UPS sizing is not a matter of adding equipment wattages and choosing the next catalogue model. The UPS must satisfy two limits at the same time: real power, expressed in kilowatts, and apparent power, expressed in kilovolt-amperes. It must also tolerate transients, support the required redundancy arrangement, leave sensible capacity for foreseeable growth and recharge its batteries without destabilising the upstream source. Battery autonomy then introduces a second sizing problem because runtime depends on load, battery characteristics, temperature, age and the manufacturer’s discharge data. A good calculation makes assumptions explicit. That is more valuable than applying a generic “add 25 per cent” rule that may either waste capacity or leave an important edge case unexamined.
Start with a load schedule, not a UPS model
Create a protected-load schedule that identifies each item or load group, its normal demand, maximum expected demand, power factor where known, starting or inrush behaviour, redundancy status and operational priority. For existing sites, measured data from reliable metering is often more useful than nameplate totals because nameplates commonly represent maximum or input ratings rather than typical simultaneous demand. Measurements should cover meaningful operating conditions, not only a quiet snapshot.
For a new facility, the schedule needs engineering judgement about diversity. Some loads will never reach nameplate simultaneously, while other process or clinical loads may need conservative treatment. Separate loads that genuinely require no-break power from those that can ride through generator transfer or be shed. This can reduce UPS size and simplify resilience. Document every exclusion. If a supposedly non-critical load later migrates onto the protected distribution without the design being updated, hidden capacity erosion can result.
Understand kW and kVA
Real power in kW represents the power that performs useful work. Apparent power in kVA reflects both real power and the effects of the load’s current and voltage relationship. In a simplified single-load case, power factor is the ratio of kW to kVA. A 90 kW load at 0.9 power factor therefore corresponds to 100 kVA. Modern UPS products are often rated with an output power factor close to unity, but the exact kW and kVA limits must still be checked.
Do not assume that a 100 kVA UPS can always deliver 100 kW. Confirm the manufacturer’s continuous ratings at the relevant operating conditions. Temperature and altitude can also affect capability. For three-phase systems, engineers should calculate from the actual phase currents and voltage conditions and consider imbalance. The procurement schedule should present both kW and kVA demand so that a unit is not inadvertently selected against only one of the two constraints.
Allow for transient and inrush behaviour
Steady-state demand is only part of the requirement. Motors, transformers, imaging equipment, power supplies and other devices can draw transient currents or behave differently during energisation. The UPS must either support that event, coordinate with another starting method, or be arranged so the load is not started from the inverter under conditions the UPS cannot tolerate. Manufacturer overload curves are therefore important.
For IT environments, the issue may be less about traditional motor starting and more about rapid changes in server demand, power-supply characteristics and downstream fault clearing. In industrial settings, large drive systems and transformers can create different challenges. State the expected load-step and overload requirement in the design brief. If a critical process requires a particular piece of equipment to restart while the site is still on UPS or generator power, test that scenario during commissioning rather than assuming that a steady-state calculation proves it will work.
Choose headroom deliberately
Headroom protects against uncertainty and growth, but more is not automatically better. A very lightly loaded UPS can represent unnecessary capital, space and maintenance burden, and efficiency may be lower away from the optimum operating range. Conversely, a design that starts close to its continuous limit leaves little room for real-world variation or future additions.
Separate three concepts: measurement uncertainty, planned growth and redundancy reserve. If the system is modular, some future growth may be met by adding power modules rather than installing the final capacity on day one. In an N+1 arrangement, the redundant module is not simply “spare growth capacity”; using it for normal growth can remove the intended fault tolerance. Write the capacity rules into operational governance so that later changes are assessed against both loading and resilience, not just whether the UPS appears to have unused kVA.
Size redundancy as an architecture
Suppose a modular UPS requires four power modules to carry the design load. An N+1 arrangement uses a fifth module so that the load can remain supported if one module is unavailable, subject to the architecture and common components. A 2N arrangement creates two independent full-capacity paths. These labels describe concepts, not guarantees. Shared switchgear, battery systems, control logic, bypass sources or distribution can reintroduce common failure points.
When sizing, calculate the load each path must carry in normal and contingency states. In an A/B data-centre arrangement, for example, a dual-corded load may be shared across paths in normal operation, but each path may need enough capacity to support the transferred or surviving demand after a failure. The “normal” percentage loading can therefore be misleading unless contingency loading is also shown. Include maintenance states: can one path be isolated while the remaining architecture still satisfies the agreed resilience objective?
Translate runtime into an operating scenario
Battery runtime should answer a scenario rather than a marketing question. If a generator normally starts within seconds, why is ten or fifteen minutes of autonomy required? There may be good reasons: repeated start attempts, switchgear sequencing, stabilisation time, operator intervention or a required safety margin. For a remote telecom site, the target may instead be hours because generator support or site access is limited.
Define autonomy at a stated load, ambient condition, battery end-of-life assumption and end voltage. Do not estimate large-system runtime by dividing nominal battery watt-hours by load; battery discharge behaviour, UPS efficiency, DC voltage limits and ageing make manufacturer-supported calculations necessary. If future growth is expected, decide whether the original battery system must provide the target runtime at future load or whether autonomy will be reviewed as capacity expands. State the requirement clearly enough that competing suppliers calculate on comparable assumptions.
Check the upstream electrical source
A UPS is also a load on the upstream system. The rectifier input, harmonic profile, power factor, battery recharge demand and response to generator voltage or frequency changes affect the source. Modern power-electronic UPS designs can present much friendlier input characteristics than older rectifier technologies, but the actual data should be requested. Generator compatibility cannot be solved by a single oversizing ratio.
After a mains failure, the generator may have to accept the protected load and simultaneously supply battery recharge, mechanical services and other essential circuits. A staged recharge limit or programmable input-current limit may help. Coordinate with the generator designer and test the combined sequence. Also consider protective-device selectivity and fault current. Inverter-limited fault current may not behave like utility fault current, so downstream protection and static bypass behaviour need to be understood as part of the sizing and coordination exercise.
A worked sizing logic
Consider a hypothetical protected load measured at 72 kW with a maximum observed apparent demand of 80 kVA. The organisation expects a validated 15 kW of additional critical load within three years. It also requires the system to tolerate a defined short-duration load step and to retain one power module as genuine redundancy. The correct selection is not simply 72 kW plus 25 per cent. The designer first establishes the future design demand, checks the corresponding kVA at the expected power factor, then selects a modular configuration in which the required number of active modules can carry that design load while one module remains unavailable.
The same process is applied to batteries. If the operational requirement is 12 minutes at the future design load after an agreed end-of-life derating, the supplier should size and document the battery against that condition. The project team can then compare proposals on the same basis. This is the practical value of a structured sizing calculation: it turns “large enough” into a transparent set of electrical and operational requirements.
Standards and official guidance may be amended. Confirm the edition and project-specific requirements with a competent professional before design, procurement or maintenance work.