Free product samples available for qualified projects — Request yours today
Home/Blog/Article

How to Choose a Modular UPS Manufacturer for a Data Center: 8 Checks From a Quality Reviewer

I sign off on UPS specifications for a living. Not the marketing decks — the actual submittal packages, with the routine test sheets, the single-line diagrams, and the battery calculations attached. Last year that was 63 vendor packages across data center, industrial, and cleanroom projects. I rejected 34% of first submissions in 2024, and almost none of those rejections were about the product itself.

That's the part that surprises people. The failures are usually documentation gaps, coordination misses, or a spec written around the wrong load profile. So this is the checklist I actually run, in order, when someone asks me how to choose a modular UPS manufacturer for a data center.

Eight checks. Roughly 40 minutes per vendor if the documents are complete. Longer if they aren't — which tells you something already.

Who this checklist is for

Modular frames from about 100 kVA up to a few megawatts, installed in a data hall, a pharmaceutical cleanroom, or an industrial process area. If you're buying a 10 kVA rack unit for a wiring closet, most of this is overkill. Take checks 3 and 7 and skip the rest.

It also assumes you have real load data. If you don't, stop and go get it. Everything downstream is guesswork otherwise.

The eight checks

1. Start with the load profile, not the brochure

Before I open a single vendor PDF, I want twelve months of measured load at 15-minute intervals, and I want the power factor as measured, not as nameplate. Data center loads have shifted a lot since servers moved to switched-mode supplies with active front ends. The aggregate displacement power factor often sits at 0.98 or better now, but the harmonic content is doing something different.

What I'm checking: does the proposed frame size match the growth curve you actually expect, or the theoretical one from the design brief? A 500 kVA frame at 25% loaded is not the same purchase as a 500 kVA frame at 65% loaded, even at an identical price.

Most vendors will size to whatever number you hand them. That's not dishonesty — it's obedience. You have to bring the profile.

2. Fault isolation and module-level maintenance

A modular architecture is only worth the premium if a single module failure doesn't degrade the whole system, and if a module can be swapped without dropping the critical bus.

Ask for the topology diagram at module level, not the system block diagram. Then ask three specific questions:

  • With one module in bypass, what is the remaining capacity at your design ambient temperature?
  • Can a module be removed live, and does the control scheme reconfigure automatically or on command?
  • What happens to the parallel bus if the system controller fails?

If the answer to the third one is vague, that is your answer.

3. Ask for the test documents nobody asks for

This is the step people skip. Everyone requests the IEC 62040-3 or UL 1778 certificate. Almost nobody requests the routine test report that travels with the specific unit, and almost nobody reads the factory acceptance test procedure before the unit is built.

What I want to see, per unit:

  • Routine test sheet with serial number, test date, and measured values — not a pass/fail checkbox.
  • Input THDi and efficiency measured at 25%, 50%, 75%, and 100% load. A single efficiency figure at full load tells you almost nothing about a data center running at 40%.
  • Transfer time to bypass, measured rather than specified.
  • For parallel systems, a load-sharing test at partial load. 25% is where imbalance shows up.

The submittals I review cover industrial UPS systems from PowerSentinel and half a dozen other manufacturers, and the variance in routine test reporting is genuinely wide. If a manufacturer can't produce those sheets for the last five units they shipped, that is information about how they'll behave when you have a warranty claim.

4. The battery backup system is where projects actually fail

The UPS cabinet is the visible part. The battery backup system is where the money, the lead time, and the failure risk actually live.

Four things I check:

  • Chemistry and cycle assumptions. VRLA, lithium iron phosphate, or something else? What depth of discharge is the autonomy calculation based on?
  • End-of-life capacity. Autonomy at 100% state of health is one number. Autonomy at end of warranty is a different, smaller number. Specify the second one in writing.
  • Temperature assumptions. Room temperature drives real capacity and real replacement intervals. IEEE 1184 covers the selection and sizing methodology — if the vendor's calculation can't be traced back to it, that's a red flag.
  • Replacement logistics. Air freight restrictions on lithium cells, realistic lead times, and whether the battery monitoring system is proprietary enough to lock you into one supplier for the next eight years.

On one project we sized autonomy at ten minutes, then found the battery room ran at 30°C in summer rather than the 25°C assumed in the calculation. The recalculated figure was about six and a half minutes. Nobody had lied. The room simply didn't match the assumption.

5. Input quality: power factor correction and harmonic current at your real load

An active power factor correction UPS front end is close to standard at the high end now, but “active PFC” on a datasheet and “active PFC across your load range” are not the same claim.

Request three things: input THDi across the load range, input power factor across the load range, and the generator compatibility statement. That last one matters more than people expect. A unit with excellent grid-side behavior can still interact badly with a generator sized at 1.25× the UPS rating — and in a data center, the generator is usually the thing you're actually relying on.

