Data Center Construction
Built for Uptime

The AI Infrastructure Shift

Artificial intelligence is fundamentally changing how data centers are designed and built. Two years ago, a high-density rack drew 10-15 kW. Today, NVIDIA H100 GPU racks draw 40-50 kW. The new Blackwell B200 and GB200 systems push past 100-120 kW per rack — well beyond what traditional air cooling can handle.

This shift means data center construction is no longer just about building rooms with raised floors and perimeter CRAC units. Modern facilities need liquid cooling infrastructure (direct-to-chip cold plates, coolant distribution units, manifold plumbing), high-capacity power distribution (400V/480V busbar systems delivering 100A+ per rack), and thermal management designed for bursty AI training workloads that can spike from idle to full power in seconds.

Colorado Quality Construction builds and retrofits data centers for this new reality. We understand the difference between training facilities (highest density, liquid-cooled, hyperscale) and inference facilities (moderate density, often air-cooled, geographically distributed for latency). We build to Uptime Institute Tier standards, ASHRAE TC 9.9 thermal guidelines, NEC Article 645, NFPA 76, and TIA-942 — and we can work in live facilities without taking your operations offline.

White Space and Gray Space Construction

Data center construction divides into two scopes that require different expertise:

Gray space is the support infrastructure — electrical rooms with switchgear, UPS systems, and transformers; mechanical rooms with chillers, CDUs, pumps, and cooling towers; generator yards; battery rooms; and telecom entrance facilities. This is heavy MEP construction: large equipment setting, medium-voltage work, chilled water piping, and fuel storage systems. Gray space is typically built first as the building shell and support infrastructure.

White space is the IT equipment room itself — where the racks live. This includes raised access flooring or slab construction, hot/cold aisle containment structures, CRAH/CRAC unit installation, in-row coolers, CDU and manifold placement for liquid cooling, busbar and PDU power distribution, overhead cable tray systems, pre-action and clean agent fire suppression piping, leak detection systems, and physical security infrastructure (mantraps, biometric readers, security cages). White space has stricter environmental controls — temperature, humidity, cleanliness, and fire suppression — and is typically phased to align with IT deployment schedules.

Cooling Infrastructure for AI Workloads

Cooling is the single most complex engineering challenge in modern data center construction. The heat generated by dense GPU clusters cannot be removed with traditional air cooling alone. We build cooling infrastructure across the full spectrum:

Air Cooling with Containment

For racks up to 30-40 kW, enhanced air cooling with hot/cold aisle containment and high-capacity CRAH units can work. We build containment structures (blanking panels, aisle doors, roof panels) and install higher-capacity cooling units. This is the most cost-effective approach for moderate-density workloads and remains common in inference and edge facilities.

Rear-Door Heat Exchangers

For 15-50 kW racks, rear-door heat exchangers (RDHx) mount on the back of server racks and remove heat at the source using chilled water. Passive RDHx handle up to 25 kW per rack; active (fan-assisted) units reach 60 kW. RDHx are retrofit-friendly — they can be installed on existing racks without major infrastructure changes, making them ideal for density upgrades in operating facilities.

Direct-to-Chip Liquid Cooling

For AI training clusters drawing 40-120 kW+ per rack, direct-to-chip (D2C) liquid cooling is becoming standard. Cold plates mounted directly on GPU and CPU dies circulate coolant that removes heat far more efficiently than air. We install the full D2C infrastructure: coolant distribution units (CDUs) sized 300-1,200+ kW, primary and secondary coolant loops, manifold piping to each rack, and Universal Quick Disconnect (UQD) fittings that allow hot-swap of servers without draining the loop. D2C supply water runs at 30-45°C — warmer than traditional chilled water — enabling higher chiller efficiency and more free cooling hours.

Immersion Cooling

For the highest-density deployments (200 kW+ per tank), immersion cooling submerges entire servers in dielectric fluid. We build immersion tank platforms, fluid handling systems, and heat exchanger infrastructure. While less common than D2C, immersion is gaining traction for specialized HPC and AI workloads where maximum heat removal is required.

Power Distribution for High-Density Racks

AI rack power density requires a fundamentally different power architecture than traditional data centers. A 120 kW rack at 208V would draw 575 amps — impractical with standard cabling. The industry is moving to:

• 400V/480V direct-to-rack distribution: Higher voltage means lower current for the same power, reducing conductor sizing, copper costs, and I²R transmission losses. Busbar systems can deliver 100A+ per rack at 400V.
• Overhead busbar (busway): Busbar trunking runs overhead with plug-in tap boxes at each rack. This enables flexible reconfiguration — adding or moving circuits without pulling cable — and supports the high capacities that AI racks demand.
• Redundant power feeds (N+1, 2N): Critical facilities require redundant power paths. We install dual busbar systems, dual PDUs per rack, and automatic transfer switches (ATS) that maintain power during utility or generator transitions.
• UPS and generator backup: AI training workloads are intolerant of power interruptions — a 50ms outage can destroy hours of training progress. We install UPS systems sized for the full IT load with adequate battery runtime, plus standby generators with fuel storage for extended outages.

All electrical work is performed by licensed electricians and inspected to NEC Article 645 standards, which governs IT equipment power distribution, grounding, and the single disconnecting means required for emergency response in data center environments.

