Concentrated by Design

—Executive Summary

AI data-center construction is concentrating in a small set of U.S. markets — led by Northern Virginia, Dallas/West Texas, Central Ohio, Phoenix, and Atlanta — that carry the deepest electrical-labor constraints and the longest grid-interconnection backlogs. The central risk is not a national labor shortage; it is concentration: demand arriving in specific markets and specific trades faster than supply can localize.

The binding constraint becomes infrastructure delivery capacity — skilled electrical and mechanical labor and power infrastructure delivered as a single, coupled system. A campus stalls equally on an unavailable medium-voltage electrician or a delayed substation; the executive question is not whether either constraint exists, but whether a specific market can resolve both on the committed timeline.

The AlphaHire verified pipeline of 49 projects, ~6.8 GW, and ~$31.3B in planned investment confirms that capital is already committed to markets where delivery capacity is the limiting variable. Version 2.0 expands the evidence base to 10 graded findings, adds an explicit labor-demand framework calibrated to public staffing observations, introduces market profiles by type, and provides a three-scenario planning frame to bound the thesis across high, base, and conservative buildout paths.

Headline findings

1. F1Demand is geographically concentrated, not nationalHigh

   More than 35 GW is under construction across North America, with 64% in frontier markets — West Texas, Tennessee, Wisconsin, and Central Ohio. Tier-1 hubs (Northern Virginia, Dallas, Phoenix, Atlanta, Chicago) are described by Georgetown's geography research as effectively sold out, and CBRE's H2 2025 data confirm record absorption, near-zero vacancy, and heavy pre-leasing. The 92% pre-commitment rate means demand is locked-in before delivery capacity is confirmed.

   Implication. Execution risk is set by contractor and grid capacity in specific markets, not national totals. Workforce planning anchored to national headcount will systematically understate local exposure.

   Sources: Pew Research · AlphaHire pipeline
2. F2The exposed labor is trade-specific, led by electricalHigh

   Electrical systems represent 45–70% of data-center construction cost (IBEW), making electricians — and within that, medium- and high-voltage and commissioning specialists — the first and sharpest constraint. BLS records ~818,700 electricians employed (2024), projected +9% through 2034 — growth that predates AI-specific demand. HVAC engineer demand is up 67% and construction roles up 30%, with rising time-to-hire. Labor shortage in trades is already driving cost overruns, delays, and disputes on AI data-center projects.

   Implication. Measure workforce risk at the trade level — electrical, mechanical, commissioning — not as aggregate construction headcount.

   Sources: IBEW · BLS
3. F3Skilled-trade scarcity is a structural constraint, not a cyclical hiring challengeHigh

   An estimated ~81,000 annual electrician shortfall persists through 2034 (WIRED, citing BLS). Skilled-trade demand is growing ~3× faster than professional roles; for every 100 young entrants to manufacturing trades, 102 leave — negative net inflow (Randstad). An estimated ~300,000 additional workers are needed by 2030 across manufacturing, construction, and operations. A pipeline with negative net inflow cannot respond to a demand surge on a project timeline regardless of wage signals.

   Implication. Workforce availability is a durable, supply-side constraint. Design around it rather than waiting for wage signals to resolve it within a project cycle.

   Sources: BLS · Goldman Sachs
4. F4Grid delivery has become a co-equal gating factor with laborHigh

   Data centers account for ~55% of forecast U.S. load growth — about 90 GW of peak-load growth (Grid Strategies). The Federal Reserve identifies electricity infrastructure as the weak point in U.S. AI competitiveness. The WEF frames grid connectivity as the strategic bottleneck and recommends site decisions start with grid capacity and interconnection queues. FERC's December 2025 order directing PJM to permit data-center colocation at power plants signals that the constraint is being treated as systemic.

   Implication. Grid-delivery timelines belong inside the site-selection decision alongside labor. A labor-ready market can still be gated by interconnection.

   Sources: FERC
5. F5Power and workforce are converging into a single infrastructure-delivery constraintModerate

   The most consequential finding is that the two risks the market still treats separately — workforce and grid — are converging. On a hyperscale critical path, the electrical rough-in, switchgear delivery, substation energization, and commissioning crew are interdependent; a campus stalls equally on a delayed interconnection or an unavailable medium-voltage electrician. The same frontier markets carry the deepest pipelines, the tightest grid headroom, and the thinnest local trade pools.

