MEP craft labor limits U.S. annual incremental data center construction capacity to about 35 GW

Overview

The report estimates the data center construction capacity that can be supported based on the supply, hiring pace, qualification requirements, and geographic distribution of U.S. mechanical, electrical, and plumbing craft labor. The conclusion is that labor does not completely block consensus market expectations, but it will limit the industry from breaking through annual incremental construction of around 35 GW and force construction models to shift toward modularization.

Core views

The nominal size of the U.S. MEP labor force is about 1.8 million, but construction workers number about 1.4 million, narrowing further to only about 790,000 after focusing on nonresidential construction, and this still requires deductions for skill mismatches, geographic inaccessibility, and labor demand from other construction projects. If the peak hiring pace of 2023–2025 continues, annual incremental data center construction capacity can rise from about 12 GW in 2026 to about 35 GW in 2030; if it falls back to the 20-year average hiring pace, 2030 capacity would be only about 25 GW. 35 GW is sufficient to support market consensus expectations for 25–35 GW of annual incremental capacity by 2030, but it is below the roughly 40 GW needed to justify medium-sized power generation equipment capacity expansion, so CAT and CMI face relatively high overbuild risk. Breaking through the ceiling mainly depends on modularization and productivity improvements, which benefits equipment manufacturers and engineering contractors with higher degrees of vertical integration.

Analysis framework

The report first uses industry interviews to estimate the number of mechanical, electrical, and plumbing craft workers and labor hours required per 100 MW of data center capacity, then deducts non-construction, residential, skill-unqualified, geographically inaccessible, and labor occupied by other construction activities from the total U.S. MEP labor force. Finally, it converts the incremental available craft labor under different hiring scenarios into annual data center construction capacity and compares it with market consensus expectations, power generation equipment capacity expansion, and the demand implied by GPU capacity expansion.

Methodology notes

• Supply constraint analysisMEP labor-constrained capacity model

  Derive the annual construction ceiling by dividing the number of MEP craft workers that can be newly added and used for data center construction by the labor requirement per unit of capacity.

  The model separately estimates the capacity contribution of mechanical, electrical, and plumbing trades, with the overall ceiling determined by the most constrained mechanical trade; it also deducts incremental labor consumed by other nonresidential construction activities.

• Scenario analysisPeak hiring and long-term average hiring scenarios

  Compare the peak hiring pace of 2023–2025 with the hiring pace in positive-growth years over the 20-year average.

  The peak scenario corresponds to an increase of about 6 GW of construction capacity per year and reaching about 35 GW in 2030; the long-term average scenario corresponds to an increase of about 2 GW per year and about 25 GW in 2030.

• Bottom-up estimateLabor economics per unit of capacity

  Estimate the number of workers, labor hours, and labor costs required per MW for each specialized trade based on industry interviews.

  Each 100 MW project requires about 30 mechanical craft workers, 48 to 50 electricians, and 10 plumbers, or about 88 to 90 on-site MEP craft workers in total.

• Addressable supply analysisSkill and geographic matching screen

  Exclude from the nominal labor pool workers who are not qualified for complex data center construction or are not near project locations.

  Data centers require specialized skills such as high-voltage work and large-diameter welding, while regions with about 70% of projects have only about 30% of MEP labor, making actual available supply significantly lower than the nominal size.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• ETN, SU, VRT, LGN

  Major beneficiaries of modular construction and vertical integration trends, and the report maintains Outperform ratings.

  Strengths

  They have equipment manufacturing, system integration, or modular delivery capabilities and can capture value shifting from the site to the factory.

  Weaknesses

  Growth still depends on data center projects being built on schedule, and the labor ceiling will limit the room for industry volumes to exceed expectations.

  Comparison

  Compared with pure on-site construction or single-product equipment suppliers, vertical integration capabilities are more favorable for capturing the modular profit pool.

  Risks

  Project delays, cooling capital expenditure, lower-than-expected modular penetration, and valuation pullback.

• PWR

  Listed as a potential beneficiary of modularization and construction model transformation, but its rating is maintained at Market-Perform.

  Strengths

  It has large-scale engineering contracting and complex infrastructure construction capabilities and can participate in data center power infrastructure expansion.

  Weaknesses

  On-site construction operations remain constrained by craft labor supply, and the degree of benefit may be weaker than that of highly factory-based equipment manufacturers.

  Comparison

  The assessment of benefit from industry trends is positive, but the stock rating is below most of the modular beneficiaries cited in the report.

  Risks

  Rising labor costs, construction schedule delays, changes in project mix, and valuation risk.

• FIX, EME

  Listed in the report as engineering contractors benefiting from the modularization trend, but not covered by Bernstein.

  Strengths

  Specialized mechanical and electrical construction capabilities position them to benefit from data center infrastructure investment.

  Weaknesses

  The report does not provide formal ratings, target prices, or complete valuation arguments.

  Comparison

  Compared with covered beneficiaries, the investment conclusion has less complete evidence.

