Data Centre Power Systems in Singapore: UPS, Generators, Busways and PDUs in Plain English

On a colocation sales tour, the host points at the diesel generators, says “100% power SLA” and “2N”, and moves on. The questions that actually protect you are quieter: how many independent paths feed my rack, how long can the site run without the grid, and will this hall cope when my racks draw twice the power in three years?

This is a beginner’s guide to the equipment behind those questions. No electrical background needed: if you can picture water flowing through pipes that narrow as they go, with tanks and backup pumps along the way, you can follow all of it.

It sits in our data-centre cluster and does not re-cover two neighbours. For the whole building — cooling, PUE, Uptime “Tier” ratings and how Singapore rations capacity — see the pillar, how data centres work in Singapore; for the design-and-operations layer, see data centre power design. Here, we explain what each piece of kit is.

The journey from the grid to the chip

Electricity behaves a little like water flowing through narrowing pipes, with filters and backup tanks along the way. Here is the whole journey in one pass:

1. The national grid feeds the site at high voltage — large Singapore sites commonly take a 22 kV feed, ideally two from different substations.
2. Transformers step it down to about 400 volts for the data halls.
3. Switchgear — the big breakers that start, stop and isolate circuits.
4. Generators wait on standby and start within seconds if the grid fails.
5. The UPS cleans the power and, on batteries, covers the gap while generators start.
6. Busways or cables carry power down the hall; PDUs split it into the sockets that feed each rack.

A serious facility duplicates the important stages; the rest of this guide opens up the boxes that do the real work.

A sprinter and a marathon runner: batteries vs generators

Two things provide backup power, and they are good at opposite jobs. UPS batteries are a sprinter — instant power, but only for minutes. A generator is a marathon runner — a few seconds to start, then hours on fuel. The everyday picture: the UPS is the power bank in your pocket; the generator is the one in the car park. You need both.

When the grid drops, the UPS carries the servers for a short window — commonly 10 to 15 minutes at full load — while the generators start. A Singapore site typically installs one more generator than it needs and stores around 24 hours of fuel. Generators usually power the cooling too, since servers that overheat still fall over — and one that is never load-tested is decoration.

Inside a UPS: what “uninterruptible” really means

UPS stands for uninterruptible power supply, and it does two jobs: it bridges the gap until the generator runs, and it continuously cleans the power, smoothing the sags and spikes the grid delivers.

Most critical halls use a “double-conversion” design, and the name explains it. Incoming grid power (alternating current, or AC) is turned into direct current (DC) by a rectifier; that DC charges the batteries and feeds an inverter, which turns it back into clean AC for the servers. So your servers never run on raw grid power — the UPS is always in the middle, which is why it is called “online”.

Two more spec-sheet terms: a static bypass routes utility power straight to the servers if the UPS itself fails, and a modular UPS is built from identical modules, so one can fail without dropping the load and you add capacity as you grow. Efficiency matters too — an oversized or ageing UPS wastes power every hour, while the best modern units run at well over 96%.

Busways and PDUs: the last few metres to your rack

Older halls run thick cables from a floor panel to each rack. Modern high-density halls hang a busway overhead instead — a rigid track of copper or aluminium bars above the cabinets. To power a new rack you clip a “tap-off” box onto the track, rather than pulling a long cable back to a distant panel — faster to change, and tidier in a live hall.

At the rack, a PDU (power distribution unit) splits one feed into many protected sockets. Floor PDUs feed rows or busways; the tall, thin rack PDUs inside the cabinet are the industrial cousin of a power board, and come in a ladder of intelligence:

• Basic — sockets only; cheap but blind.
• Metered — shows total current, so nobody overloads a circuit.
• Monitored — reports voltage, current and power over the network.
• Switched — turns sockets on and off remotely, so a stuck server reboots without a site visit.
• Intelligent — meters every socket, for fair billing, capacity planning and finding idle “ghost” servers.

This is where redundancy becomes real or imaginary. A good hall gives each rack two independent feeds, an “A” side and a “B” side, from separate busways and UPS paths, and a proper server has two power supplies, one on each. Any single feed, PDU or UPS can then fail without dropping the server. Plug both cords into the same strip, though, and you have a tidy-looking single point of failure. Human error, not equipment, causes most outages.

Batteries: lead-acid or lithium?

