A data center UPS has one job that nothing else in the building can do: hold the critical load up during the seconds when the utility is gone and the generator has not yet taken it. Everything else it does — conditioning, isolation, metering — is secondary to that ride-through obligation. This guide is the entry point to the UPS systems catalog and to the four detailed guides beneath it, and it is written to be checked against the standards rather than taken on trust.
Key Takeaways
- The UPS sizing question is a kW question, not a kVA question, and the two diverge whenever output power factor is below unity.
- Topology is classified formally by IEC 62040-3 as VFI, VI or VFD; marketing names map onto those three classes and nothing else.
- The energy store, not the converter, sets floor loading, room temperature limits, fire code obligations and refresh cost.
- Redundancy configuration usually pushes a UPS to part load, and part-load efficiency is where the operating cost of a resilient design actually shows up.
- The secondary market for UPS equipment is more selective than for generators, because battery condition, capacitor age and firmware support all age independently of the frame.
What the UPS is actually for#
The ride-through window#
The design case is not a long outage. It is the interval between utility loss and the standby generator accepting load, which for a correctly maintained diesel set is a matter of seconds rather than minutes. The UPS covers that interval, and the battery or flywheel behind it is sized against that interval plus a stated margin.
Runtime beyond the generator start window is a separate decision with a separate justification. Sites without generation, sites with a graceful-shutdown requirement, and sites carrying a contractual hold time all buy runtime for reasons that have nothing to do with ride-through. Confusing the two is the most common way a battery room ends up two or three times the size the resilience case required.
Power conditioning as the second job#
A double-conversion UPS regenerates the output waveform rather than passing the input through, which is why it isolates the load from most upstream disturbance. That isolation is real and worth paying for in a data hall. It is also the reason the topology carries a conversion loss that eco-mode operation is designed to avoid, and the reason eco-mode is a deliberate trade rather than a free efficiency gain.
Where the UPS sits in the power chain#
Upstream sits the utility service, the transfer arrangement and generation. Downstream sit the distribution boards, remote power panels and busway and finally the rack PDUs. Between the UPS output and dual-corded load, static transfer switches are what let a single-corded device sit on a two-path design at all. The whole arrangement is only as concurrent as its least concurrent element, which is the point N+1 and 2N redundancy planning exists to make.
UPS topologies#
The three classes that matter#
IEC 62040-3 defines UPS performance by output dependency: VFI (voltage and frequency independent), VI (voltage independent) and VFD (voltage and frequency dependent). Every commercial topology is one of those three. Double conversion is VFI, line-interactive is VI, and standby is VFD.
Double conversion dominates data center use because it is the only class that holds output voltage and frequency independent of input under normal operation. Line-interactive equipment is common at the edge and in network closets, where the load is smaller and the tolerance wider. Standby units belong at the desk, not in the data hall.
Eco-mode and the efficiency argument#
Eco-mode runs the load on bypass with the inverter available, which recovers most of the conversion loss at the cost of a transfer event when the input goes out of tolerance. Manufacturers have narrowed transfer times substantially, and modern implementations detect and transfer fast enough for most IT power supplies to ride through. Whether that is acceptable is a site decision, and it belongs in a written operating mode policy rather than in a commissioning default.
The honest framing is that eco-mode moves risk from the electricity bill to the transfer event. Sites with well-behaved utility feeds and modern loads often take that trade. Sites with a dirty feed or legacy equipment generally should not.
Rotary and flywheel machines#
Rotary UPS and diesel rotary systems store energy mechanically and, in the diesel rotary case, integrate the engine into the same machine. They occupy a narrow but real niche at large single-block ratings, where the footprint and the absence of a battery room outweigh the mechanical maintenance burden. Flywheel storage is treated in detail in the UPS energy storage comparison.
The full topology comparison, including where each class fails and what the transfer behaviour looks like under fault, is in UPS topologies explained.
| Class (IEC 62040-3) | Common name | Output held independent of input | Typical data center role |
|---|---|---|---|
| VFI | Double conversion / online | Voltage and frequency | Primary critical load |
| VI | Line-interactive | Voltage only | Edge, network closet |
| VFD | Standby / offline | Neither | Workstation, non-critical |
Battery chemistry and stored energy#
VRLA#
Valve-regulated lead acid remains the reference case: lowest purchase price, highest footprint, shortest service life, and a maintenance and testing regime set out in IEEE 1188. The string is a consumable inside the life of the UPS, so the replacement cycle belongs in the acquisition case rather than in the operating budget.
Float temperature is the variable most often ignored. Elevated room temperature shortens VRLA life materially, which is why the battery room specification and the UPS specification cannot be written independently.
Lithium-ion#
Lithium-ion UPS strings — NMC and increasingly LFP — buy cycle life, calendar life, weight and footprint at a higher unit price. They also bring a battery management system, a thermal design requirement and a fire code conversation that VRLA does not, governed in the United States by NFPA 855. The installation cost of that conversation is a real line item, not a formality.
The general guidance for battery selection and sizing in UPS applications sits in IEEE 1184, which is the document to hand a vendor when their sizing worksheet is doing something you cannot reconstruct. Chemistry-specific maintenance obligations then split between IEEE 450 for vented lead-acid and IEEE 1188 for VRLA.
Flywheel#
A flywheel carries the load for a short, well-characterised interval and does it without a chemistry that degrades on a calendar. That makes it a strong fit where the only requirement is bridging to a generator that reliably starts, and a poor fit where any extended hold time is required. The trade-offs against both battery chemistries are set out in the energy storage comparison and priced in the UPS pricing index.