  Risks

  Uncovered status, labor tightness, execution risk, and project cycle volatility.

• CAT, CMI

  Medium-sized power generation equipment expansion faces potential overbuild risk.

  Strengths

  They have mature medium-sized power generation equipment products and supply capabilities and can benefit from backup or on-site generation demand from data centers.

  Weaknesses

  The rationale for the related capacity expansion requires U.S. annual data center construction to reach about 40 GW by 2030, above the report’s 35 GW base-case ceiling.

  Comparison

  Compared with modular electrical equipment and engineering contractors, the two are more sensitive to construction demand assumptions exceeding labor constraints.

  Risks

  Capacity utilization below expectations, inventory increases, price competition, and downward revisions to data center orders.

• DLR, EQIX, CSQR

  Data center operations-related names for which the report maintains Outperform ratings.

  Strengths

  Direct exposure to growth in data center capacity demand, with the industry base case still supporting significant expansion through 2030.

  Weaknesses

  Labor shortages may extend delivery timelines and raise construction costs.

  Comparison

  The report does not list them as the primary beneficiaries of the shift in the modular profit pool; the main value drivers remain capacity demand and project execution.

  Risks

  Project delays, financing costs, rising construction costs, and new supply below plan.

Key data

• Base-case construction ceiling in 2030About 35 GW/yearAssumes continuation of the 2023–2025 peak MEP craft labor hiring pace.
• Annual incremental construction scale in 2026About 12 GW/yearThe growth starting point estimated in the report.
• Annual increment in the base caseAbout 6 GW/yearCorresponds to a compound growth rate of about 30% from 2026 to 2030.
• Long-term average hiring scenarioAbout 25 GW/year in 2030Annual construction capacity increases by about 2 GW, corresponding to a compound growth rate of about 20%.
• Nominal MEP labor force sizeAbout 1.75 million to 1.8 million peopleOf which about 1.37 million to 1.4 million are employed in construction.
• Nonresidential MEP labor forceAbout 790,000 peopleFurther deductions are still needed for qualification, geography, and other project labor constraints.
• Geographic mismatch between projects and labor70% of projects correspond to about 30% of MEP laborLabor is not concentrated in the areas where future data center construction will be most intensive.
• On-site labor for a 100 MW projectAbout 88 to 90 MEP craft workersAbout 30 mechanical craft workers, 48 to 50 electricians, and 10 plumbers, excluding supervisors and management personnel.
• Peak annual hiring scaleAbout 15,000 mechanical, about 30,000 electrical, and about 15,000 plumbing workersThe report believes this pace is closer to the hiring ceiling rather than a sustainable midpoint.
• Theoretical ceiling under full labor reallocationAbout 85 GW/yearAssumes all available U.S. MEP craft labor shifts to data center construction, which is not achievable in reality.
• Construction scale needed for medium-sized power generation equipment expansionAbout 40 GW/yearAbove the 35 GW base-case ceiling, creating potential overbuild risk for CAT and CMI.
• Selected maintained target pricesLGN $103; ETN $534; HUBB $584; PWR $748LGN, ETN, and HUBB maintain Outperform ratings; PWR maintains a Market-Perform rating.

Impact & implications

Labor constraints mean data center equipment demand can still maintain strong growth, but cannot be extrapolated indefinitely. Medium-sized power generation equipment, GPUs, and related supply chains may face insufficient demand realization if they expand capacity based on U.S. annual construction scale exceeding 35 GW. At the same time, modularization, prefabrication, and factory integration can reduce on-site labor hours, shifting the profit pool from construction sites to manufacturing and enhancing the relative value of vertically integrated OEMs and large engineering contractors.

Risks

• MEP hiring has slowed from the 2023 peak, so the base case may overestimate sustainable hiring capacity.
• The share of MEP craft workers in the U.S. labor force is at a more-than-20-year high, leaving limited room to further expand supply.
• Training depends on existing skilled craft workers, skill formation cycles are long, and the labor force cannot be replicated quickly in the short term.
• Geographic mismatch between data center projects and labor may cause localized schedule delays and wage increases.
• A rebound in other nonresidential construction activity would crowd out MEP labor and depress the actual data center construction pace.
• If data center construction does not reach about 40 GW/year, medium-sized power generation equipment expansion may result in surplus capacity.
• If modular productivity improvements are slower than expected, the industry will struggle to break through the construction ceiling of around 35 GW.

What to watch

• Whether annual hiring of mechanical, electrical, and plumbing craft workers returns close to the 2023–2025 peak.
• Productivity and training pace of mechanical trades as the tightest constraint.
• Wages, worker migration, and project delays in major U.S. data center states.
• Penetration of modularization, prefabrication, and factory integration in data center projects.
• Whether data center construction’s share of U.S. nonresidential construction labor continues to rise.
• Capacity, orders, inventories, and cancellations at medium-sized power generation equipment manufacturers such as CAT and CMI.
• Whether actual annual incremental data center capacity can enter the consensus expectation range of 25–35 GW.