Published: June 2026 Technical Level: Advanced Category: Power Systems Design
This paper presents the complete electrical power system design for a 10 MW Tier III colocation data center, documenting the design decisions, calculation basis, and equipment selection that translate the owner's reliability and efficiency requirements into a code-compliant, constructable electrical system. The facility is designed to meet Uptime Institute Tier III concurrent maintainability requirements: any single power component or distribution path can be taken out of service for maintenance without interrupting power to any IT cabinet. The design achieves a predicted PUE of 1.28 using high-efficiency double-conversion UPS with lithium-ion batteries, transformer-less distribution to IT load, and airside economizer cooling. The case study addresses the engineering decisions that most frequently drive cost and schedule risk in data center power system projects: utility service configuration, UPS sizing and battery runtime, generator paralleling and automatic transfer, and arc flash hazard mitigation strategy.
The facility is a 100,000 square foot purpose-built colocation data center in Phoenix, Arizona, with 10 MW of installed IT load capacity in 8 data halls of 1,250 kW each. The utility service is 69 kV from a nearby substation, stepped down to 13.8 kV at the facility-owned primary substation, and then to 480Y/277 V at each data hall's electrical room. The facility is designed for Class A2 operating environment per ASHRAE TC 9.9 (18–27°C, 8–80% relative humidity at the server inlet).
The power reliability requirement is Tier III: concurrent maintainability on all power paths. The efficiency requirement is PUE ≤ 1.30 annualized. The mechanical cooling design uses airside economization for free cooling when the outdoor temperature and humidity permit, active direct evaporative cooling for intermediate conditions, and mechanical chilled water cooling for the hottest and most humid periods — a strategy well-matched to the Phoenix climate where dry-bulb temperatures below 65°F occur for approximately 3,500 hours per year, providing significant free cooling opportunity.
The 69 kV service enters the site through two geographically separated feeders from the same utility substation, connected to separate 69-13.8 kV step-down transformers at the facility primary substation. The two transformers are each rated for 15 MVA — 50 percent above the 10 MW facility IT load at unity power factor — providing capacity margin for the facility infrastructure loads and for load growth to 12 MW IT in Phase 2. Each transformer feeds a separate bus section of the 13.8 kV switchgear, with a normally-open bus tie breaker between sections. Under normal operation, each bus section serves half the facility load; upon loss of one utility feed, the normally-open tie breaker closes automatically (via an automatic transfer scheme incorporating bus under-voltage detection and a synchronism check relay) and one transformer serves the full facility load.
The selection of 69 kV as the service voltage, rather than 12.47 kV or 13.8 kV distribution class, was driven by the facility's load magnitude and the utility's available service voltage at the delivery point. At 10 MW, the fault current at a 12.47 kV secondary service would be sufficient to require switchgear with 40 kA interrupting rating in several circuit breaker positions. Stepping down from 69 kV at a facility-owned transformer allows the engineer to select transformer impedance to limit fault current to a level manageable with standard 25 kA or 31.5 kA medium-voltage switchgear, reducing equipment cost.
Each data hall is served by a dedicated UPS system consisting of two 750 kW modular double-conversion UPS frames, each containing three 250 kW power modules, configured in a 2N architecture. Under normal operation, one UPS frame (750 kW capacity) supplies the data hall's critical load bus; the second frame is on standby, ready to assume full load within the UPS static transfer switch's transfer time of 4 milliseconds. The 2N architecture meets Tier III concurrent maintainability: either UPS frame can be taken out of service for maintenance while the other sustains the full 1,250 kW data hall load.
Each UPS frame is equipped with a lithium iron phosphate (LFP) battery system providing 10 minutes of runtime at full 750 kW discharge rate. The 10-minute runtime is the design basis for the time between utility power loss and the time that the engine-generator set assumes the facility load: 20 to 30 seconds for generator start and load acceptance, plus margin for the generator voltage and frequency stabilization period. The 10-minute battery provides more than 20 times the required margin, which is appropriate for a colocation facility whose clients' SLAs specify maximum unplanned downtime of less than 1.6 hours per year (Tier III benchmark).
