Published: June 2026 Technical Level: Advanced Category: Power Systems Design
This case study documents the electrical power system design, construction, and commissioning of a 50 MW hyperscale data center in Phoenix, Arizona, designed to Uptime Institute Tier IV concurrent maintainability and fault tolerance standards. The design achieves a power usage effectiveness (PUE) of 1.22 at design conditions through a combination of liquid-cooled server infrastructure, modular uninterruptible power supply (UPS) architecture with eco-mode operation, and an integrated 20 MW solar PV plus 40 MWh battery energy storage system that reduces both energy cost and demand charges. This paper presents the engineering basis for the major design decisions: the 2(N+1) redundant electrical distribution topology, the UPS sizing and configuration, the generator plant design for extended-duration outages, the arc flash and protection coordination study for the medium-voltage distribution, and the commissioning methodology that verified Tier IV fault tolerance through integrated systems testing. Measured operational data from the first year confirm the design PUE within the expected seasonal range and validate the redundancy architecture against a documented utility outage event.
The facility serves a major cloud services provider requiring 50 MW of critical IT load with expansion capability to 75 MW, constructed on a 45-acre greenfield site in Phoenix. The Phoenix location imposes a demanding thermal environment — a 115°F (46°C) summer design dry-bulb temperature — that drives both the cooling system design and the derating of electrical equipment. The utility service is provided by Arizona Public Service at 230 kV transmission, with 40 kA of available three-phase fault current at the service entrance and an energy rate structure of 15/kW for demand.
The design criteria established four primary objectives: Tier IV reliability (99.995 percent availability, equivalent to 26 minutes of downtime per year), a design PUE no greater than 1.25, at least 30 percent of annual energy from renewable sources, and an 18-month schedule from design start to commissioning. The renewable energy and PUE objectives are coupled to the Phoenix climate: the high solar resource (approximately 6.5 kWh/m²/day average) makes on-site solar generation economically attractive, while the high ambient temperature makes cooling efficiency the dominant factor in achieving the PUE target.
Uptime Institute Tier IV certification requires that the facility be both concurrently maintainable (any component can be removed from service for maintenance without affecting the IT load) and fault tolerant (any single fault does not affect the IT load). These requirements are met by a 2(N+1) distribution topology: two completely independent electrical paths (the A path and the B path), each sized to carry the full IT load with N+1 redundancy in its own UPS and generator capacity.
The 230 kV utility service feeds two independent 230/13.8 kV, 40 MVA main power transformers. Each transformer feeds a 13.8 kV medium-voltage switchgear lineup (Switchgear A and Switchgear B) that are electrically isolated under normal operation. The two switchgear lineups distribute power to independent 13.8 kV/480V unit substations serving the A and B power paths to each computer room. Each IT rack is served by dual cord-connected power supplies, one from the A path and one from the B path, so that the loss of either complete path does not interrupt the IT load.
The UPS system provides ride-through power during the transfer from utility to generator following a utility outage, and conditions the power to the IT load against utility voltage and frequency variations. The design uses modular UPS units in an N+1 configuration on each power path. The original conventional design called for 48 UPS modules (24 per path) to provide 2N redundancy; a design optimization using AI-assisted load modeling determined that the actual diversity of the IT load permitted an N+1 configuration with 36 modules (18 per path), reducing the UPS count by 25 percent while maintaining the required fault tolerance.
The UPS modules operate in a high-efficiency eco-mode that bypasses the double-conversion path during normal operation when utility power quality is within tolerance, achieving 99 percent UPS efficiency compared to 96 percent for continuous double-conversion. The eco-mode transfer to double-conversion operation occurs within 2 milliseconds upon detection of a power quality event, fast enough to protect the IT load. The 3 percent efficiency improvement from eco-mode operation, applied across 50 MW of UPS throughput, represents approximately 1.5 MW of reduced loss — a direct and substantial contribution to the PUE target.
The generator plant provides extended-duration backup power for utility outages exceeding the UPS battery ride-through time. The plant consists of twelve 2.5 MW diesel generators arranged as two independent groups (six per power path), each group providing N+1 redundancy for a 12.5 MW path load (six generators rated 2.5 MW each, with one redundant).
The generators are sized with derating for the Phoenix altitude (1,086 ft) and the 115°F ambient design temperature. The combined altitude and temperature derating per the manufacturer's curves is approximately 8 percent, so the 2.5 MW standby-rated generators provide approximately 2.3 MW of usable output at site conditions. The N+1 group of six therefore provides of firm capacity with one generator out of service — adequate for the 12.5 MW path load when combined with the simultaneous reduction in cooling load that occurs during the brief generator-only operating periods.
The generator step-loading sequence is coordinated with the UPS battery discharge: the generators start and reach rated speed within 10 seconds, synchronize to the path bus, and accept the UPS rectifier load in steps that keep the generator frequency and voltage within the tolerance required by the UPS input. The cooling system loads are added after the critical IT load is secured on generator power.