   Implication. Adopt infrastructure delivery capacity as the unit of site-selection and execution risk. Score markets on the joint availability of trades and power delivery.

   Sources: FERC · IBEW
6. F6Demand is front-loaded while supply elasticity is low, making the constraint durableModerate

   Construction demand is arriving now and is heavily pre-committed, while the trades pipeline responds on a multi-year apprenticeship-and-licensing cadence against an aging workforce. Negative net inflows into the trades (102 leave per 100 entrants in manufacturing). Electrician employment projected to grow only ~9% over a decade against near-term demand acceleration. The gap cannot close on a project timeline.

   Implication. Plan for rebalancing lag as a structural feature of the lead markets. Treat workforce development as co-equal to capital, not a downstream fix.

   Sources: BLS · Goldman Sachs
7. F7Rural dispersion shifts the constraint toward mobilization rather than removing itModerate

   More than 1,500 planned facilities are in development, with most construction in the South and Midwest, particularly rural areas (Pew). Facility count is dispersing without diluting capital concentration in frontier clusters. FERC's December 2025 PJM colocation order steers gigawatt-class campuses toward power-plant nodes with no construction-labor history. Per-diem, lodging capacity, travel logistics, and regional contractor relationships become first-order determinants of schedule.

   Implication. In rural and generation-adjacent markets, model mobilization and retention logistics explicitly. Metro contractor-depth assumptions will not transfer.

   Sources: Pew Research · FERC
8. F8Construction labor competes within a stacked megaproject pipeline, not in isolationModerate

   The same electricians, linemen, pipefitters, and civil trades that data centers require are demanded simultaneously by semiconductor fabrication, battery manufacturing, grid modernization, and utility-scale renewables. Data centers are ~55% of incremental load growth — one large claimant among several electrification-driven programs competing for the same trades. Cross-sector competition is under-quantified in joint regional workforce models.

   Implication. Assess trade availability net of competing megaproject demand in the same market. AI-only demand understates the true draw on the pool.

   Sources: FERC · AlphaHire pipeline
9. F9Operations labor forms a persistent, location-anchored demand layer beyond the construction peakModerate

   Workforce demand does not end when construction peaks. As more than 50 GW comes online at a 24% CAGR, facilities operations, critical maintenance, network, and controls technicians become a durable, location-anchored layer. The Hamm Institute forecasts operations employment growing steadily through 2030 as facilities come online. Operations roles overlap with construction electrical and controls skills, compounding local tightness.

   Implication. Plan for both the construction surge and the operations long tail. The same markets carry both, and they compete for overlapping skills.

   Sources: FERC · AlphaHire pipeline
10. F10Aggregate load and near-term pipeline may run below peak forecasts — a bounded thesisModerate

    Grid Strategies: utility forecasts may overestimate 2030 data-center load by ~25 GW; ~65 GW is the more likely through-2030 track; data-center load is likely ~10% of peak by 2030. FERC data show capacity under construction dipped modestly from ~6,350 MW (2024) to ~5,995 MW (end-2025). Community resistance, zoning, and moratoria are redirecting some projects. None of this overturns concentration or scarcity, but it argues against anchoring to upside-only projections.

    Implication. Frame workforce planning around high/medium/low buildout scenarios. Do not anchor capital or training commitments to upside-only projections.

    Sources: FERC · Pew Research

Strategic implications

Sources: Goldman Sachs · FERC · IBEW · BLS · Pew Research · AlphaHire pipeline — Confidence: Moderate

6.13Market profiles

Northern Virginia (saturated primary) — High exposure. The deepest data-center ecosystem in the country and simultaneously one of the most constrained. Record absorption and near-zero vacancy leave no slack; the deep local trade base is fully utilized, so incremental demand competes directly with committed work and bids up rates. Grid headroom in the PJM footprint is limited.

Dallas / West Texas (primary-to-frontier) — High exposure. A primary metro feeding a frontier corridor selected for land and power. The ERCOT large-load queue is among the most constrained in the country, and the West Texas trade base is thin relative to the committed pipeline. Labor and grid are most likely to bind together.

Central Ohio (emerging-frontier) — High exposure. A frontier market absorbing significant new capacity with a thin historic MEP labor base and a constrained PJM connection. Capital is arriving faster than local delivery capacity can form.

