Why I Over-Bought 200 kVA of UPS Capacity — and Why Capacity Wasn't the Real Problem

Four years ago I signed off on an invoice for roughly $47,000 — no, $52,000 once you count the battery replacement (that one still stings). Three online UPS units, a three-phase setup, external battery strings. I'd run the sizing calculation twice, walked the load list three times, and left a 25% margin because I'd already been burned once in 2017 on a smaller job. On paper it was a clean design.

Six months later, the facility was running at 60% load. A scheduled battery test that was supposed to hold for 15 minutes dropped at 8 minutes 40 seconds. One of the 300 kVA units was sitting at 82% capacity for most of the afternoon with efficiency noticeably below what we'd quoted to the client. The cooling loop couldn't keep up. And I had to explain to a very patient project manager — for the third time that quarter — that no, actually, we had not measured "load" the way the UPS measures it.

That's the part nobody warns you about. You size the system, you check the box, you move on. Then the real world shows up.

The Problem You Think You Have Isn't the Problem

If you've ever looked at industrial UPS systems and thought "the sizing calculator gave me a number, so I'm done," I get it. Most sizing guides tell you to add up your loads, apply a growth factor, pick the next standard size up. It reads like a solved problem.

Here's what I didn't understand until much later: the kVA number on a UPS is rated at a specific power factor, usually 0.8 or 0.9 depending on the model. Your actual load's power factor is almost never exactly that. IT equipment sits around 0.95–0.99. Motor loads and transformer-fed distribution sit lower — 0.8 or even 0.7 depending on the mix. When your load PF doesn't match the UPS's rated PF, you don't get the full kW output. The kVA rating is still "true," it's just not useful for figuring out whether your actual load will fit.

I learned this the expensive way. The client's load mix was about 68% IT, 22% HVAC and motor loads, 10% miscellaneous. Our number assumed near-unity PF because we'd copied a sizing worksheet from an earlier data hall project that was almost all IT. The UPS was rated correctly. Our input assumptions weren't.

The Deeper Reason Sizing Guides Keep Letting People Down

It took me roughly four years and about 60 installations to actually internalize this — I'm a slow learner on some of this stuff (my wife would say "all of it, honestly"). The sizing calculator only tells you whether the UPS can serve the load. It tells you almost nothing about whether the UPS can serve the load well. Those are different questions, and the gap between them is where most of my mistakes live.

1. The efficiency curve isn't flat

A modern online UPS hits peak efficiency somewhere around 50–75% of rated load. At 25% load you might be eating 4–6 points of efficiency you didn't budget for. This matters less for short bursts and a lot more for a UPS that runs at 30% load for three years because the facility never grew the way the original plan projected. The energy waste is annoying. The heat rejection inside the electrical room is the actual problem — it forces you to oversize the room cooling, which I forgot to do, which is why the electrical room was running 6°C hotter than design that first summer.

2. Battery runtime is quoted under conditions you don't have

Every battery spec sheet gives you a runtime curve: typically 25°C ambient, new cells, specific discharge rate, aging assumption of roughly 20% capacity loss by end of life. Our test that failed wasn't a defective battery. It was a battery string sized against a nameplate curve that assumed a fresh install at perfect temperature. Five months in, with a warm electrical room and battery age already ticking, the real runtime was 30–40% lower than the number on the quote. That's not a defect. That's the difference between a spec sheet and a physics lab.

3. Nobody talks about how the UPS interacts with what's upstream

This is the one that still keeps me up sometimes. Switch-mode power supplies — which is most of your IT load — throw current harmonics back onto the input. If the UPS is feeding from a shared transformer or a generator, those harmonics don't just disappear. They heat the transformer windings, they circulate in the neutral, and they can interact with the UPS input rectifier in ways that damage the input filter over time. I have watched two different clients order a perfectly sensible UPS and then under-spec the K-rated transformer upstream because the spec sheet made the UPS look "self-contained." It isn't. Nothing in a power chain is.

Bottom line: the spec sheet tells you whether a UPS will turn on with your load. The engineering tells you whether it will still be doing that in year four.

What This Actually Cost

Let me put some numbers on the failures. The 2022 site: $52,000 in UPS and battery — reasonable. Then $8,400 in extra room cooling because the electrical room overheated past design in summer. Then $3,200 in rework when the client insisted we re-commission the battery test with a proper derating calculation. Then the part nobody prices: two weeks of schedule slip while we sorted out the actual load profile, which meant the client's IT migration got pushed, which meant I spent a month managing relationships instead of doing new work.

We also over-bought. Our 900 kVA total installed capacity was sized for a facility that's still running at 60% load two years later. Roughly 200 kVA of that capacity has been sitting idle since day one. Somewhere between $18,000 and $25,000 of hardware is being depreciated for no operational reason (I'd have to pull the asset register to give you an exact number — honestly I don't want to).

The triggering event for how I think about this now was that failed battery test in September 2022. Since then I keep a one-page pre-check on every UPS project. We've caught 14 issues on the last 22 jobs that would have turned into similar problems — wrong PF assumptions on 6, runtime derating missing on 4, upstream transformer sizing overlooked on 4.

How I Size UPS for a Data Center Now (Briefly)

I'm not going to lay out a full procedure — there are good references for that and I'm not writing another generic "10-step sizing guide." What changed for me was a shorter list.

1. Measure the actual load PF, don't assume it. Meet with the facilities engineer, pull the last 30 days of power logs. If you can't get real data, use a power analyzer for a week before you commit to a rating.
2. Size to the efficiency curve, not just the kVA ceiling. Ask the vendor for the efficiency curve across the 20–100% load range, not just peak. If your facility is projected to sit at 40% load for its first two years, the UPS you want is different from the one the sizing calculator hands you.
3. Derate the runtime number by 25–30% for year-one real-world performance. It's not conservative; it's been my experience on nearly every install.
4. Look one box upstream. Is there a transformer? A generator? What's the K-rating, and what harmonics will the UPS reflect onto it? If the vendor can't answer that, get one who can.

One honest limitation to this whole piece: this is written from the perspective of mid-size data halls and industrial facilities — roughly 200 kVA to 2 MVA range. For a small IT closet or edge site, half of this is overkill. For that class of load, a properly sized rack online UPS is fine and the transformer harmonics question mostly doesn't apply. And if you're dealing with a facility where uptime is measured in seconds, not minutes, the equation changes again — redundancy architecture starts driving the decision more than sizing does. I don't pretend to have that conversation covered here.

There's a version of this where I tell you the PowerSentinel three-phase UPS system we spec on most new jobs solved all of this. It didn't. What changed was the pre-check. What changed was the habit of forcing the vendor conversation to be about load curves and derating instead of catalog kVA. PowerSentinel's industrial UPS systems hold up really well when you actually run the numbers the way I've described — but so do one or two of their competitors. The system didn't save us. The process did.

If there's one takeaway I'd tape to a monitor: the sizing calculator is the start of the work, not the end. Every UPS I've ever had go sideways started with a number I trusted. None of them started with a UPS that was actually undersized.