Why Colorado for Data Centers

Colorado offers several natural advantages for data center development, particularly for facilities that can leverage the climate for free cooling:

• Free cooling hours: Denver's cool, dry climate provides 5,000-6,500 hours per year of airside economizer potential and 4,000-5,500 hours of waterside economizer potential. Well-designed Colorado facilities can achieve PUE of 1.15-1.25 — significantly better than warmer, humid climates.
• Dry climate advantage: Low humidity enables highly effective adiabatic and evaporative cooling at minimal water usage. Colorado's low wet-bulb temperatures mean cooling towers and evaporative systems perform exceptionally well.
• Fiber connectivity: Denver is a major interconnection hub. CoreSite and Equinix operate carrier hotels with dozens of network providers, cloud on-ramps (AWS Direct Connect, Azure ExpressRoute, Google Cloud Interconnect), and transcontinental fiber routes to Chicago, Silicon Valley, Los Angeles, Dallas, and Seattle.
• Lower seismic risk: Colorado has moderate seismic risk (USGS zones 2-4%g) — lower than California, reducing the need for costly seismic hardening while still requiring equipment anchorage and flexible connections per IBC.

Colorado-Specific Challenges We Address

• Water usage (WUE): Colorado is a drought-prone headwater state with complex water rights. We design low-WUE cooling systems — air-cooled chillers, dry coolers, adiabatic systems with minimal water use, and closed-loop cooling towers — to meet water efficiency targets under 0.5 L/kWh where required.
• Wildfire smoke: During fire season, PM2.5 levels can spike to 100+ µg/m³. Airside economizers must shut down or switch to MERV 13-15 filtration with gas-phase (activated carbon) filters for VOC and corrosive gas removal. We install air quality monitoring systems tied to economizer controls and redundant filtration for outside air intakes.
• Utility coordination: Xcel Energy interconnection studies for large data center loads can take 12-24+ months. We coordinate early with utilities on capacity, rate structures, and renewable energy procurement (on-site solar, battery storage, PPAs) to keep projects on schedule.
• Expansive soils: Many Front Range sites have expansive clay soils that require special foundation design for data center floor loading — particularly important for liquid cooling infrastructure and battery rooms that can exceed 200+ PSF floor loads.

Building in Live Data Centers

Many of our projects are retrofits and upgrades in operating facilities — the client cannot afford downtime. Construction in a live data center requires protocols that go far beyond standard commercial construction:

• Dust containment: Temporary barriers with negative air pressure and HEPA filtration prevent particulates from entering white space. Even small amounts of dust can clog server fans and damage sensitive electronics.
• Vibration monitoring: Sensors detect vibrations that could affect disk drives and sensitive equipment, particularly during demolition, drilling, or concrete work. Work stops immediately if thresholds are exceeded.
• Phased work windows: Noisy or disruptive work is scheduled during maintenance windows — often nights, weekends, or planned outages. We coordinate months in advance with the facility operations team.
• Fire system interlocks: We coordinate with the facility's VESDA detection and clean agent suppression systems to prevent accidental discharge during construction. Pre-action sprinkler systems are locked out during hot work.
• Power isolation: Strict lockout/tagout procedures when working near energized electrical systems. We identify and isolate circuits before any work near live power distribution.
• Security protocols: Background-checked, badged, and escorted personnel in secure areas. We comply with each facility's access control and visitor management requirements.

Commissioning and Handover

Data center commissioning is a rigorous, multi-level process that verifies every system works individually and together before IT equipment is deployed. We support the full five-level commissioning process:

• Level 1 — Factory Acceptance Testing: Equipment tested at the factory before shipping. We verify factory test reports and confirm equipment meets spec.
• Level 2 — Site Acceptance Testing: Equipment inspected and tested on-site after delivery. We verify no shipping damage and confirm proper installation.
• Level 3 — Subsystem Testing: Individual systems (electrical, mechanical, fire suppression) tested independently. We support the testing contractor and resolve any installation issues identified.
• Level 4 — System Integration Testing (SIT): All systems tested together under load using load banks to simulate IT load. This includes failure scenario testing (utility loss, generator failover, cooling failure). Required for Uptime Tier III/IV certification.
• Level 5 — Integrated System Testing under IT Load: Testing with actual IT equipment installed and operational. Validates performance under real thermal and power loads.

We deliver full as-built documentation including drawings, equipment submittals, warranty information, operation and maintenance manuals, and commissioning reports. This documentation package is essential for facility management, warranty claims, and future expansion planning.

Project Types We Deliver

• Colocation build-outs: Multi-tenant facility construction with dedicated suites, secure cages, and shared infrastructure. Speed to market is critical — we deliver in 8-16 weeks for single-suite build-outs.
• Enterprise data centers: Private, single-tenant facilities for corporations, financial institutions, and healthcare systems. Built to the client's specific availability targets (typically Tier III).
• Edge and micro data centers: Small-footprint modular and containerized facilities for edge computing and distributed networks. Often deployed near cell towers, regional offices, or military installations.
• Retrofits and density upgrades: Converting legacy air-cooled facilities to support higher-density workloads. This includes installing liquid cooling infrastructure, upgrading power distribution, and adding containment — all in live facilities.

Data Center Construction FAQ