Technical explainer

How data centre power distribution works

11 August 2026 15 minute read

Data centre power distribution is the chain of equipment carrying electricity from the utility connection down to an individual server. Every stage in that chain exists to do one of three things: convert voltage, hold the load up when the supply fails, or subdivide what is left into smaller protected circuits. Read it that way and the equipment list stops being a list.

Three jobs, not nine boxes

A single-line diagram of a data hall shows eight or nine pieces of equipment in a row, and read as a row it is hard to remember. Grouped by what each one is for, it collapses to three:

• Conversion. Utility power arrives at a voltage no rack can use. Medium-voltage switchgear, a step-down transformer and the low-voltage switchboard bring it down and put protective boundaries around it.
• Continuity. Two separate mechanisms cover two different failure durations — a UPS for the seconds, an engine-generator for the hours. Neither substitutes for the other.
• Subdivision. One large low-voltage feed becomes hundreds of small protected circuits, through a floor PDU, an RPP or busway, and finally the rack PDU at the cabinet.

The terms, before the rest of the article uses them

Two of these words are used loosely across the industry. Neither "PDU" nor "RPP" is fixed by a single IEC or IEEE component standard — both are trade usage, and the floor-PDU-versus-RPP distinction is not applied consistently between operators. These are the definitions this article uses; on an unfamiliar site, confirm what each labelled box actually contains rather than assuming the label matches.

Conversion: getting to a voltage a rack can use

Why the supply arrives at medium voltage

Utility power to a large site normally arrives at medium voltage rather than low voltage. Carrying the same power at MV over site distances reduces conductor size and distribution losses, so an incoming feed of this scale is practically never brought in at final utilisation voltage.

The equipment that receives that feed — metal-enclosed switchgear rated above 1 kV up to 52 kV — is standardised as its own class under IEC 62271-200, separately from the transformer it feeds. The switchgear and the transformer are specified, tested and procured against different standards even though they sit next to each other on the diagram.

Before any of that equipment is specified, the connection has to be agreed with the network operator. In Great Britain the Grid Code, administered by the National Energy System Operator, sets the technical and connection rules for a large demand customer connecting to the transmission network. Large new demand connections have become a significant share of connection requests, and connection and queue rules have been under active revision — the connection process itself, not just the switchgear order, deserves planning time.

The transformer and the switchboard it feeds

A power transformer steps the incoming MV supply down to low voltage, under 1000 V AC, before it reaches the main LV switchboard. Transformers in this role sit under IEC 60076, whose Part 1 carries the general specification, rating, connection, marking and testing requirements.

What the transformer feeds is split between two standards, divided by what the equipment is rather than by product line. The switchboards, panelboards and motor control centres downstream — the assemblies — are covered as a family by IEC 61439. The individual switching and protective devices mounted inside those assemblies are standardised separately under IEC 60947, which explicitly excludes assemblies from its own scope.

A buyer is therefore specifying against two standards at once: one for the enclosure and busbar system, one for the breakers inside it. A component that satisfies IEC 60947 does not by itself make the assembly around it compliant with IEC 61439, and a supplier who conflates the two has not understood the question.

Continuity: two mechanisms for two different durations

A UPS and a generator are often described as though one backs up the other. They do not. They cover different lengths of outage, and a facility needs both because neither can do the other's job.

The UPS covers the seconds

An engine-generator takes time to start and reach stable output. Across that interval the load is carried from stored energy in the UPS. That is the whole of the UPS's role in an outage — it is a hold-up device, not a source of supply.

IEC 62040-3 groups UPS equipment by how tightly output voltage and frequency track the input: voltage-and-frequency-dependent passes input disturbances through; voltage-independent regulates voltage but still follows input frequency; and voltage-and-frequency-independent — double-conversion, the type most used for critical IT load — decouples output entirely from input variation. This is not only an IEC classification: in the United States, UPS equipment above certain size thresholds falls under a federal energy-conservation test procedure that references the same grouping.

Because a UPS sits in series with the load, its own failure is a load failure. That is why one is commonly fitted with a maintenance bypass — the same redundant-path logic applied at component rather than facility scale.

The generator covers the hours

An engine-generator set sized to the facility load connects through an automatic transfer switch, which senses loss of utility supply and moves the load onto standby power. This path runs alongside the utility feed rather than in series with it, so it can pick up the load without waiting for the failed path to be repaired.

Three standards cover different parts of it. The generating set sits under the ISO 8528 series. The transfer switch is covered by UL 1008. The system as a whole — generator, transfer switch, controls and wiring — is governed by NFPA 110, which classifies emergency power supply systems on two axes: a Type, being the maximum time to deliver rated load after the normal source fails, and a Class, being the minimum run duration without resupply. Those two axes are the citable way to state a requirement.

Emissions rules constrain when a generator may run

Most standby generators are diesel-fuelled, valued for reliability and fast start; natural gas is an alternative in some facilities, trading simpler refuelling logistics for a different run-time and emissions profile.

In the United States an emergency generator is regulated as a stationary source of air pollution under the Clean Air Act, and non-emergency running is treated as a compliance question separate from the electrical design. Guidance permits an emergency generator to run in defined non-emergency circumstances — as part of a demand-response arrangement, within an annual hour limit — without losing its emergency classification. Anyone planning non-emergency use should check that limit before designing around it.

Subdivision: one large feed becomes hundreds of small ones

Floor PDU, RPP and busway

A floor PDU is a room-level unit, typically distinguished by carrying its own step-down transformer, sitting downstream of the UPS output switchboard and feeding onward. An RPP takes one upstream feed and splits it into the branch circuits running to rack rows; set against the floor PDU, its distinguishing feature is what it lacks — no transformer, no voltage step-down, breakers and busbar fed at final utilisation voltage.

