In short. A data centre generator is sized and proven against the whole chain it supports, not the IT load alone. It must carry the UPS input including battery recharge and the cooling plant with its compressor and pump restarts, with an alternator that tolerates the UPS’s harmonic current and a transfer that completes before the UPS batteries run down. Capacity is arranged as N+1 or 2N, and readiness is proven by integrated tests throughout the site’s life.
The load is the whole facility, not the racks
The generator sees more than the IT load. The UPS rectifier draws the IT load divided by UPS efficiency, plus battery recharge current once the set is online, which can add 10 to 25 percent for the recharge period. Cooling for an air-cooled facility in Hyderabad’s climate typically adds 30 to 50 percent of the IT load, and it arrives as compressor and pump motor starts rather than as a steady draw. Building services and the generator’s own auxiliaries complete the picture.
| Load component | Typical share of IT load (illustrative) | Generator-relevant behaviour |
|---|---|---|
| UPS input at steady state | 105 to 110 percent | Non-linear current, THD depends on rectifier type |
| Battery recharge after outage | 10 to 25 percent extra for 30 to 120 minutes | Step increase as soon as the UPS sees generator supply |
| Mechanical cooling | 30 to 50 percent, higher at peak ambient | Motor starts, often several at once unless sequenced |
| Building and auxiliaries | 5 to 10 percent | Mostly linear, includes generator radiator fans and fuel pumps |
Worked example (illustrative). A 500 kW IT hall with 95 percent efficient UPS presents about 525 kW at the rectifier, plus 75 kW of recharge, plus 200 kW of cooling and 40 kW of building load, giving roughly 840 kW running load during the recharge period and a starting event when the chillers return. A single 1,000 kVA set at 0.8 power factor, 800 kW, would be at full load before the chiller start. The design therefore moves to two 1,000 kVA sets in N+1, or a larger single set if the facility accepts no redundancy. The arithmetic for your own facility can be built with the generator sizing guide.
The UPS and the generator must be proven together
A diesel set that carries a resistive load bank at full load is not thereby proven for a UPS. The UPS rectifier draws harmonic current, and that current across the alternator’s reactance distorts the voltage at the bus. Older six-pulse UPS can produce input current THD above 30 percent; modern units with active front ends are below 5 percent. The alternator is chosen for the actual UPS, usually with a sub-transient reactance of 10 to 12 percent and often one frame size larger than the engine needs. The UPS input window for voltage and frequency, and its behaviour when the window is exceeded, must be checked against the set’s governor and AVR response. The interaction is covered in UPS and non-linear loads.
Timeline of an outage
The design is tested against the clock. At zero seconds the utility fails and the UPS carries the IT load on batteries while cooling, lighting and everything else goes dark. At about 2 to 5 seconds the mains failure is confirmed and the set cranks. At 10 to 15 seconds the set is up to voltage and frequency and the transfer switch moves. At 15 to 20 seconds the UPS sees generator supply and begins recharging. From 20 seconds to a few minutes the cooling plant restarts, chillers in sequence so that the set is not hit with simultaneous starts. With no cooling a dense hall can rise several degrees a minute, so server thermal shutdown can arrive before the chillers are back. UPS autonomy, typically 5 to 15 minutes, is set to cover a failed first start and a second attempt, not merely the normal 15 seconds. The restart sequencing is discussed under critical-load restoration and restart sequencing.
Capacity architecture
N+1
Enough sets to carry the full design load plus one spare. Loss of any one set, or removal of one for service, leaves full capacity. It is the usual arrangement for enterprise and colocation facilities and needs paralleling switchgear and a master that keeps N+1 online.
2N
Two complete independent systems, each able to carry the full load, with no shared component between them. It tolerates a failure in one system plus maintenance on the other and is the basis of higher-tier designs.
Shared dependencies
The count of generators does not settle the matter. Two sets on one day tank with one transfer pump share a fuel failure. Two sets on one control PLC share a logic failure. Two sets through one switchboard share a bus fault. The architecture is N+1 only if the fuel, control, cooling and distribution behind the sets are also N+1. The principles are set out in N, N+1, N+2, 2N and distributed-redundant architectures.
Fuel for the stated autonomy
Data centres state autonomy in hours at full load, commonly 12, 24 or 48. At roughly 0.2 to 0.25 litres per kWh for a loaded diesel set, a 1 MW facility running 24 hours needs about 5,000 to 6,000 litres of usable fuel after allowing for unusable tank bottoms, plus a refuelling arrangement that can be relied on during a regional outage. Day tanks at each set are filled from bulk storage by duplicated transfer pumps on the generator supply. Stored diesel degrades, so long-held stock needs polishing and periodic testing. Tank arithmetic is in fuel-tank sizing.
Testing that proves readiness
A weekly no-load start proves the engine cranks and little else. Monthly or quarterly loaded tests, with a load bank where the IT load cannot be disturbed, prove the set carries load at temperature and keep it clear of wet stacking. Annually, and at commissioning, an integrated test pulls the utility breaker and watches the whole chain: UPS on battery, start, transfer, recharge, cooling restart, steady running, utility return, retransfer and cooldown, with the hall temperature and the UPS state logged throughout. A failed-start scenario is injected at least once so that the second set’s start and the UPS autonomy are seen to cover it. The load-bank side of this programme is described under load-bank sizing, and temporary capacity for maintenance windows or growth under data centre rental power.
Common mistakes
- Sizing the set from the IT load or the UPS output rating. Recharge, cooling and auxiliaries push the running load 60 to 80 percent higher and the set is overloaded on its first real outage.
- Accepting a resistive load-bank test as proof of UPS compatibility. The UPS drops to battery on the first live transfer because of voltage distortion or frequency slew.
- Allowing all chillers to restart together. The simultaneous motor start stalls the set or dips the voltage enough to trip the UPS rectifier.
- Counting generators without counting shared fuel, control and distribution. The facility is N+1 on paper and N in reality.
What to send us
Send the IT load in kW now and at build-out, the UPS make, rating, rectifier type and battery autonomy, the cooling plant schedule with motor ratings and starting methods, the target availability or tier, the required fuel autonomy in hours, and the single-line diagram if one exists. For an existing facility, the last integrated test record and any UPS alarm logs from a generator transfer tell us most of what we need. Use the application engineering enquiry.
Questions engineers ask
How big should a data centre generator be relative to the IT load?
As a first estimate, 1.6 to 1.8 times the IT load in kW covers UPS losses, battery recharge, cooling and building services in an air-cooled facility in Indian conditions, before adding redundancy. The actual figure comes from the cooling schedule and UPS data, and from checking the largest motor start against the set’s transient capability.
Why does the UPS go to battery when the generator takes over?
The generator’s voltage or frequency is leaving the UPS input window during the transfer or the recharge step, often because the alternator is too small for the UPS’s harmonic current or the governor overshoots. Fixes include a lower-reactance alternator, a recharge current limit on the UPS, and widening the UPS input window where the manufacturer allows it.
Is N+1 enough for a data centre?
For most enterprise and colocation facilities, yes, provided the fuel, controls and distribution are also N+1 and the spare set is kept online or starts automatically on a trip. Facilities that must tolerate a failure during maintenance move to 2N or a distributed redundant arrangement.
How much fuel should a data centre keep on site?
Enough for the stated autonomy at full load plus a margin for unusable tank volume and a failed delivery, commonly 24 to 48 hours. A 1 MW facility uses roughly 200 to 250 litres an hour at full load.
Technical content reviewed by Srinivas Reddy, Managing Partner, Reddy Generators. Last updated 19 October 2026.