Tracing the 2 MW hall's power chain from utility service to rack — voltage levels, distribution method, and why this facility distributes at higher voltage than a typical enterprise data center.
The chain starts at the utility service entrance, stepped down through an on-site substation transformer to the facility's medium-voltage switchgear, then to 480V (line-to-line) distribution — one level higher than the 208V distribution common in legacy enterprise data centers. That choice matters at this rack density: at 40 kW/rack, a 208V circuit would need to carry roughly 111A per rack to deliver the load (P = V × I × √3 for three-phase, so I = P / (V × √3) = 40,000 / (208 × 1.732) ≈ 111A), pushing conductor and PDU sizing into impractical territory across dozens of racks. Distributing at 415V/480V to the rack cuts that current roughly in half for the same power, which is why higher-voltage distribution has become the default for GPU-density facilities rather than an exotic choice.
From the 480V distribution board, power runs to row-level Power Distribution Units (PDUs) via overhead busway rather than individual branch-circuit whips — busway's plug-in tap boxes let a row's rack layout be reconfigured (a real operational need as GPU generations turn over every 1-2 years) without re-pulling conductors. Each PDU feeds rack-level power strips (rPDUs), typically dual-corded from two independent PDUs fed from separate UPS sources — the physical basis for the redundancy Step 4 formalizes. This project's 40 kW/rack load, split across two independent 30A/415V three-phase whips per rack (roughly 21.6 kW capacity each, giving headroom above the 40kW/2=20kW per-path requirement), is a realistic circuit design a facility this size would actually use.
NEC Article 645 (Information Technology Equipment) governs much of this distribution once it's inside the recognized IT equipment room, including allowances for under-floor and overhead cabling methods not permitted elsewhere in the NEC — but Article 645's provisions require the room to meet specific construction and detection criteria first. Getting the distribution topology and voltage level right at this stage is what makes Step 3's UPS sizing a real engineering exercise instead of a guess — the UPS output voltage and the downstream distribution voltage have to match, and retrofitting a voltage mismatch after equipment is ordered is an expensive mistake.