Sizing a UPS system#
kW, kVA and power factor#
The load is a kW quantity. The UPS nameplate is often a kVA quantity, and the two are separated by the output power factor. Most current three-phase machines are rated at unity output power factor, where the numbers coincide; a great many installed machines are not, and that is where like-for-like replacement quietly oversizes or undersizes the room.
Runtime#
Runtime is set by the generator start time plus a stated margin, unless a separate hold-time requirement exists. State the margin explicitly and state the end-of-life condition it must hold at — a string that meets runtime when new and misses it at 80% capacity has not met the requirement.
Redundancy and part load#
An N+1 or 2N arrangement means the installed capacity exceeds the load by design, so each unit runs below its rated output. Double-conversion efficiency falls away at low load, and a 2N design can sit at a load point where that penalty is significant. The efficiency curve at the actual operating load point, not the headline efficiency figure, is the number to compare across bids.
Module size drives the cost of redundancy more than frame size does, which is the argument behind modular versus monolithic frame architecture. The full method, including the power factor tolerance window and worked examples, is in how to size a UPS.
| Input | Source | Common error |
|---|---|---|
| Critical load (kW) | Measured or IT-provided, not nameplate sum | Summing nameplate ratings |
| Output power factor | UPS datasheet | Assuming unity on legacy frames |
| Runtime | Generator start time plus margin | Sizing to a hold time nobody required |
| Redundancy | Availability requirement | Ignoring the resulting part-load point |
| End-of-life capacity | Battery datasheet | Sizing at beginning-of-life capacity |
Where BESS fits, and where it does not#
Grid-scale battery energy storage and a UPS solve overlapping but distinct problems. A UPS is a sub-cycle transfer device sized in seconds and located inside the critical power path. A BESS is an energy asset sized in hours and located at the service, where it can also serve peak shaving and demand response obligations that a UPS cannot.
Storage cost and performance benchmarks published by NREL and the U.S. Energy Information Administration are the right reference point for the energy-asset case, and they are not interchangeable with UPS pricing. Where a BESS displaces generator runtime rather than UPS ride-through, the comparison is against fuel and maintenance, not against a battery string. Treating the two as substitutes is how projects end up with an oversized BESS and an under-specified critical path.
The major platforms#
The installed base at data center scale is concentrated in a handful of platforms: Vertiv Liebert, Schneider Electric and APC, Eaton, ABB, Mitsubishi Electric, and Piller and Hitec on the rotary side. Model-level specifications, ratings and configuration data for the platforms we hold are in the UPS systems catalog, and the broader distribution picture is in the guide to switchgear, transformers and distribution inventory.
Platform choice constrains service more than it constrains performance. Parts availability, firmware support policy and the local service network differ more between manufacturers than the converters do, and those differences are what actually govern a fifteen-year holding period.
Buying UPS equipment on the secondary market#
Why the used market is more selective#
A generator ages as one machine. A UPS ages as three: the frame and power electronics, the DC capacitors, and the energy store — each on its own clock, and each replaceable on its own schedule. A frame with low hours and a tired capacitor bank is not a bargain, and a nameplate rating tells you nothing about which of the three you are buying.
What is worth buying used#
Large static frames, flywheel systems and distribution equipment hold value well and are routinely redeployed. The economics improve further where the acquisition includes a service history and a manufacturer support path. Where those are missing, price the refurbishment before pricing the asset.
What to inspect#
Ask for capacitor age and replacement records, battery capacity test results under IEEE 1188 or IEEE 450, firmware revision and support status, and the current service contract position. A capacity test result is worth more than any statement about age. Where the seller cannot produce one, the string should be treated as a replacement item in the acquisition case.
Comparable equipment, indicative ranges and lead-time context sit in the UPS pricing index, and battery-side inventory in the energy storage inventory. For the surrounding electrical scope — switchgear, transformers, distribution — see the switchgear catalog.
Frequently asked questions#
What is a data center UPS? An uninterruptible power supply is the device that carries the critical IT load without interruption when utility power fails, holding it up until a generator accepts load or the site shuts down cleanly. In a data hall it is almost always a double-conversion (VFI) machine.
How long does a data center UPS run on battery? Long enough to cover generator start plus a design margin, which is a matter of seconds to a few minutes rather than hours. Longer hold times are a separate requirement and are sized separately.
What is a double conversion UPS? A UPS that rectifies incoming AC to DC and inverts it back to AC continuously, so the output waveform is regenerated rather than passed through. IEC 62040-3 classifies it as VFI — output voltage and frequency independent of the input.
Should data centers use lithium-ion or VRLA batteries? Lithium-ion costs more up front and returns it in cycle life, calendar life, weight and footprint; VRLA is cheaper to buy and is a consumable inside the life of the UPS. The deciding factors are usually floor loading, room temperature and whether the NFPA 855 installation requirements are already satisfied.
How do you size a UPS for a data center? Start from measured critical load in kW, apply the machine''s output power factor to reach a kVA rating, set runtime from generator start time plus margin at end-of-life battery capacity, then apply the redundancy configuration and check the resulting part-load efficiency point.
What is the difference between a UPS and a BESS? A UPS is a sub-cycle device inside the critical power path, sized in seconds. A BESS is an energy asset at the service, sized in hours, and useful for demand charges and load shifting that a UPS cannot address.
How much does a data center UPS cost? Pricing varies by rating, topology, redundancy and storage type more than by manufacturer. Current indicative ranges and lead times are maintained in the UPS pricing index.
Can you buy a used UPS? Yes, and large static and flywheel systems are routinely redeployed. The condition assessment has to cover the capacitor bank, the energy store and firmware support separately, because those three age independently of the frame.