Rack density measures the total electrical power consumed by a single IT cabinet. Driven by AI training hardware, rack densities have surged from a historical average of 5-7 kW to over 100 kW per rack, fundamentally obsoleting the electrical and cooling infrastructure of legacy data centers.
Legacy enterprise data centers were engineered to deliver 5 to 10 kW of power and cooling to a standard 42U rack using raised floors and forced air. A modern AI rack packed with NVIDIA DGX systems or high-end ASIC miners demands 40 to 120 kW. Delivering this requires bringing 415V or 480V 3-phase power directly to the rack PDUs, deploying in-row liquid cooling, and structurally reinforcing the floor to handle the immense weight of liquid-filled servers.
High-Voltage Distribution
Power is distributed to the racks at 415V 3-phase (rather than 208V or 120V) to reduce amperage and minimize massive copper cable sizing.
Busway Deployment
Overhead continuous busways replace underfloor whips, allowing operators to drop 60A or 100A tap boxes directly above high-density racks.
Liquid Cooling Integration
Rear Door Heat Exchangers (RDHx) or Direct-to-Chip manifolds are installed to neutralize the 100kW thermal load before it cooks adjacent racks.
Structural Reinforcement
A fully populated 100kW AI rack with liquid manifolds can weigh over 3,500 lbs, requiring slab-on-grade construction rather than legacy raised floors.
- Stranded Capacity: A facility designed for 10 MW across 1000 racks can now hit its 10 MW limit with just 100 AI racks, leaving 90% of the building empty but un-leasable.
- Weight Constraints: Multi-story data centers cannot structurally support the point-loads of 3,500 lb racks, relegating AI clusters to ground floors.
- Electrical Lead Times: Sourcing the massive switchgear and transformers required to feed 100kW racks takes 18-24 months due to supply chain backlogs.