Start with the physics, because the whole confusion lives in one ratio. Watts are real power: what your electricity meter bills, what the server's power supply converts into heat and computation. Volt-amps are apparent power: voltage times current, which is what wiring, breakers and the UPS's electronics have to carry regardless of whether the load converts it usefully. The ratio between the two is the load's power factor. APC's FAQ on the subject defines it cleanly — "the ratio of the Watt to VA rating is called the 'Power Factor'" — and notes that for computer equipment the two "can differ significantly, with the VA rating always being equal to or larger than the Watt rating." For a resistive space heater they are equal. For a server with a modern PFC power supply they are close. For the sloppy mix of devices most people actually plug in, they are not.
The UPS industry sized for that uncertainty with a convention. APC again: "it is a standard in the industry that the Watt rating is approximately 60% of the VA rating, this being the typical power factor of common loads" — and, critically, "Neither the Watt nor the VA rating of a UPS may be exceeded." The sizing rule follows: keep the sum of your loads' nameplate ratings below 60% of the UPS's VA, "and it is impossible to exceed the Watt rating of the UPS." This is also why a naive "my rack is 900 VA, so a 1000 VA unit" calculation fails in the field: 900 VA of nameplate loads can exceed the ~600 W ceiling of a classic 1000 VA unit the moment measured watts drift above nameplate assumptions. APC's own documentation recommends its sizing tools precisely because nameplate ratings "are often in VA, which makes it difficult to know the Watt ratings."
The 0.6-to-1.0 spread: why "same VA" units are not the same
The 60% rule is a convention for the worst case, not a property of all hardware — and this is where 2026's buyer has an advantage the rule predates. Newer UPS designs ship with power factors of 0.9 or unity: APC's Smart-UPS VT documentation notes its sizing and runtime calculators "are based on Watt ratings" with VA converted using an approximation that assumes "low VA loads have a power factor of around 0.6 and high power loads have a power factor of nearly 1.0." Compare two APC units at the same nominal size: the older Smart-UPS 1500 VA class rated at 1000 W (PF 0.67) versus the SRT-class 1500 VA units rated at 1350–1500 W (PF 0.9–1.0). Same box size, same "1500 VA" label, 35–50% more real capacity. The practical corollary for the used market: a secondhand "1500 VA" unit from 2012 and one from 2022 are different products wearing the same number. Check the watt rating, not the VA, every time.
Your load's power factor matters less than it used to, which is a quiet win most sizing guides skip. Modern server and desktop power supplies ship with active PFC and draw at 0.95–0.99 — the measurement our home-lab power-bill post recommends (a plug-in watt meter) reads real watts directly, which is the number that must fit under the UPS's watt ceiling. The legacy 60% rule exists precisely for the cases where you cannot or did not measure. If you measured: size watts-to-watts with 20–30% headroom. If you did not: apply the 60% rule and accept an oversized, longer-running UPS as the price of not measuring — APC's own docs call the oversized result "a larger run time than expected," which is the rare failure mode that points the right way.
Runtime: the second number nobody reads correctly
Capacity sizing answers "will it hold"; runtime answers "for how long" — and the two come from different places. Runtime is set by the battery's stored energy, it falls steeply and nonlinearly as load rises, and the full-load runtime on the box (typically 3–7 minutes at capacity for consumer line-interactive units) is the worst data point on the label. Our arithmetic for the always-on home server case: a lab idling at 80 W — the realistic range our power-bill post measures for a small server stack — draws about 13% of a 600 W-rated unit's capacity, and in the light-load region runtime scales close to inversely with load, so the battery that holds 600 W for ~5 minutes holds 80 W for roughly 35–45 minutes. Two caveats keep that honest: it is an inverse-load approximation, not a curve fit — the vendor's runtime chart for your exact model is the authority — and a 3–4 year old battery delivers materially less than its rated curve, which is why runtime claims have a date stamp whether or not the chart prints one.
There is also a sizing inversion most first-time buyers get backwards. The goal for a home server is not long runtime on the UPS — it is enough seconds to ride through flickers plus enough minutes to shut down cleanly, with the UPS's USB/network signaling telling the host to power off. Long outages are the generator's problem at datacenter scale and "nobody's" problem at home scale; sizing a home lab for hours of battery runtime means buying battery capacity you will use a handful of times a decade. External-battery-capable units (the standard escape hatch on Smart-UPS-class hardware) exist for the cases where that math genuinely differs — a lab that must stay up — and cost proportionally. Everything else should size for graceful shutdown, which is a much cheaper UPS than the runtime fantasy suggests.
The checklist
One: measure the load in watts with a plug-in meter over a normal day — idle, backup jobs running, boot surge if you can catch it. Nameplate watts overstate real draw by 30–50%; measured watts are the number that binds. Two: read the UPS's watt rating and require load ≤ 70–80% of it. If only VA is known, apply the 60% rule from APC's guidance and treat the result as a ceiling, not a target. Three: check both limits — "neither the Watt nor the VA rating of a UPS may be exceeded," and whichever your load hits first is the real size. Four: pull the vendor's runtime chart for your model and your measured load; ignore the full-load number on the box. Add a battery-replacement date the day you install the unit, because the chart's clock starts then. Five: wire the shutdown signaling before you need it — a UPS that holds power for five minutes but cannot tell the server to shut down is a battery that dies loudly instead of cleanly. The network gear matters here too: the router and switch belong on the UPS even when the servers do not, or the shutdown command dies before the servers do — the same "chain fails at its weakest link" logic our network post applies to bandwidth, and the datacenter version of it is the ten-second rule. A home lab will never need Type 10 compliance, but "the backup chain itself fails more often than the grid" holds at every scale. Size it once, correctly, and the lights going out becomes a logging event instead of a recovery project.
Comments