Construction & Power Layer

The construction layer is where the AI factory becomes physical reality. This is civil engineering at the scale of airports and power plants — but on 18-month timelines with first-of-kind technical requirements.

The Value Chain

graph LR
    A[Site Selection] --> B[Permitting & Approvals]
    B --> C[EPC Contract]
    C --> D[Ground Up Construction]
    D --> E[MEP Installation]
    E --> F[Commissioning]
    F --> G[GPU Installation]

1. EPC Contractors (Engineering, Procurement, Construction)

Key Players: Turner, Hensel Phelps, Mortenson, DPR, Jacobs, Bechtel, Fluor, Black & Veatch, McDermott, Kiewit, PCL, Swinerton

Delivery Models:

• Design-Bid-Build: Traditional, slower, owner controls design
• Design-Build: Single contract, faster, contractor optimizes
• EPC (Turnkey): Fixed price/date, contractor takes risk — preferred for AI factories
• Construction Management at Risk (CMAR): Early contractor involvement, GMP

2. Specialized MEP Subcontractors (The Critical Path)

Key Players: Cupertino Electric, Rosendin, Helix Electric, Faith Technologies, M.C. Dean, EMCOR, Comfort Systems, TDIndustries

3. Power Infrastructure (The Ultimate Constraint)

Key Players: Utilities, GE Vernova, Siemens Energy, Mitsubishi, Schneider, Eaton, ABB, Vertiv, Generac, Kohler, Caterpillar

Grid Connection (The #1 Bottleneck)

On-Site Generation (Bypassing the Queue)

Power Density Requirements

4. Thermal Management at Scale

Key Players: Vertiv, Schneider Electric, Delta Electronics, SPX Cooling, Baltimore Aircoil, EVAPCO, Johnson Controls, Trane, Carrier, Daikin

Water Usage (Critical Constraint):

• Evaporative cooling: 1-3 million gallons/day per 100MW
• Dry cooling: 0 water, 1.3-1.5x capital cost, 5-10% efficiency penalty
• Closed-loop + cooling tower: 500K-1M gallons/day
• Immersion: Minimal water, but fluid cost $50-100/gal

5. Site Selection Criteria (The Real Estate Decision Matrix)

Top US AI Factory Regions (2024):

1. Northern Virginia (Data Center Alley) - Power constrained, expensive
2. Phoenix/Mesa, AZ - TSMC fab, good power, water risk
3. Dallas/Fort Worth, TX - Cheap power, land, growing fiber
4. Atlanta, GA - QTS campus, good power, incentives
5. Columbus, OH - Intel fab, hyperscale, renewables
6. Salt Lake City, UT - NSA, NSA, cheap power, fiber
7. Reno/Sparks, NV - Tesla, Switch, geothermal, no tax
8. Memphis, TN - xAI Colossus, TVA power, river cooling
9. North Dakota - Applied Digital, wind, flared gas, cold
10. Central Washington - Hydro power, cool climate, Microsoft

6. Permitting & Regulatory (The Hidden Timeline)

Strategies to Accelerate:

• Pre-permitted campuses (Switch, Digital Realty, Vantage)
• Over-the-fence agreements with existing industrial
• Brownfield redevelopment (former aluminum, paper mills)
• Federal land (DOE, DOD, BLM) for strategic projects
• Emergency/expedited authority (defense production act)

Construction Timeline (18-Month Target)

Months 1-3:   Site prep, permits, long-lead equipment orders
Months 3-6:   Foundations, underground utilities, substation civil
Months 6-9:   Steel structure, roof, envelope, major equipment delivery
Months 9-12:  MEP rough-in, chiller/cooling tower install, switchgear
Months 12-15: MEP fit-out, CDU/manifold install, controls, testing
Months 15-18: Commissioning, burn-in, GPU installation, go-live

Critical Path Items (Order by Month 1):

1. 500kV transformers & switchgear (24-36 mo)
2. Large centrifugal chillers (12-18 mo)
3. Generators & UPS (12-18 mo)
4. Cooling towers / dry coolers (9-12 mo)
5. CDUs & manifolds (6-9 mo)
6. Structural steel (4-6 mo)

Cost Breakdown (Per 100MW AI Factory Campus)

The construction layer is where AI ambition meets physical reality. The 18-month build cycle for a 100MW+ AI factory requires orchestrating thousands of skilled trades, managing hundred-million-dollar equipment lead times, and navigating a regulatory maze that can add years. Companies that master this — Turner, Hensel Phelps, Mortenson, and the specialized MEP subs — are the unsung enablers of the AI revolution.