Busway is the other common way to carry power the last distance: a busbar trunking run, overhead or under floor, used instead of feeder cable, with tap-off points along its length. It is not a separate standards regime — busway is an assembly and falls under the same IEC 61439 family covering the switchboards upstream. Choosing between busway and feeder cable is a layout and serviceability decision, not a difference in governing standard.

Where the connector rating steps down

The rack PDU is the final distribution device: an outlet strip fed by one or more branch circuits, connecting servers and network equipment to the supply. Its connectors mark the point where current rating drops sharply.

Feeder-level connections upstream commonly use industrial-pattern plugs, sockets and couplers under the IEC 60309 series, rated up to 1000 V and 800 A. A server's cord, by contrast, terminates in an IEC 60320 appliance coupler, rated up to 250 V AC and 16 A — a general-purpose class built for equipment current, not feeder current. That step between connector families is a physical illustration of how far the chain has narrowed by the time it reaches a server.

Redundancy: what fails, and what survives it

Start from what N means: the capacity actually needed to serve the load, nothing more. Every topology is a statement about how much capacity is installed above N, and across how many independent paths.

N+1 adds one spare unit beyond N. It survives loss of a single unit — a module failure, or one taken down for maintenance — because enough capacity remains. It does not survive loss of whatever those units share: if every module sits behind one switchboard, losing that switchboard takes all of them down together regardless of spare module count.

2N duplicates the whole capacity requirement into two independent, fully rated paths, each alone able to carry the load. That is a materially different guarantee, since it survives loss of an entire path rather than one unit within a shared path. The cost is proportional: roughly double the installed capacity of N, idle in normal operation.

Distributed redundancy spreads capacity across three or more paths, each sized above N divided by one less than the number of paths, so losing any one path still leaves enough aggregate capacity to cover N — most of 2N's path-level protection without a full duplicate of every path.

A worked comparison at 300 kW

The figures below are chosen for the illustration, not taken from any installation. Take a critical load of 300 kW.

N        3 × 100 kW modules            = 300 kW installed
N+1      4 × 100 kW modules            = 400 kW installed
2N       2 × 300 kW paths              = 600 kW installed
Dist.    3 × 200 kW paths (each > N/2) = 600 kW installed

Losing one module under N+1 leaves exactly 300 kW — enough. Losing the shared switchboard those four modules sit behind takes all 400 kW down regardless of module count. Under 2N, losing either path leaves the full 300 kW on the other. Under distributed redundancy, losing any one of the three paths leaves 400 kW across the remaining two.

2N and distributed redundancy land on the same installed total here because both survive loss of a whole path. They differ only in how many paths that capacity is split across, and therefore in how much of it is idle at any moment.

Redundancy does not transfer between stages

These topologies apply independently at each point in the chain — utility feeds, generators, UPS modules, switchboards — so a facility can and often does mix them, say 2N at the UPS stage with N+1 at generation. Resilience to a generator failure is then governed by the N+1 stage, not by the 2N stage elsewhere. Redundancy bought at one point does not carry over to a less redundant point.

This article assigns no availability or downtime percentages to any topology. NFPA 110's Type and Class axes are the citable way to state a requirement of that kind.

Reading a single-line diagram as a fault-analysis tool

The diagram earlier in this article is not only a layout. Equipment in series with the IT load — the UPS, its output switchboard — interrupts that load if it fails, because there is no path around it. Equipment in parallel to the load — a lighting or mechanical panel fed from the same switchboard — can fail without directly interrupting IT load. Where a device sits relative to the load determines how its failure propagates, which is why the series and parallel distinction is worth marking on the drawing rather than inferring later.

Reserve capacity carries a running cost, not only a capital one. Transformers and UPS units are most efficient in the middle-to-upper part of their rated load and lose efficiency sharply when lightly loaded, so idle redundancy is one of the larger controllable sources of electrical inefficiency in a facility. Oversizing UPS capacity for redundancy is a capital decision and an operating-cost decision at the same time.

One architectural alternative is worth naming without overstating its status: medium-voltage DC distribution is an active area of technical work, evaluated as an alternative internal architecture against the conventional AC chain described here. It is not in general use.

Frequently asked questions

What is the difference between a PDU and an RPP in a data centre?

An RPP is a breaker-and-busbar panel with no transformer of its own, fed at final utilisation voltage. A floor PDU typically includes its own step-down transformer and feeds onward to RPPs or busway. Usage varies between operators, so check a given site's own diagram rather than assuming.

What does N+1 redundancy mean for power distribution?

N is the capacity the load actually needs. N+1 adds one spare unit beyond that, so one module, generator or path can fail or go down for maintenance without interrupting the load. It does not protect against loss of a shared distribution path feeding every module.

Why does a data centre need both a generator and a UPS?

They cover different durations. The UPS carries the load from stored energy across the seconds between a utility failure and the generator reaching stable output. The generator then carries it for as long as fuel supply and run-time allow. They are complementary, not interchangeable.

What voltage does power arrive at before it reaches the racks?

Utility supply typically arrives at medium voltage and is stepped down on site to low voltage, under 1000 V AC, before reaching the main switchboard. In many designs it is stepped down once more at the floor PDU before the rack PDU.

Is a rack PDU the same thing as a power strip?

Functionally related, built to a different class. A rack PDU distributes one or more branch circuits to multiple outlets inside a cabinet, often with per-outlet monitoring or switching, and connects upstream through industrial-pattern connectors rated well above a household strip's.