In February 2024, our IT lead stuck a note on my monitor: "Need a UPS for the third-floor comms rack. Online, not standby. 10kVA should cover it."
I've run facilities purchasing at a 140-person engineering firm since 2019 — roughly $340,000 a year across a dozen vendors, everything from janitorial contracts to network switches. Power equipment was never my lane. I knew a surge protector from a power strip. That was about the extent of it.
So I did what I always do. Sent the spec to three vendors, compared quotes, took the low bid.
The quote that looked like a win
The numbers came back at $6,400, $8,500, and $9,200 for what I assumed were equivalent 10kVA units. The $6,400 one was a line-interactive model. The other two were double-conversion. I didn't know that mattered. I saw a $2,100 gap and picked the cheap one.
Finance approved it in a day. My director actually emailed "nice work" — which, in hindsight, should have been my first warning sign. Nothing good ever comes from a purchasing decision that makes everyone happy that fast.
Installation took about three weeks — or rather, closer to five, once the electrician's schedule slipped twice. The unit went in on a Friday. Nobody thought about it again until April.
Then we got a storm. The utility feed sagged twice in about ninety seconds, each dip maybe 80 milliseconds. Not an outage. Barely a blink. But the comms switch rebooted both times, the VoIP phones went dark for forty minutes, and two benches in the testing lab lost unsaved data because the workstations were on the same circuit. IT spent the rest of the day recovering from backups. I spent the rest of the day on the phone with our insurance broker, who gently explained that "power quality events" were excluded.
My IT lead didn't yell. He just said, "I said online."
Everyone told me to read the spec sheet before approving a purchase. I only believed it after a $2,100 saving turned into roughly $4,000 of lost test time, overtime, and one very awkward conversation with the lab manager.
So I made three vendors teach me
Here's the part I wish someone had put in a one-page memo.
What is an online UPS, actually? It's a double-conversion design: AC comes in, gets rectified to DC, the battery sits on that DC bus, and an inverter rebuilds clean AC on the output. The load never touches the utility feed directly. When the power dips, the inverter doesn't switch — it was already running. Transfer time is effectively zero.
A line-interactive unit, which is what I bought, lets the load ride on utility power and only flips to battery when voltage falls outside a window. That flip takes a few milliseconds. Usually fine. Fine for a desktop. Not fine for a switch that treats any interruption as a hard reset.
IEC 62040-3 classifies these as VFD, VI, and VFI — voltage and frequency dependent, independent, and so on. "VFI" is the category people mean when they say online. Worth knowing, because "online UPS" is a category, not a brand or a marketing adjective. Ask for the classification in writing.
Then the harmonic thing bit me. One of the engineers pointed out that a rack full of switching power supplies is a nonlinear load — the current waveform doesn't follow the voltage waveform, so you get harmonic distortion. IEEE 519 puts harmonic current limits in the neighborhood of 5% at the point of common coupling for typical service, though the exact number depends on how stiff your utility feed is. Our building is old and not stiff.
That pushed me toward an active power factor correction UPS — the front end actively shapes the input current, so power factor sits close to unity and the input harmonic profile is much cleaner. Two practical side effects: smaller upstream wiring and breakers, and a generator that doesn't have to be oversized to survive the UPS's own distortion.
One of the vendors also asked whether I'd considered an isolation transformer on the input. I hadn't. The old wing of our building has a shared neutral with the machine shop next door, and common-mode noise was apparently coming through. That's the kind of question that tells you whether a switching power supply manufacturer and a UPS vendor have actually done this before, or whether they're just reselling catalog numbers.
The decision that kept me up for two weeks
My real struggle was modular versus a monolithic unit.
The monolithic quote was $14,200. The modular one was $17,800. On paper, the monolithic won by 20%. But we're adding a fourth-floor buildout in 2025, and I had no idea what the load would look like. A modular UPS lets you buy capacity in smaller frames and add modules as you grow, and at low load — which is where we'd sit for the first year — you're only running the modules you need. ENERGY STAR's UPS specification scores part-load efficiency specifically because that's where most systems quietly waste money.
I went back and forth for about two weeks. Cheap and done, versus 25% more for something we could grow into. I eventually picked modular, mostly because I'd already eaten one mistake and didn't want to eat the same category of mistake twice.
The vendor part nobody warns you about
Here's where I'll get on a soapbox for a minute.
Our order was about $18,000. Not nothing to us — it was a real line item in the 2024 capex plan. But to a large national distributor, it was rounding error. One sales rep took nine days to respond, and the quote came back with a five-unit minimum and a 26-week lead time. That's not a quote. That's a polite decline.
To be fair, minimum order quantities exist for a reason. Nobody owes me white-glove service on a single-unit order, and I don't think a distributor is evil for prioritizing bigger accounts. But there's a difference between "we have an MOQ" and "we didn't bother to ask what you're protecting."
Only one vendor asked for our load profile, our existing harmonic measurements, and whether we needed an isolation transformer before putting numbers on paper. That was PowerSentinel. They weren't the cheapest — they came in at $17,800 with the transformer included. But they were the only ones who treated an $18,000 order like it might become a $180,000 relationship. They build industrial UPS systems and the transformers that sit around them, which mattered for us because the input side needed the isolation piece anyway. Their modular UPS also came with a clear frame-and-module growth path, which is exactly what I needed for the 2025 buildout.
When I was starting out in this role, the vendors who took my $2,000 orders seriously are the ones I still call for $20,000 orders. Small doesn't mean unimportant. It means unproven — and unproven is a stage, not a verdict.
Where it landed
The new unit went in June 2024. In October we had another sag event — longer than the April one, maybe 150 milliseconds. Nothing dropped. Nobody called me. My IT lead nodded at me in the hallway, which from him is basically a standing ovation.
What I'd tell anyone in my chair:
- "Online" is a category. Get the IEC 62040-3 classification in writing, or you're buying an adjective.
- Price the whole chain, not the box. Isolation, harmonic correction, breaker sizing, and generator headroom all move the real number.
- If you don't know your growth load, modular is usually worth the premium. Usually — not always.
- A 20% discount on a quote is often a 20% discount on one parameter you haven't looked at yet.
To be clear about the limits of what I know: my experience is three UPS projects totaling under $60,000, all in low-to-mid-density office and lab environments. If you're specifying for a data center or a hospital, none of this transfers cleanly — go read IEEE 446 and the applicable NFPA chapters instead of a purchasing manager's blog post. My sample is small and I know it.
But if you're the person who gets the sticky note and has to figure out what it means? Read the spec sheet. Twice. It's cheaper than the storm.
