The runtime figure printed on a UPS box — "provides 15 minutes of backup power" — is measured at a specific load the datasheet rarely states clearly, usually somewhere between 30% and 50% of the unit's rated capacity. Plug in equipment drawing closer to the unit's actual maximum, which is exactly what happens when someone sizes "close enough" instead of doing the arithmetic, and that 15 minutes can drop to 4 or 5 without anything being wrong with the UPS. Runtime isn't linear with load — it drops off faster the harder a battery is pushed — so the gap between the box's number and reality gets worse, not just proportionally smaller, the closer you run to capacity.

Watts vs VA: the spec that trips up most sizing

UPS capacity is rated in VA (volt-amps) and, separately, watts — and the watt figure is always equal to or lower than the VA figure, governed by a power factor that's typically around 0.6-0.7 for older UPS designs and closer to 0.9-1.0 for modern ones. A UPS rated "1500VA / 900W" can only deliver 900 real watts no matter what its VA rating suggests, and the equipment being protected draws watts, not VA — so sizing off the bigger, more prominent VA number on the box is a reliable way to end up with less real capacity than planned. The fix is simple once you know to look for it: always size against the watt rating, and if a UPS's listing only shows VA with no watt figure or power factor stated, assume the conservative end (0.6) rather than the optimistic one until you can confirm it.

Nameplate wattage isn't what equipment actually draws

The wattage printed on a device's nameplate or PSU label is a maximum rating, not a typical draw — manufacturers size that number for worst-case, fully-loaded, all-peripherals-active conditions, which real-world idle-to-moderate usage rarely reaches. Adding up nameplate wattages across every piece of equipment and sizing a UPS to that total produces a UPS that's comfortably oversized for actual draw, which isn't dangerous, just an avoidable extra cost. The more accurate approach is measuring actual draw with a plug-in power meter under real working conditions for the equipment in question, or — when that's not practical — using a server or device's published typical power consumption figure from its spec sheet rather than its maximum rated PSU wattage, which can overstate real draw by 30-50% or more depending on the hardware.

Runtime doesn't scale down linearly with load

Lead-acid battery capacity (the chemistry in the overwhelming majority of UPS units) follows Peukert's law: the effective capacity of a battery decreases as the discharge rate increases, meaning a battery delivers noticeably less total energy when drained quickly than the same battery drained slowly, even though its rated amp-hour capacity is a fixed number. In practice this means doubling the load on a UPS doesn't halve the runtime — it typically cuts it by more than half, because the battery is simultaneously being asked to deliver more current AND is doing so less efficiently at that higher current. This is the mechanism behind vendor runtime charts looking curved rather than a straight line when plotted against load percentage, and it's exactly why the datasheet's one headline runtime figure, measured at a specific moderate load, badly understates how much runtime degrades once a UPS is pushed toward its rated maximum.

The sizing math that actually holds up

Three numbers matter: total real watt draw of everything the UPS protects (measured or conservatively estimated, not summed nameplate maximums), the UPS's rated watt capacity (not VA), and the load percentage that results — rated watts divided into actual draw. Keeping that load percentage at 50-60% or below, rather than sizing a UPS to just barely cover present draw, buys back two things simultaneously: it sits on the flatter, more efficient part of the runtime curve described above, and it leaves headroom for adding equipment later without needing a whole new UPS. A UPS sized to exactly match today's load at 90%+ utilization is both delivering noticeably less runtime per the Peukert effect and has zero room to grow — the two problems compound rather than being separate concerns.

Worked example: a small server closet with two 1U servers measured at 180W each under real load, a switch at 60W, and a couple of access points at 15W each — 450W total actual draw. A UPS rated "1500VA / 900W" run against that load sits at exactly 50% utilization (450W ÷ 900W), which is the sweet spot described above. The same equipment on a UPS sized off the servers' 350W-each PSU nameplate rating instead — 850W total nameplate draw, rounded up "to be safe" to a 1000W UPS — would actually be running at 45% of real capacity while looking, on paper, like it was sized tightly. Nameplate-based sizing doesn't just waste money in this case; it can also mislead in the other direction, making a UPS look more comfortably sized than its real-world headroom actually is once battery age is factored in.

Battery age quietly erodes every number above it

None of this math holds for the life of the UPS — sealed lead-acid batteries, the standard chemistry in most UPS units, typically lose meaningful capacity after 3-5 years even with correct charging and moderate temperatures, and significantly faster in a warm server closet or garage, since battery lifespan roughly halves for every 8-10°C above the chemistry's rated temperature. A UPS correctly sized on day one, with a comfortable load percentage and a clean runtime margin, can quietly degrade to something closer to the "running near maximum" scenario described above purely from battery aging, with no change to the connected load at all — which is why a runtime self-test (most UPS units and their management software support scheduling one) every few months is worth doing, rather than finding out the real runtime during an actual outage.

Topology changes what "sizing" even means

Standby and line-interactive UPS designs — the cheaper, more common category for small deployments — stay on utility power most of the time and switch to battery within a few milliseconds of detecting a problem; that transfer time is small enough to be invisible to most equipment, but not all of it. Double-conversion (online) UPS units instead continuously convert incoming AC to DC and back to AC, so the connected equipment never actually sees a transfer at all — output is generated from the inverter at all times, utility power or not. The sizing math for watts, load percentage and runtime is the same across both, but the choice of topology changes what that runtime is actually protecting: a standby unit sized perfectly on paper can still let a brief voltage sag or transfer glitch through in a way a double-conversion unit, run at the identical load percentage, won't — which matters far more for equipment sensitive to even momentary power anomalies than for something that tolerates a clean, fast transfer without issue.

Inrush current catches sizing based on steady-state draw alone

Equipment with motors, compressors or large power supplies — network switches with many PoE ports powering up simultaneously, older spinning-disk arrays, anything with a transformer — can draw several times its steady-state wattage for a brief moment at power-on. A UPS sized exactly to steady-state draw with no margin can trip into overload protection or distort its output waveform at the exact moment everything is trying to restart after an outage, which is the worst possible time for it to fail. This is a second, independent reason the 50-60% load-percentage headroom recommended above matters: it's not just about runtime curve efficiency, it's also slack for the inrush spike that steady-state wattage math never accounts for in the first place.