Most UPS sizing errors are made in the first hour of a project, when a load figure of unknown provenance is multiplied by a growth factor of unknown origin. The result is either a unit that cannot carry the design load, or one that spends its life at 30% load where its efficiency curve is at its worst. This is the sizing method behind the model data in the UPS systems catalog, written to be checked rather than trusted.
This guide is one of four spokes beneath the data center UPS and energy storage hub, which sets out how topology, storage, sizing and frame architecture fit together. Read it alongside modular vs monolithic frame architecture and UPS energy storage compared.
Key Takeaways
- kVA and kW are separated by the power factor. A 225 kVA nameplate at 0.90 output power factor is a 202.5 kW machine, and buying to the kVA figure oversizes the replacement by 11%.
- Most current three-phase machines are rated at unity output power factor, where kVA and kW are numerically identical. Legacy units frequently are not, which is where like-for-like replacement goes wrong.
- The load power factor tolerance window is a real selection criterion on mixed mechanical and IT loads, and the windows differ widely between machines.
- Build the load from metered readings, then add committed load, growth and unprotected load as three separate visible lines.
- Redundancy changes the equipment purchased, never the load number. Decide N, N+1 or 2N before the frame, because 2N doubles the electrical room.
kVA and kW are separate quantities#
kVA is apparent power and kW is real power. The ratio between them is the power factor, and it is a property of the load and of the UPS output rating together. A 0.9 power factor rating means a 200 kVA unit delivers 180 kW.
Most current three-phase machines are rated at unity output power factor, where kVA and kW are numerically the same. The Schneider Electric Galaxy VL is rated 200 to 500 kW at unity, and the Vertiv Liebert EXM2 covers 100 to 250 kVA/kW at unity. Older machines frequently are not.
225 kVA at 0.90 output power factor is 202.5 kW — the Mitsubishi Electric 9900AEGIS publishes its ratings on that basis, so a like-for-like replacement bought to the kVA nameplate arrives 22.5 kW larger than the requirement, an 11% oversize.
This is the single most common error in a legacy replacement. A 225 kVA nameplate on the old machine describes a 202.5 kW requirement, and the specification should carry the kW figure through.
Table 1 — Output power factor across current and legacy machines#
| Model | Output power factor | Published rating | kW at that rating |
|---|---|---|---|
| Schneider Electric Galaxy VL | Unity | 200–500 kW | 200–500 kW |
| Vertiv Liebert EXM2 | Unity | 100–250 kVA | 100–250 kW |
| Mitsubishi Electric 9900AEGIS | 0.90 | 225 kVA | 202.5 kW |
| Mitsubishi Electric 7011A Series | 0.70 load pf | 10.0 kVA | 7.0 kW |
The last row is the widest gap in the table and the easiest to miss: single-phase equipment can sit much further from unity, and a 10.0 kVA unit is a 7.0 kW unit.
Check the load power factor window#
A unity-rated UPS still has to tolerate the actual load power factor. Modern IT power supplies sit close to unity; chillers, pumps and older equipment do not. Manufacturers publish a tolerance window, and the widths differ enough to decide a selection.
Table 2 — Load power factor tolerance windows#
| Model | Tolerance window without derating |
|---|---|
| Schneider Electric Galaxy VL | 0.5 leading to 0.5 lagging |
| Schneider Electric Galaxy VS | 0.7 leading to 0.7 lagging |
| Vertiv Liebert EXL S1 | 0.7 leading to 0.4 lagging |
0.5 leading to 0.5 lagging — the Galaxy VL's window is the widest of the three, which matters on a mixed mechanical and IT load where the aggregate power factor moves through the day.
Building the load number#
Work from measured data wherever it exists. Nameplate ratings on IT equipment are worst-case figures and typically overstate real draw by a wide margin, which is how rooms end up with UPS running far below their efficient band. Use metered readings at the PDU or the existing UPS, taken over a period that includes peak activity.
Then add three things, separately and visibly:
- Known committed load not yet installed.