The LFP battery selection was driven by three advantages over valve-regulated lead-acid (VRLA) technology that are significant at the 10 MW scale: the higher cycle life (3,000 to 5,000 cycles versus 300 to 500 for VRLA at comparable depth of discharge) reduces the expected replacement frequency from every 4 to 5 years for VRLA to every 10 to 12 years for LFP; the lower weight per kWh (approximately 50 percent lighter than VRLA) reduces the structural floor loading requirement, eliminating the need for structural reinforcement in the electrical rooms; and the narrower temperature operating range of LFP (allowing operation up to 45°C without accelerated degradation, versus 25°C maximum for VRLA) improves compatibility with the data center's airside economizer cooling approach.
The facility is served by six 2,500 kW diesel engine-generator sets, each connected to a separate bus section of the 13.8 kV medium-voltage switchgear. The generators are sized such that any three of the six can supply the full 10 MW IT load plus facility infrastructure overhead at 1.25 PUE (12.5 MW total), providing N+1 redundancy at the generator level. Generator paralleling is managed by a digital synchronizing and load-sharing control system that monitors voltage, frequency, and phase angle on each generator bus and the main switchgear bus, and issues close commands to the generator breakers in sequence after verifying synchronism within ±5° and ±0.5 Hz.
The automatic transfer scheme operates as follows: upon utility power loss, all six generators start simultaneously. The first generator to reach 90 percent rated voltage and frequency — typically within 10 seconds — connects to its bus section and accepts the facility load for that section. Subsequent generators connect and load share as they come online, with full load sharing achieved within 25 to 30 seconds. The UPS systems sustain IT load from battery during this interval.
The arc flash study for this facility was performed per IEEE 1584-2018 using SKM PowerTools, with fault current data from the short-circuit study. The highest incident energy in the facility occurs at the 480 V UPS output switchboard, where the available fault current is 65 kA and the upstream protective device is a 2,500 A insulated-case circuit breaker with an 18-cycle (0.3 second) short-time delay. The incident energy at the working distance of 455 mm is 38 cal/cm² — within Category 4 but above the practical limit for most PPE category table entries. Arc flash mitigation at this location is achieved through zone-selective interlocking between the 2,500 A main and the 1,000 A feeder breakers, which reduces the clearing time for faults on the feeders to 4 cycles (instantaneous) and reduces the incident energy to 8 cal/cm² for the most common maintenance task locations.
The analysis in this paper connects to several companion studies in this library. Readers concerned with the upstream and downstream engineering will find Data Center Power System Design develops a closely related aspect of the same problem, while Case Study extends the treatment into an adjacent domain. For the broader methodological context, Emergency and Standby Power System Design provides complementary depth.
The 10 MW Tier III colocation design documented in this paper demonstrates how the owner's concurrent-maintainability and efficiency requirements translate into a complete, code-compliant electrical system, achieving a predicted PUE of 1.28 through high-efficiency double-conversion UPS with lithium-ion batteries and an optimized distribution architecture. The central engineering conclusion is that Tier III concurrent maintainability is an architectural property that must be designed into every distribution path from the utility service through the IT cabinet, because the requirement that any single component be removable for maintenance without interrupting IT power cannot be retrofitted onto a topology that was not designed for it. The arc flash analysis included in the case study reinforces that worker safety and system reliability are designed together, since the same protective-device decisions that establish coordination also govern incident energy. For the practicing engineer, the case study illustrates that a successful data center design is the disciplined propagation of a small number of top-level requirements — Tier level, PUE target, capacity — through every subsequent sizing and topology decision.
[1] Uptime Institute, Tier Standard: Topology, Uptime Institute, 2017.
[2] IEEE Standard 1584-2018, Guide for Performing Arc-Flash Hazard Calculations, IEEE, 2018.
[3] NFPA 70, National Electrical Code, Articles 517, 700, 706, 2023 edition, NFPA, 2023.
[4] ASHRAE TC 9.9, Data Center Power Equipment Thermal Guidelines and Best Practices, ASHRAE, 2019.
[5] The Green Grid, PUE: A Comprehensive Examination of the Metric, The Green Grid White Paper #49, 2012.
[6] IEEE Standard 446-1995, Recommended Practice for Emergency and Standby Power Systems, IEEE, 1995.
[7] NFPA 110, Standard for Emergency and Standby Power Systems, 2022 edition, NFPA, 2022.
[8] U.S. DOE, Data Center Energy Efficiency Strategies, DOE, 2024.