How fast can the U.S. build data centers amid labour concerns?

Bernstein projects U.S. data-center construction could rise from 12 GW/yr in 2026 to a labour-limited ~35 GW by 2030 (about +6 GW/yr; ~30% CAGR), but specialised MEP hiring constraints imply only ~25 GW/yr at longer-term hiring rates. The gap vs Bernstein’s ~40 GW requirement for mid-scale power-generator manufacturing raises potential overbuild risk for power-equipment suppliers like Caterpillar and Cummins. Analyst commentary also links the broader S&P 500 weekly rise to reduced rate-hike expectations.

Analysis

The key market mechanism is not “slower data center growth” but a shift in who captures the margin pool. If site labor remains the bottleneck, the value migrates away from commoditized equipment volume and toward vertically integrated electrical, cooling, and field-services firms that can package labor, design, and hardware into a constrained supply offering. That argues for relative outperformance in ETN, VRT, PWR, EME, and FIX versus pure-play power-equipment names where unit growth assumptions look more fragile.

CAT and CMI are the clearest overbuild-risk names because generator demand is being modeled off an installation rate that may be unattainable without a step-up in specialized labor. The second-order effect is order deferral, not necessarily cancellation: customers can stretch lead times, re-spec to smaller modules, or shift toward distributed power architectures, which compresses near-term backlog quality and can force multiple contraction before any revenue disappointment shows up. This is a 1-3 month catalyst risk as the market starts interrogating backlog conversion rates and 2026-2030 capacity assumptions.

The contrarian read is that the labor ceiling is bullish for the “pick-and-shovel” integrators because scarcity raises pricing power and increases the appeal of modular/off-site construction. If modular adoption accelerates, the winners are those with software, controls, and integration capabilities rather than just shipped metal. The structural implication over 6-18 months is a widening valuation gap between constrained-capacity beneficiaries and equipment vendors exposed to a false sense of TAM expansion.

A key falsifier is evidence that recruitment or modular build techniques materially improve throughput: if quarterly backlog conversion and order intake in CAT/CMI hold up through the next two earnings cycles, the overbuild thesis weakens quickly. Conversely, any commentary from hyperscalers on delayed power delivery or re-phasing of capex would be an immediate negative for CAT/CMI and a relative positive for ETN/VRT/PWR.

Key Decisions for Investors

• Pair trade: long ETN or VRT / short CAT for 1-3 months. Thesis: labor scarcity and modularization favor integrated power-management franchises over generator OEM volume. Risk: if hyperscaler capex accelerates faster than labor supply, CAT can outperform on backlog leverage.
• Short CMI on strength into earnings or conference-season commentary. Use a 3-6 month horizon and size modestly: the risk/reward hinges on whether generator lead times convert to realized shipments or just headline demand. Falsifier: sustained backlog growth and unchanged delivery schedules.
• Long PWR or EME as relative winners in a labor-constrained buildout. These names monetize scarce field execution and should see pricing power before pure equipment suppliers do. Best entry is on any selloff tied to generalized capex fears rather than company-specific execution misses.
• Avoid chasing CAT/CMI on data-center hype until there is proof of capacity expansion in the specialized labor pool or modular offset. If you need exposure, prefer a call spread rather than outright equity to limit multiple-compression risk.