- A growth allowance stated as a percentage with a date attached.
- Any load that must ride through but is not currently on protected power.
Keep them as separate lines so the assumptions can be revisited without rebuilding the model. Facility-level growth assumptions can be sense-checked against the Lawrence Berkeley National Laboratory data center energy usage report and the US Department of Energy's data center and server program material. For a first-pass view of how the sized load interacts with generation and grid timing, the Power Intelligence tool models the same inputs at project scale.
Redundancy changes the frame#
N is the capacity that carries the load. N+1 adds one redundant unit or module. 2N duplicates the whole system. The load number is identical across all three; the equipment purchased is not.
On a modular frame, redundancy is priced in module steps. The ABB DPA 250 S4 installs 300 kW as six 50 kW modules and is rated 250 kW N+1, so the redundancy costs one module. On a monolithic pair, redundancy costs a whole unit: two Eaton Power Xpert 9395P frames sized above the load each carry it alone.
Decide the scheme before the frame. A 2N decision doubles the electrical room; a modular N+1 decision does not. Where the capital difference between schemes is the deciding factor, current benchmarks by rating are in the data center UPS pricing guide.
Runtime is a separate specification#
Autonomy is set by the energy store, so it is specified apart from the converter. Where standby generation exists and is tested to NFPA 110, the UPS only has to cover the start and transfer sequence. Where it does not, the store has to cover an orderly shutdown, which is a much larger number.
That dependency runs in one direction: the UPS sizing assumption rests on a generator being available on the date the room goes live, so check current generator lead times and the wider backup power lead times before the runtime figure is set.
Quote runtime at end of battery life and at the design load. A figure at beginning of life and half load describes a different machine. Cross-check the tested efficiency curve at the resulting load point using ENERGY STAR data, because a unit sized with too much headroom pays for that headroom every hour it runs.
Before the order goes out#
Confirm the classification code and test basis under IEC 62040-3. Confirm input voltage and any transformer requirement. Confirm the room can take the frame under NFPA 70 clearances, since working space and overcurrent protection are difficult to revisit once the layout is fixed.
Cooling load follows from the efficiency figure, so size it against the ASHRAE Datacom series using the real number at the real load point. Finally, sanity-check the resilience assumption against the Uptime Institute annual outage analysis, which continues to identify power as a recurring cause of significant site outages.
FAQ: UPS sizing#
How do you convert kVA to kW for a UPS?#
Multiply the kVA rating by the output power factor. At unity they are the same number. At 0.90, a 225 kVA unit is 202.5 kW. At 0.70, a 10.0 kVA unit is 7.0 kW. The power factor is on the datasheet, and it is the number to check first on any legacy machine being replaced.
How much headroom should a UPS have?#
Enough to carry committed load and a dated growth allowance, and no more. Headroom is not free: a unit running at 30% load sits at the poor end of its efficiency curve and pays for the unused capacity every hour. Size to the documented load plus stated growth, and revisit the growth line on its date instead of building it into the frame permanently.
What is the difference between N, N+1 and 2N?#
N is the capacity that carries the load. N+1 is N plus one redundant unit or module. 2N is two complete systems, each able to carry the load alone. The load figure does not change between them — only the equipment purchased does, and the electrical room that has to house it.
Should you size a UPS from equipment nameplate ratings?#
No. Nameplate ratings on IT equipment are worst-case figures and typically overstate actual draw by a wide margin. Metered readings at the PDU or the existing UPS, taken across a period that includes peak activity, are the correct basis. Nameplate-based sizing is the most common route to a permanently underloaded room.
How long should a UPS run on battery?#
Long enough to cover the generator start and transfer sequence, where standby generation exists and is exercised to NFPA 110. Without generation, the store has to cover an orderly shutdown of the protected load, which is a substantially larger figure. Quote the requirement at end of battery life and at design load.
Next step#
Filter by rating, output power factor and load power factor window in the UPS systems hub to shortlist frames against a sized load.