Where AI Data-Center Construction Is Concentrating

Executive summary

Every tracked data center is a front-loaded construction claim before it is compute. The verified pipeline — a multi-state project set across 12 states and 14 markets, representing multi-gigawatt new load and multi-billion-dollar disclosed capex — is a forward order book for a narrow band of construction trades, not an IT line item.

It is happening now: the majority (61%) are already under construction, and 100% target an online date in 2026–2027. The crews for that wave are being hired today.

Concentration is extreme, not diffuse: Northern Virginia alone is ~31% of tracked MW; the Southeast is ~44% of MW; the top three states (Virginia, Ohio, Iowa) carry ~57% of capacity.

Every project is high-intensity. The tracked projects sit in the top two workforce-exposure bands and carry critical-or-high electrical pressure — a concentrated draw on electricians, controls and commissioning crews, and project leadership.

It collides with the grid: the same metros pulling this load (Virginia first) are where new generation supply and grid readiness are already constrained — three claims on one regional workforce.

Key metrics

Why this signal leads

A project enters this tracker once a developer has committed enough to disclose capacity, capital, and a target online date — typically one to three years ahead of operation. Between that disclosure and go-live sits the entire construction cycle: site civil work and foundations, structural shell, the enormous electrical scope (substations, switchgear, UPS and distribution), mechanical cooling systems, and an unusually long commissioning tail. Crews are mobilized in the run-up to the online date, not at it.

Data centers are electrically singular. A hyperscale campus concentrates more electrical and controls construction into a small footprint than almost any other building type, which is why a disclosed megawatt figure is, in labor terms, a near-term claim on exactly the trades the rest of the Lab's reporting flags as scarcest. Reading the data-center pipeline is reading the order book for that labor before the orders are filled — and because the builds cluster, the order book is lumpy by metro, not spread evenly across the country.

One caution up front: a tracked project is a commitment, not a guarantee. Capex and capacity figures are self-reported, and timelines slip. This pipeline is read as a directional, front-loaded demand signal — the shape and concentration of demand — rather than a precise build schedule, and treated accordingly throughout.

The tracked pipeline

The verified universe is a multi-state project set totaling multi-gigawatt load and multi-billion-dollar disclosed capital across 12 states and 14 markets. The project-type mix is the first labor tell, because each type pulls a different scale and trade blend.

By project type: Hyperscale (27 projects, 55.1%) · Colocation (18 projects, 36.7%) · Hyperscale campus (4 projects, 8.2%)

Hyperscale-class builds — standalone hyperscale plus multi-building campuses — are 31 of 49 (63%) of the pipeline. These are the largest, most electrically intensive, and most labor-concentrated projects, and they are what makes the metro-level concentration matter: a single campus can absorb a regional electrical workforce on its own. By status, 30 projects are already under construction and 19 are in committed planning — so most of the pipeline is past the decision point and into the field.

A broader monitoring set of ~112 projects across ~45 states (including unverified drafts) is tracked internally. This paper uses only the verified, publicly displayable subset of 49 to keep every figure defensible.

Geographic concentration

The pipeline is geographically lopsided to a degree that few construction signals match. The Southeast carries nearly half of all tracked load; the top three states carry more than half of capacity.

By region: Southeast (19 projects, 3,001 MW, 44.3%) · Midwest (10 projects, 1,788 MW, 26.4%) · Southwest (8 projects, 814 MW, 12.0%) · Pacific NW (4 projects, 550 MW, 8.1%) · South Central (4 projects, 264 MW, 3.9%) · Mountain West (2 projects, 250 MW, 3.7%) · Mid-Atlantic (2 projects, 100 MW, 1.5%)

By state (top five): Virginia (13 projects, 2,331 MW, $8.79B disclosed, 7 under construction) · Ohio (4 projects, 800 MW, $4.50B, 2 under construction) · Iowa (3 projects, 750 MW, $3.70B, 3 under construction) · Oregon (4 projects, 550 MW, $2.75B, 3 under construction) · Arizona (5 projects, 434 MW, $2.26B, 3 under construction)

By market: Northern Virginia (12 projects, 2,081 MW) is the standout — roughly a quarter of all tracked projects and nearly a third of tracked load in one metro. Columbus, the Iowa Hyperscale Corridor, and Portland form a clear second tier. Phoenix leads on project count (5 projects, 434 MW) even where its per-project megawatts run smaller.