Phoenix (saturated primary) — Moderate exposure. Record absorption and heavy pre-leasing with a moderate trade base and somewhat better grid headroom than the top-constrained metros. The risk is incremental tightening as the existing base saturates.

Tennessee and Wisconsin (emerging-frontier) — Moderate exposure. Rising-concentration frontier markets with thin local depth selected primarily for power and land. Grid headroom is comparatively better, but construction-labor is shallow and mobilization risk is forming as pipelines deepen.

Rural corridors and generation-adjacent nodes — Moderate (widest dispersion). The thinnest local capacity of any market type, carrying the largest facility count and increasingly gigawatt-class loads steered by FERC colocation rules toward power-plant nodes with no construction-labor history. The constraint is almost entirely mobilization, housing, and retention.

10Planning scenarios

Base case — managed rebalancing lag. Concentration persists; lead and frontier markets absorb demand with schedule slippage and cost escalation concentrated in electrical and commissioning trades. Capital continues; delivery stretches. Aggregate load lands between the conservative and aggressive reads. This is the most evidence-consistent path and should anchor planning.

Upside-pressure case — coupled constraint tightens. Grid and labor bind simultaneously in top-tier and frontier markets, widening the announced-versus-built gap and pushing activity toward rural and generation-adjacent nodes faster than their delivery capacity can support. Cross-sector competition intensifies the draw on the shared trade pool.

Cooling case — partial relief. Modular construction, pre-assembled power systems, standardized campus designs, accelerated colocation rules, and efficiency gains reduce on-site labor intensity and interconnection delay enough to ease the binding constraint at the margin. Aggregate load tracks the conservative read (~65 GW through 2030). The constraint softens without disappearing.

The constraint is present in every scenario; only its magnitude varies. Plans anchored solely to the aggressive case carry overcommitment risk; plans anchored solely to the conservative case carry under-provisioning risk.

12Limitations

This publication is an intelligence assessment, not a forecast. Concentration and grid findings rest on government, grid-operator, institutional, and primary-research sources and are publication-grade. The quantitative labor-exposure framework (§6.6) is an illustrative planning lens calibrated to publicly documented peak-staffing observations; it carries Emerging confidence and should be applied at the unit and market level, not cited as a nationally measured estimate. Granular market-by-market matching of pipelines, interconnection queues, and trade availability is fragmentary in public sources; AlphaHire datasets are the intended bridge. Cross-sector competition for trades is under-quantified in joint regional workforce models. Aggregate load may run below the most aggressive forecasts; the thesis is framed with explicit scenario width to accommodate this. Social-license risk (community resistance, zoning, moratoria) is a distinct channel that interacts with, but is not reducible to, workforce availability.

@techreport{WILFLAG20262DCX,
  title       = {Concentrated by Design: AI Infrastructure Workforce Concentration and the Convergence of Construction, Grid, and Skilled-Trade Capacity Across U.S. Markets},
  author      = {AlphaHire Workforce Intelligence Lab},
  institution = {AlphaHire Workforce Intelligence Lab},
  type        = {Workforce Concentration Risk Series},
  number      = {WIL-FLAG-2026.2-DCX},
  year        = {2026},
  note        = {Version 2.0; methodology WIL-2026.1},
  url         = {https://library.alpha-hire.com/library/p/concentrated-by-design},
}

TY  - RPRT
AU  - AlphaHire Workforce Intelligence Lab
TI  - Concentrated by Design: AI Infrastructure Workforce Concentration and the Convergence of Construction, Grid, and Skilled-Trade Capacity Across U.S. Markets
PY  - 2026
PB  - AlphaHire Workforce Intelligence Lab
M1  - WIL-FLAG-2026.2-DCX
ET  - Version 2.0
UR  - https://library.alpha-hire.com/library/p/concentrated-by-design
AB  - Version 2.0 of the Workforce Concentration Risk Series flagship. Establishes the Q2 2026 baseline with 10 evidence-graded findings: AI data-center demand is concentrating in specific markets and trades faster than workforce or grid capacity can rebalance. The binding constraint is infrastructure delivery capacity — skilled electrical and mechanical labor and power infrastructure as a single coupled system. Covers geographic concentration, trade-specific scarcity, structural pipeline constraints, grid co-gating, operations long-tail demand, cross-sector competition, and bounding evidence across three planning scenarios.
ER  -