AI workloads have pushed rack densities from single digits to well over 100kW, but the electrical infrastructure behind those racks hasn’t kept pace, and Kevin Cruts, Senior Director of Data Center Solutions at Turtle, walks through why that gap now determines project timelines.
He points to a “spatial paradox” where the equipment needed to support high-density racks eats into the footprint originally reserved for compute, and to switchgear and transformer lead times stretching well over a year, which means procurement decisions can no longer wait until design is finalized.
Cruts also flags a structural shift underway: GPU-dense clusters are moving toward 800-volt DC distribution, forcing project teams to manage AC and DC architectures side by side on the same campus, each with its own protection scheme and coordination requirements. His takeaway for project teams is to start procurement earlier, specify for the architecture the facility will actually run, and treat AC/DC coordination as one integrated design problem rather than two separate tracks.
For the original byline for Data Center Dynamics, click here.
Getting the basics right
Five years ago, a standard data center rack drew five to eight kW. New AI facilities are designed for 15 to 50kW per rack. GPU-dense AI configurations reach 100 to 250kW. New builds are on track to average 45kW per rack, with 80 percent incorporating liquid cooling from the ground up.
The compute layer has changed. The electrical infrastructure supporting it has not kept pace.
This is the gap that owners, general contractors, electrical contractors, and engineers of record are all navigating right now. Decisions made during design and early procurement determine whether a facility reaches commercial operation on schedule or sits structurally complete, waiting on power.
The spatial paradox
As rack density rises, the upstream electrical infrastructure required to support it consumes more of the physical footprint originally allocated to compute. Engineers call this the spatial paradox. It forces a complete rethinking of how electrical scope is sequenced, specified, and built into the project.
The problem compounds at the equipment level. For AI-density configurations, switchgear and transformers typically carry lead times of 52 to 78 weeks for distribution-class units. Large power transformers run 128 weeks or longer, depending on the manufacturer and configuration. High-voltage utility interconnections, on-site substations, and redundant distribution paths all require design lock and procurement decisions well before a project’s civil and structural work is complete. Equipment delivery now anchors the construction cycle, ahead of labor availability and site readiness.
Project teams that treat electrical procurement as a post-design activity consistently reach the same outcome: structural work finishes on schedule and sits idle, waiting on switchgear.
A new electrical architecture
In new facilities designed around AI workloads, the shift from alternating current to 800-volt direct current distribution is accelerating. Most data center construction still operates on AC infrastructure. For facilities built around GPU-dense AI clusters, DC distribution is becoming the architecture of record. It represents a fundamental change in how power moves from the utility feed to the rack.
DC-capable systems require new categories of switchgear, DC-ready circuit breakers, solid-state transformers, and cable architecture designed for the ampacity demands of high-density AI clusters. In GPU-dense deployments, busbars are displacing traditional cabling because they provide the flexibility and current-carrying capacity that conventional cable systems cannot match at extreme rack densities.
Electrical contractors entering these projects for the first time find that the scope is not simply larger. It is structurally different.
Most electrical contractors carry deep experience with AC infrastructure. Teams that can design, specify, and commission DC-capable systems are in short supply. Recognizing that gap early and staffing for it is not a competitive advantage. It is a scheduling requirement.
Two architectures, one campus
A single data center campus today may need to support both architectures in parallel. Traditional enterprise and cloud workloads continue to run on AC infrastructure. AI clusters require high-density DC systems. Each architecture carries its own protection scheme, monitoring layer, and power path.
Managing two concurrent electrical architectures introduces coordination demands that most facilities teams have not encountered before. The interconnection between those systems, the handoff points, and the protection coordination across both must be designed together, not sequenced independently.
Electrical contractors entering these projects for the first time find that the scope is not simply larger. It is structurally different.
What the project team needs to do differently
Three shifts in project execution directly address the infrastructure gap.
First, initiate electrical procurement before the final design is complete. Equipment lead times make it necessary. Switchgear and transformer specifications can be developed in parallel with design development. Distribution partners with real-time visibility into manufacturer availability support a parallel track.
Second, specify for the electrical architecture the facility will actually operate, not the one that was standard when the design team last built a data center. DC-ready equipment, high-ampacity busway systems, and power modules designed for liquid-cooling integration are available now. The right distribution partner can translate specification requirements into available inventory and flag gaps before they become schedule risks.
Third, coordinate both electrical architectures as a single integrated system from the design phase. Protection scheme design, monitoring integration, and commissioning sequencing across AC and DC systems require early collaboration between the engineer of record, the electrical contractor, and the equipment suppliers.
The data center switchgear market alone is projected to reach $13.6 billion by 2031, growing at 16 percent annually. That growth reflects real construction volume. The projects driving it will be delivered by teams that recognized the infrastructure gap early and built their execution strategy around it.
The intelligence gap in AI data center infrastructure is not in the compute layer. It sits in the electrical design decisions made before a single rack is installed. Project teams that resolve it early will deliver. Those who discover it late will wait.
Let’s discuss your project requirements