The UPS’s batteries buy you those first critical minutes, and there are two common types. The choice is not only about price — in Singapore’s heat and costly floor space, it is a real design decision.

Lithium-ion is not always the answer: in a cool room with a shorter horizon, lead-acid can be the cheaper, sensible choice. UPS-grade lithium is not a laptop cell — it ships with battery management and fire-safety engineering — but insurance, the building’s fire strategy and the vendor’s record still belong in the decision.

You cannot manage what you cannot measure

Hardware keeps the lights on; software tells you whether they will stay on. A serious facility meters power at every stage — feeds, UPS, generators, busways and PDUs — and pulls it into monitoring systems (often labelled EPMS or DCIM), with battery monitoring watching individual cells rather than trusting the UPS to simply report “OK”.

For a tenant the test is simple: can the operator show you, in near real time, how many kilowatts you use, where they go, and how close a circuit is to tripping? Without metering at the PDU, nobody can bill fairly, plan your next rack, or prove an efficiency claim.

Singapore's efficiency bar: why wasted watts cost you

Singapore has little land, imports almost all its energy, and rations new data-centre capacity by policy — so efficiency here is close to a licence to operate. The headline measure is PUE (Power Usage Effectiveness): total facility power divided by the power reaching the IT equipment. A PUE of 1.5 means every 100 watts of computing costs another 50 in overhead; lower is better. Our pillar guide covers PUE and the capacity rules in full.

A few rules shape power and cooling choices. The Green Data Centre Roadmap targets at least 300 MW of new capacity and nudges operators toward efficient kit via the Energy Efficiency Grant.^[1] The Tropical Data Centre Standard, SS 697:2023, shows how to run halls safely at 26°C and above, saving 2–5% of cooling energy per 1°C, on cooling that can be up to 40% of a site’s energy.^[2] SS 715:2025 sets efficiency levels for the IT equipment itself, targeting at least 30% less energy.^[1] And SS 726:2026 is the world’s first standard for liquid cooling in tropical data centres.^[4]

The bar for new capacity shows the direction. Winners of the latest allocation, DC-CFA2, must earn Green Mark for Data Centres Platinum, hit a PUE of 1.25 or better at full IT load, use SS 715-compliant equipment, and draw at least half their power from green pathways such as low-carbon hydrogen, ammonia, biomethane or novel fuel cells.^[3] The takeaway for a buyer is blunt: a hall that wastes power is more expensive to sit in, and a small PUE gain is a large number on a Singapore electricity bill.

How to choose: the questions that matter

You do not need the most redundant, most expensive hall for every workload. Match resilience to what an hour of downtime actually costs you — a brochure that says “Tier III” or “Tier IV” describes what a design can tolerate, not a guaranteed uptime figure (the Uptime Institute removed those availability percentages back in 2009).^[5] With that framing, ask:

1. Two independent utility feeds, on different routes from different substations?
2. What UPS design, and how many minutes of battery at your load — not an empty hall?
3. How many generators, how many hours of fuel, and when was the last full-load test?
4. Does each rack get two feeds (A and B) from separate paths, and is your kit cabled to use both?
5. What PUE, and is power metered at the PDU so you see and are billed for what you use?
6. How much power per rack today, and can it grow when your racks get hungrier?

Vague answers tell you more than a glossy specification. A well-designed power system is invisible when it works: no flicker, no surprise on the invoice, no Saturday-night generator drama.

Compare data centre providers in Singapore

Browse directory profiles for colocation and data-centre operators, with recorded facility and power details where available.

Browse data centres in Singapore

Frequently asked questions

Sources

1. IMDA — Green Data Centre Roadmap — Infocomm Media Development Authority, 2024-05 official checked 2026-09-06
2. IMDA — Tropical Data Centre Standard (SS 697:2023) — Infocomm Media Development Authority, 2023 official checked 2026-09-06
3. IMDA & EDB — Launch of second Data Centre Call for Application (DC-CFA2) — Infocomm Media Development Authority, 2025-12-01 official checked 2026-09-06
4. IMDA & EnterpriseSG — Singapore launches the world's first liquid cooling standard for tropical data centres (SS 726:2026) — Infocomm Media Development Authority, 2026-08-27 official checked 2026-09-06
5. Uptime Institute — Tier Standard and Tier Certification — Uptime Institute official