Power · UPS

Data Center UPS Systems for AI Power Trains

A UPS is the bridge that keeps compute alive through a utility dip or the seconds before generators take over — and it conditions incoming power so sensitive GPU hardware sees a clean sine wave. For AI halls that means large modular, double-conversion units, scaled and made redundant by hot-swap power modules.

What it is

An uninterruptible power supply (UPS) sits in the power train between the switchgear and the rack PDUs. A double-conversion (online) UPS continuously rectifies incoming AC to DC and inverts it back to AC, so the load is always fed from a clean, regulated source and rides through sags, surges and short outages from its battery with zero transfer time. That conditioning matters as much as the ride-through for dense GPU loads.

Modern data-center UPS systems are modular: a frame holds hot-swappable power modules, so capacity scales by adding modules and redundancy (N+1) comes from running one spare module. A 1500 kVA modular frame anchors a sizeable power block; larger halls parallel several frames. High-efficiency designs reach ~97% in double-conversion and ~99% in eco mode, which materially cuts losses on a megawatt-scale load.

When you need it

A UPS is non-negotiable for production AI capacity — the question is topology, sizing and how much redundancy the tier demands.

  • Any production AI/HPC hall — the UPS conditions power and bridges to generator start.
  • High-density power blocks needing large three-phase capacity (hundreds of kVA to multiple MVA).
  • Redundancy tiers: N+1 hot-swap modules, or 2N for the highest-availability designs.
  • Sizing where GPU power factor and headroom make kVA and kW differ — both must be sized.
  • Efficiency / OPEX targets, where ~97% online (99% eco) meaningfully cuts megawatt-scale losses.

Quick facts & specs

TypeModular 3-phase
Rating1,500 kVA / 1,500 kW
TopologyDouble-conversion online
Efficiency≤ 97% (eco 99%)
Key specifications — Galaxy VXL UPS 1500 kVA
TypeModular 3-phase UPS
Rating1,500 kVA / 1,500 kW
TopologyDouble-conversion online
Efficiency≤ 97% (eco 99%)
FormFrame, hot-swap modules
Size it

UPS Sizing Calculator for GPU Racks

Size real power (kW), apparent power (kVA) and the number of UPS units from IT load, power factor and N / N+1 / 2N.

Open the calculator

Source it

Add the equipment to your quote list — pricing, availability and lead times are confirmed after you ask, usually within a business day.

Compare
Vertiv · Power Infrastructure

Rack PDU 415V 60A

415V · 60A · metered vertical

16–20 wkNew · Used · Refurb

From $1,180 / unit

Min order 4 units

Browse all power infrastructure equipment

Frequently asked

What does a data center UPS do?

A UPS conditions incoming power and bridges outages. A double-conversion (online) UPS continuously rebuilds the AC waveform so the load always sees a clean, regulated supply, and it rides through utility sags and short outages on battery with zero transfer time — long enough for generators to start and take over.

How do I size a UPS for GPU racks?

Start from the IT load in kW, divide by the power factor to get apparent power in kVA, add headroom, and apply your redundancy scheme (N, N+1 or 2N) to get the number of UPS modules or frames. kW and kVA differ when power factor is below 1, so both must be sized. Our UPS sizing calculator for GPU racks does this end to end.

N+1 vs 2N UPS redundancy — which do I need?

N+1 runs one spare power module (or unit) beyond the load, so a single module can fail or be serviced without losing power — efficient and common. 2N runs two fully independent power trains (A and B), tolerating a whole-system failure, for the highest-availability tiers. Modular UPS frames make N+1 as simple as adding a hot-swap module.

Why is a modular UPS preferred for AI data centers?

Modular UPS frames hold hot-swappable power modules, so you scale capacity by adding modules and get N+1 redundancy from one spare — no forklift upgrades, and failed modules swap live. A 1500 kVA modular frame anchors a large power block, and multiple frames parallel for whole-hall capacity at ~97% efficiency.

Want a second opinion on a build?

Our engineers scope power, cooling and compute together. No payment, and quotes come back in about a business day.

Talk to an engineer