the recent pullback as a uniform AI-infrastructure buying opportunity. It is more likely an opportunity to separate execution quality: EME offers the cleanest earnings sensitivity but is vulnerable to multiple compression if its mix normalizes; PWR has the strongest structural grid thesis but has already embedded substantial growth expectations; CAT provides a more diversified, lower-purity power-demand expression. Brokerage target dispersion should not be used as valuation support without updated EV/EBITDA, free-cash-flow conversion, and data-center revenue concentration data.

Key Decisions for Investors

• Initiate a 3-6 month long EME / short PWR pair on a beta-neutral basis if EME underperforms PWR by an additional 5-7%. Thesis: final-stage electrical/mechanical deployment should convert into earnings faster than utility-program revenue, while the short leg hedges AI-infrastructure and industrial-factor risk. Exit if EME's next reported operating margin or cash conversion misses guidance, which would indicate project mix or labor pressure.
• Maintain PWR as the preferred 12-18 month structural grid exposure, but add only after confirming backlog conversion and electric-infrastructure margin guidance at the next earnings release. A guidance cut tied to permitting, customer funding, or working-capital needs would falsify the thesis; absent that confirmation, treat the name as a watch item rather than chase a rebound.
• Use CAT as a 6-12 month tactical long only if management quantifies power-generation order growth, capacity utilization, and incremental margin on the expanded engine platform. The upside is a scarcity premium for behind-the-meter generation; key downside trigger is evidence that utility interconnection timelines are shortening or that gas-engine lead times/pricing are normalizing.
• Avoid using NVDA as a clean hedge for these positions. A hyperscaler AI-capex slowdown would pressure both semiconductor and construction beneficiaries, but NVDA's valuation sensitivity and product-cycle drivers are materially different; use XLI or an equal-weight industrial basket for broad cyclicality hedging instead.