Site Selection
Site selection is the most consequential data center design decision. The site determines power availability, natural disaster risk, connectivity options, labor market, tax environment, and long-term operational cost. Poor site selection cannot be corrected after construction.
Power Availability
Verify that the utility can provide sufficient power at the required voltage and reliability level. For large data centers (10+ MW), engage the utility early — transmission infrastructure upgrades can take 2–5 years. Assess utility reliability history, redundant feed availability, and rate structures.
Natural Disaster Risk
Evaluate flood risk (100-year and 500-year flood plains), seismic risk (USGS seismic hazard maps), tornado and hurricane risk, and wildfire risk. Mission-critical facilities should avoid high-risk zones or design for the specific risk (seismic bracing, flood barriers, hardened construction).
Connectivity
Assess fiber connectivity options — number of carriers, diverse entry points, and available bandwidth. For colocation facilities, carrier diversity is a key differentiator. For enterprise data centers, verify that required connectivity can be provisioned at the site.
Labor Market
Data center operations require skilled technical staff. Assess the local labor market for electrical engineers, mechanical engineers, and data center technicians. Remote locations may offer lower land and power costs but face staffing challenges.
Tax and Incentive Environment
Many jurisdictions offer tax incentives for data center investment: sales tax exemptions on equipment, property tax abatements, and economic development grants. These incentives can significantly reduce total cost of ownership and should be evaluated as part of site selection.
Civil & Structural Design
Floor Loading
Data center floors must support the weight of IT equipment, raised floors, and infrastructure. Traditional IT equipment averages 100–150 lbs/sq ft. AI-dense deployments with liquid cooling infrastructure can exceed 300 lbs/sq ft. Structural design must account for concentrated loads from high-density racks.
Raised Floor vs. Overhead Distribution
Raised floors (12–24 inches) provide underfloor space for power and cooling distribution. Overhead distribution (cable trays, busway, overhead cooling) is increasingly preferred for new construction — it provides better visibility, easier maintenance, and eliminates the floor loading constraints of raised floors.
Ceiling Height
Minimum 12-foot clear height for the data center floor. Higher ceilings (14–16 feet) provide better airflow management and accommodate overhead infrastructure. Generator and UPS rooms require additional height for equipment clearances.
Power System Design
Power system design must balance reliability, efficiency, and flexibility. Key design decisions:
Utility Entry
Design for two utility feeds from diverse substations. Utility entry points should be physically separated to prevent a single event (vehicle impact, fire) from affecting both feeds. Underground utility entry is preferred for protection against weather events.
UPS Architecture
For Tier III facilities: distributed UPS (one UPS per power zone) provides better fault isolation than centralized UPS. For Tier IV: 2N UPS architecture with independent A and B power paths throughout. Lithium-ion UPS is increasingly preferred for new construction due to longer life and smaller footprint.
Generator Placement
Generators should be located outside the data center building, with fuel tanks positioned to minimize fire risk. Exhaust routing must comply with local air quality regulations. Generator enclosures should be weatherproof and provide sound attenuation.
Future Capacity
Design power infrastructure for 150–200% of initial load to accommodate growth. Conduit and busway systems should be sized for future capacity even if initial equipment is smaller. Adding power capacity after construction is significantly more expensive than designing for it initially.
Cooling System Design
Cooling Architecture Selection
Select cooling architecture based on target rack density. For mixed-density environments, design for the highest-density zone and use appropriate cooling for lower-density areas. Designing for future density is critical — retrofitting liquid cooling infrastructure is expensive and disruptive.
Chiller Plant Design
For large facilities, a central chiller plant provides the most efficient cooling. Design for N+1 chiller redundancy. Water-cooled chillers are more efficient than air-cooled but require cooling towers and water treatment. Free cooling (economizer mode) can dramatically reduce cooling energy in cooler climates.
Liquid Cooling Infrastructure
For AI-dense deployments, design liquid cooling infrastructure from the start: chilled water or facility water loops to each rack row, leak detection systems throughout, manifolds and quick-disconnect fittings at rack level, and secondary containment for liquid cooling areas.
Network Infrastructure Design
Spine-Leaf Architecture
Modern data centers use spine-leaf network architectures: leaf switches connect to servers; spine switches connect to all leaf switches. Any server can reach any other server in two hops. Provides predictable, low-latency connectivity and easy horizontal scaling.
Diverse Fiber Entry
External fiber should enter the building at two physically diverse points. Fiber should be routed through diverse paths within the building to the main distribution frame (MDF). A single fiber cut should not affect all external connectivity.
Cable Management
Structured cabling with color coding, labeling, and documentation from day one. Overhead cable trays for horizontal distribution; vertical cable managers within racks. Separation of power and data cables to minimize electromagnetic interference.
Physical Security
Physical security must be designed in layers — perimeter, building, data center floor, and cage/cabinet level:
- Perimeter: Fencing, vehicle barriers, CCTV, and controlled vehicle access
- Building: Controlled entry with badge readers, visitor management, and security desk
- Data center floor: Mantrap entry (two-door airlock), biometric authentication, CCTV coverage of all aisles
- Cabinet level: Locked cabinets with electronic access control for multi-tenant environments
Security systems should be on independent power (UPS-backed) and separate from the IT network. Security video should be retained for 90 days minimum.
AI-Optimized Design
Data centers designed for AI workloads have specific requirements that differ from traditional enterprise data centers:
- Power density: Design for 30–120 kW per rack; traditional 5–10 kW per rack designs cannot support AI infrastructure
- Liquid cooling: Design liquid cooling infrastructure from the start; retrofitting is expensive
- Floor loading: AI servers with liquid cooling infrastructure can exceed 300 lbs/sq ft
- Network: InfiniBand NDR (400 Gb/s) requires specific cabling (copper DAC for short distances, active optical cables for longer runs)
- Power quality: AI servers with switching power supplies require careful harmonic analysis and power factor correction
- Expansion flexibility: AI hardware generations change rapidly; design for easy reconfiguration
Commissioning
Commissioning validates that the as-built facility performs as designed. A well-designed facility can fail in operation due to poor commissioning — installation errors, configuration mistakes, and integration failures that are not caught before go-live.
Commissioning Levels
- Level 1 (Component): Individual equipment tested in isolation
- Level 2 (Subsystem): Related components tested together (e.g., UPS + batteries + bypass)
- Level 3 (System): Complete systems tested under simulated load (e.g., full power path from utility to rack)
- Level 4 (Integrated): All systems tested together under actual load, including failure scenarios
- Level 5 (Operational): Facility operated under normal conditions with all systems active
Integrated systems testing (Level 4) is the most valuable commissioning activity — it tests the interactions between systems that are most likely to reveal design or installation errors. Failure scenarios tested should include: utility power failure, UPS failure, generator failure, cooling failure, and fire suppression activation.