The two numbers, and why one of them lies by omission
Every PoE switch publishes two power figures: maximum per port (15.4W af, 30W at, up to 90W bt) and total PoE budget — the sum the power supply can deliver across all ports at once. Datasheets print both; buyers read the first. The second is the one that blacks out cameras in January. A 24-port switch with 30W ports and a 250W budget can fully power eight 30W devices, or twenty-four 10W devices, or any mix summing to 250W — the per-port figure is a ceiling per socket, not a promise multiplied by ports.
Serious datasheets make this easy to check: ONV's L3 flagship, for example, lists 400W and 600W budget versions of the same 24-port switch, because the manufacturer knows the same switch serves a fixed-camera floor (low sum) and a PTZ floor (high sum). Choose the version by your arithmetic, not by hope.
The method: list, worst-case, sum, headroom
Step one: list every powered device with its worst-case draw — for cameras, the night figure with IR and heaters on; for PTZ, motors moving; for APs, all radios at full load. Manufacturer datasheets state maximum consumption; use it, not the idle figure from a review. Step two: sum. Step three: multiply by 1.25 — headroom for the devices you will add, the cable losses on long runs, and the optimistic edge in every nameplate. Step four: compare against the switch's total budget and buy the version that clears it.
Worked example — 24 fixed cameras: nameplate max 12W each (IR on). Sum: 288W. With headroom: 360W. A 250W budget switch fails this design on paper (and would fail it in winter, intermittently, at 3 a.m.). The 400W version clears it. The same 24 ports with 7W cameras sum to 168W → 210W with headroom → the 250W version is correct and the 400W is waste. The arithmetic takes ten minutes and decides purchases worth thousands.
The drains people forget
Extenders consume budget to repeat it — a far-end cluster behind an extender draws switch watts plus the extender's own overhead, all from one source port, which also caps the cluster at that port's class. Splitters convert with losses. Long cable runs waste a slice as heat. And budgets interact with temperature: a switch in a hot cabinet delivers its rating less happily than the lab unit — another reason the 1.25 factor is a floor, not a luxury.
Finally, growth is a power event. The spare ports you wisely left will be filled by someone, someday, with devices nobody budgeted. If the estate has any expansion habit at all, buy the bigger-budget version of the switch while the model is in production; power supplies are rarely upgradeable later.
Budgeting the network you will actually have
One more layer of realism: the budget you compute today describes the network on commissioning day, and networks only grow. The camera count creeps, the spare ports fill, the APs gain radios. Handle this structurally rather than hopefully. Where a model offers budget versions (the 400W/600W choice on ONV's 24-port L3 model is the type specimen), the price step between versions is usually trivial next to a mid-life switch replacement — take the headroom when the device list has any growth story at all. Where it does not (a fixed, sealed design), the 1.25 factor already covers honesty.
And put the arithmetic in the project file: the device list, the worst-case figures, the sum, the factor, the chosen budget. Budgets fail silently and are diagnosed expensively; a one-page power schedule turns a 3 a.m. January fault into a ten-minute lookup, and it is the document your successor will thank you for. The switches will be replaced someday; the arithmetic discipline should be permanent.
The version question: when two budgets share one model number
Manufacturers increasingly sell one switch in two power versions — same ports, same firmware, different internal supply (ONV's L3 24-port in 400W and 600W versions is the pattern). Buyers read this as a pricing trick; it is actually the budget method made physical, and it deserves a decision rule. If your computed requirement (with the 1.25 factor) lands within 10% of the smaller version's ceiling, buy the larger: the margin you calculated will be spent by the first unlisted device, guaranteed. If it lands below 70% of the smaller ceiling and the design is genuinely frozen — fixed cameras, no growth story — the smaller version is the honest buy and the saving is real.
The deeper point is what the version's existence confesses: the manufacturer knows its switch serves PTZ floors and fixed-camera floors whose budgets differ by hundreds of watts, and has priced the difference honestly as copper and silicon in the power supply rather than as a different model name. Treat the choice with the same honesty in return — as an arithmetic result, recorded next to the device list that produced it.
The decision, compressed
The whole guide fits in one sentence: sum every device's worst-case watts, multiply by 1.25, and buy a switch whose published total budget clears the result — while its per-port class covers the single hungriest device you own. Everything else — the versions choice, the night draw, the extender losses, the growth headroom — is that sentence applied carefully. If a quotation reaches you without the device list and the sum attached, it is not a quotation; it is a guess with a price on it. Do the ten minutes of arithmetic, file it, and your cameras will still be on in January.
One last practical note on procurement language: put the budget requirement in the purchase specification itself, as a number with its derivation — 'total PoE budget ≥ 400W, derived from the attached device schedule at worst-case draw × 1.25' — rather than as a model name. Model names get substituted; numbers get checked. A substitution that clears the stated budget is a legitimate equivalent; one that quietly halves it is a defect the paperwork can catch before the winter does. This single habit — specifying the arithmetic, not the artefact — generalises across every guide on this site, and it is the cheapest engineering control a buyer owns.
And if you inherit an estate with no power schedule at all — the common case — the recovery is a day's work with a clamp meter and a ladder: record each switch's port loads at night, at duty, and rebuild the sums from measurement rather than nameplate. Measured numbers end arguments that nameplates start, and the schedule you produce becomes the baseline every future change is priced against. Estates are rarely re-documented on a schedule; they are re-documented once, properly, after the first avoidable outage — or, in the disciplined ones, before it.