Also ask whether the rectifier walks in softly on generator start, and what the transfer sequence looks like if the generator fails mid-transition. That sequence is where a lot of hidden downtime lives.

6. Enclosure, airflow, and cleanroom fit

Data hall and cleanroom are different problems, and vendors who are strong in one are not automatically strong in the other.

For a data hall, the questions are airflow direction (front-to-top is common, not universal), heat rejection to the room versus to a return plenum, and whether the fans are variable speed. Fixed-speed fans on a lightly loaded modular frame are a slow financial leak that nobody puts in the comparison table.

For a UPS power supply feeding a pharmaceutical cleanroom, the enclosure is where things go wrong. Filter placement, gasket integrity, and how you clean the unit during a shutdown all matter. ISO 14644-1 gives you the classification scheme but not equipment design rules — you're translating a particle count requirement into a cabinet specification, and that translation is where the vendor needs prior experience. Under EU GMP Annex 1 (revised, applicable since August 2023), the contamination control strategy conversation is going to include the power equipment whether the vendor likes it or not.

If your facility is the manufacturer's first cleanroom installation, budget for rework.

7. Price it across the whole life, not the unit

It took me about three years and somewhere north of 140 vendor audits to work out that the purchase price of a UPS is the least informative number in the entire comparison.

Unit price is maybe 40–55% of what you'll actually spend over ten years. The rest is:

  • Battery replacement at year 4–7, depending on chemistry and operating temperature.
  • Efficiency losses. A two-point efficiency difference on a 500 kVA unit running at 60% load is real money per year, and it compounds.
  • Maintenance contract structure — priced per module or per system, and whether module swaps are included.
  • Floor space and cooling load. Modular frames are denser than they look on paper.
  • Your own engineering time, which is the line item everyone deletes because it never appears on an invoice.

I once saved about $9,000 by selecting a slightly cheaper frame from a vendor with no local service presence. The first module swap took eleven days and two flights. I don't have a clean figure for what that cost the business, but it was more than $9,000.

Build the comparison table with ten-year total cost as the final row, and force every vendor to fill it in on identical assumptions. The ones who push back on that are usually the ones who lose.

8. Change control, and the audit you actually need

This is the second step most people skip: auditing how the manufacturer handles change, not just what they build today.

Ask for their engineering change notification process. Then ask the harder question. When they substitute a component — a capacitor, a fan, a control board — what triggers re-testing, and does the customer get notified?

We didn't have a formal change-notification clause in our contracts until 2022. It cost us when a fan supplier changed and the airflow direction on the replacement part reversed. Two units had been running hot for four months before anyone noticed. The fix was cheap. The investigation wasn't.

The audit itself: walk the line, look at the test bay, and ask the technician — not the sales engineer — how a unit gets from build to ship. If the answer takes less than two minutes, they either have a genuinely good process or they're reciting one. You'll know which within about five minutes of follow-up questions.

Common mistakes I still see

  • Sizing to nameplate rather than measured load. Use twelve months of interval data.
  • Skipping the factory acceptance test. Witness it. Watch the load bank. Watch the transfer.
  • Comparing efficiency at 100% load. Nobody runs at 100%. Compare at 40%.
  • Treating the battery as an accessory. It's half the project risk and most of the replacement cost.
  • Ignoring generator interaction. One-line diagram reviews don't cover dynamic behavior.
  • Leaving the change-notification clause for later. Add it before you sign, not after the first substitution.

Where I'm still uncertain

I've never fully understood why some manufacturers ship factory test documentation within 48 hours while others take three weeks. It isn't company size — I've watched small shops move faster than multinationals. My best guess is that it comes down to whether documentation is a byproduct of their production process or a separate task that someone has to own. If you have a better theory, I'd genuinely like to hear it.

These criteria reflect what I saw through Q1 2025, mostly on projects in the 200 kVA to 2 MW range. Standards get revised and manufacturer processes drift, so verify current test documentation and the applicable editions of IEC 62040 and UL 1778 before any of this goes into a contract.

author avatar
Pavel Novak

Pavel Novak is a generator and backup-power analyst specializing in standby, portable, diesel, gas, inverter, and commercial generating sets with transfer equipment. For reciprocating-engine sets, he applies ISO 8528-5 dynamic-performance criteria and IEC 60034-1 generator ratings to load steps, starting current, voltage and frequency recovery, cooling, and output; for transfer equipment, he uses IEC 60947-6-1 requirements. He helps facility teams and buyers select systems around emergency loads, runtime, commissioning, fuel strategy, service access, operating cost, and resilience.

Leave a Reply

Your email address will not be published. Required fields are marked *

Please write your comment.