Northern Virginia is the single densest concentration of data-center construction labor demand in the country, and it sits on top of the largest existing data-center footprint. The density in one metro is what makes the grid-collision risk acute: three claims on one regional electrical workforce.

Labor implications

Data-center construction pulls a distinctive, electrical- and controls-heavy trade blend, and the overlap with the broader construction market is concentrated in exactly the roles flagged as most constrained.

Electrical. Medium-voltage distribution, substations, switchgear, UPS, and power-distribution scope — the binding constraint, and the same electrical pool that generation and grid-reinforcement work draw from.

Controls and commissioning. Building-management systems, controls integration, and an unusually long commissioning tail — specialized roles that cannot be expanded quickly.

Mechanical and cooling. Liquid- and air-cooling systems, piping, and mechanical fit-out — scaling fast with AI-density designs.

Civil and structural. Earthwork, foundations, and shell — front-loaded, and heavy on the large campus builds.

Project leadership. Superintendents, project managers, commissioning leads, and field engineers — the seats whose scarcity gates how many mega-projects can run concurrently in one metro.

The electrical and leadership lines are the ones to watch. The Data Center Pipeline™ is the load side of a three-way competition for one workforce: the Power Generation Pipeline™ measures the generation that must be built to feed this load; the Grid Constraint Monitor™ measures whether each state's grid can absorb it; and the Workforce Scarcity Index™ measures how scarce the electricians and project leaders are to begin with. Reports should never be read in isolation; this one is the load that the other three respond to.

How to apply this

Methodology and limitations

Primary source. The pipeline is built from the AlphaHire data-center activity tracker — a curated registry of major U.S. data-center projects with disclosed capacity, capital, project type, status, and target timeline. This paper uses only the verified, publicly displayable universe (a multi-state set across 12 states and 14 markets). A broader monitoring set, including unverified drafts, is tracked internally but excluded here pending verification, so that every published figure is defensible. Published under methodology DCP-v1.0.

Workforce-exposure, electrical-pressure, and leadership-pressure bands are AlphaHire classifications derived from project scale, type, and scope; they are directional labels, not engineered labor counts. Capacity (MW) and capital figures are as disclosed by developers and may revise.

What this report shows. The concentration, timing, type mix, and labor-exposure profile of verified major data-center construction — read as a directional, front-loaded indicator of electrical, mechanical, and project-leadership demand.

What this report does not show. It is not a forecast or a national census. It is a curated tracker of major disclosed projects — useful for locating concentration, not for totaling national data-center demand. Capacity and capital figures are self-reported and may revise; online dates slip. Because all 49 are curated major builds, the all-top-band exposure profile reflects the sample, not all data centers. No firm, developer, project, or site is named.

Confidence level. High on the existence and concentration of the tracked projects. Moderate/directional on the exposure and pressure bands (AlphaHire composites). Sample-based/illustrative on the totals, which represent a curated, non-exhaustive set rather than national demand. Trust the concentration and the direction; do not read the totals as a national figure.

@techreport{WILIP20262,
  title       = {Where AI Data-Center Construction Is Concentrating},
  author      = {AlphaHire Workforce Intelligence Lab},
  institution = {AlphaHire Workforce Intelligence Lab},
  type        = {Intelligence Paper},
  number      = {WIL-IP-2026.2},
  year        = {2026},
  note        = {Version 1.0; methodology DCP-v1.0},
  url         = {https://library.alpha-hire.com/library/p/data-center-pipeline},
}

TY  - RPRT
AU  - AlphaHire Workforce Intelligence Lab
TI  - Where AI Data-Center Construction Is Concentrating
PY  - 2026
PB  - AlphaHire Workforce Intelligence Lab
M1  - WIL-IP-2026.2
ET  - Version 1.0
UR  - https://library.alpha-hire.com/library/p/data-center-pipeline
AB  - A data center is a construction project long before it is a cloud region. The AlphaHire-tracked pipeline of verified major U.S. hyperscale and colocation projects represents multi-gigawatt new load and multi-billion-dollar disclosed capex across 12 states and 14 markets — with the majority already under construction and 100% targeting a 2026–2027 online date. This paper reads that pipeline as a leading indicator of electrical, mechanical, and project-leadership labor demand, maps its extreme geographic concentration (Northern Virginia alone ~31% of tracked MW), and connects it to the broader grid-and-generation collision that makes the labor claim uniquely acute.
ER  -