The 13.8 kV medium-voltage distribution required a complete protection coordination study and arc flash hazard analysis per IEEE 1584-2018. The available fault current at the 13.8 kV switchgear is approximately 23 kA, producing incident energy levels at the switchgear of 18 to 32 cal/cm² depending on the protective device clearing time. The coordination study set the main breaker and feeder breaker relays to achieve selective coordination while minimizing clearing time, reducing the arc flash incident energy at the feeder positions to below 8 cal/cm² through the application of zone-selective interlocking and reduced instantaneous trip settings during maintenance (a maintenance switch mode that lowers the trip threshold when personnel are working in the switchgear).
The arc flash study produced labeled PPE requirements for each equipment location and informed the operational procedures for energized work, including remote racking of the 13.8 kV breakers to keep personnel outside the arc flash boundary during racking operations.
The 20 MW solar PV plant is interconnected at the 13.8 kV distribution level through dedicated inverters compliant with IEEE 1547-2018, configured to operate in a non-export mode that supplies the data center load but does not back-feed the utility. The 40 MWh battery energy storage system provides two functions: peak shaving to reduce the demand charges, and a fast-response reserve that supplements the UPS ride-through during the generator start sequence.
The combined solar and storage system reduces the facility's utility demand charges by approximately 67 percent by shaving the peak demand and shifting the timing of grid import. The renewable energy fraction achieved is approximately 32 percent of annual energy, meeting the 30 percent design objective. The battery dispatch is co-optimized with the cooling load: during high-solar periods, the battery charges from excess solar generation, and during the evening peak demand period, the battery discharges to reduce grid import.
The commissioning program followed an integrated systems test methodology in which the complete electrical and mechanical systems were tested together under simulated load using load banks representing the full 50 MW IT load. The Tier IV fault tolerance was verified by deliberately introducing single faults — opening individual breakers, simulating UPS module failures, tripping individual generators — and confirming that the IT load (represented by the load banks) experienced no interruption in any single-fault scenario.
The first-year operational data validated the design. The annual average PUE was 1.28, slightly above the 1.22 design value due to the seasonal cooling load in the Phoenix summer, but within the expected operational range. A documented utility outage event in the first year — a 47-minute loss of the 230 kV service following a regional transmission fault — was managed by the generator plant and UPS system with zero interruption to the IT load, confirming the redundancy architecture under actual outage conditions. The commissioning achieved zero critical deficiencies on the final integrated systems test, attributed to the early integration of commissioning into the design phase, which the project team estimated prevented the majority of startup issues that typically arise when commissioning begins only after construction completion.
The most consequential finding from the 50 MW build is that reliability and efficiency, long treated as opposing objectives in data center design, were reconciled by the same architectural choices rather than traded against each other. The 2(N+1) topology that delivers Tier IV fault tolerance also enabled the modular UPS to run in eco-mode at high part-load efficiency, because redundancy at the system level removed the need to operate each module conservatively. The design decision that bought reliability also bought the PUE.
The most common implementation failure in facilities of this scale is the generator plant that meets its rating on the datasheet but not on the site, because altitude and temperature derating were applied to the prime rating rather than to the standby duty the application actually imposes. A plant sized without that correction discovers its shortfall during the first hot-weather utility outage — the one moment the analysis exists to survive.
The engineer carrying this forward should next quantify the dynamic interaction between the renewable interconnection and the UPS during a utility disturbance, because the case study verified steady-state integration but the behavior that governs Tier IV availability is the sub-second transient — the coordination of the UPS ride-through, the generator start, and the renewable inverter's response — during the transfer that the commissioning test exercised only in isolation.
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 Data Center Power System Design extends the treatment into an adjacent domain. For the broader methodological context, Emergency and Standby Power System Design provides complementary depth.
[1] Uptime Institute, Data Center Site Infrastructure Tier Standard: Topology, Uptime Institute, 2018.
[2] IEEE Standard 1584-2018, IEEE Guide for Performing Arc Flash Hazard Calculations, IEEE, 2018.
[3] IEEE Standard 1547-2018, Standard for Interconnection and Interoperability of Distributed Energy Resources, IEEE, 2018.
[4] NFPA 70, National Electrical Code, Articles 645, 700, and 705, 2023 edition, NFPA, 2023.
[5] ASHRAE, Thermal Guidelines for Data Processing Environments, 5th ed., ASHRAE TC 9.9, 2021.
[6] NFPA 110-2021, Standard for Emergency and Standby Power Systems, NFPA, 2021.
[7] IEEE Standard 446-1995 (Orange Book), IEEE Recommended Practice for Emergency and Standby Power Systems, IEEE, 1995.
[8] The Green Grid, PUE: A Comprehensive Examination of the Metric, The Green Grid White Paper 49, 2012.