Home / Articles / 5 October 2026
Theme of the month: Data centresN+1, 2N and the common point: why redundancy looks doubled when it is not
Two supply paths, two UPS systems, two generators — and the whole facility stops on one breaker. How to read a redundancy diagram and where the points hide at which the two paths quietly become one.
A data centre is the one kind of facility where the requirement is not stated as power but as time. Not “400 kW”, but “no more than so many hours of unavailability per year” and “maintenance without shutdown”. Everything else — the topology, the size of the room, the number of units and the capital cost — follows from that answer.
The European framework for this is the EN 50600 series: Part 1 introduces the availability classes, Part 2-2 covers power supply and distribution. Alongside it sits the Uptime Institute classification, in which the two key terms are concurrently maintainable — any element can be taken out for maintenance without stopping the facility — and fault tolerant — any single failure does not cause an interruption. The difference between them is not an availability percentage but whether an outage counts as planned or not.
What N actually means
N is the number of units required to carry the design load. If 400 kW is carried by two 250 kVA UPS units, N = 2. From there:
- N — no redundancy. Lose one module and the load drops. Used only for non-critical parts.
- N+1 — one spare module covers the loss of any working one. Three 250 kVA modules for a load needing two. Covers a single failure and allows one module to be serviced.
- 2N — two independent systems, each able to carry the entire load alone. Two incomers, two UPS systems, two distribution systems to the server’s two power supplies.
- 2(N+1) — two independent systems, each of which is itself N+1. Found at the highest classes.
Between N+1 and 2N there is a substantial difference the labels do not show. N+1 makes the equipment redundant; 2N makes the path redundant. With N+1 the three UPS units usually work onto a common output busbar — and if that busbar fails or has to be serviced, the spare does not help.
The five places where two paths become one
Almost every diagram drawn as 2N has at least one of the following. The check is simple: trace the chain from the medium-voltage incomer to the server power supply and ask of every element, “if this burns out, how many paths go down?”
- A common transformer. Two low-voltage incomers fed from one transformer are one incomer. The same goes for a single medium-voltage cable from a single substation.
- A common output busbar. The classic error in N+1 presented as redundant. The busbar is one element and has no spare.
- A common route. Two cables in one containment are one cable as far as fire, mechanical damage or flooding are concerned. Separation has to be physical, not only on the diagram.
- The static transfer switch. An STS with a single output merges the two paths immediately before the load. For a single-corded consumer that is acceptable and often necessary; for a dual-corded server it is a single point inserted for nothing.
- The controls. One controller commanding both branches is a single point of failure in the logic, however impeccably the power side is split.
The one number that settles the argument
Every discussion of redundancy reaches the question of whether it is worth it. The answer is arithmetic, not conviction. Two expressions are compared: the capital difference between the topologies divided by the years of service, against the expected cost of unavailability — hours of interruption per year multiplied by the value of one stopped hour for that particular client.
For a colocation hall with contractual penalties the second term is large and 2N is easily justified. For a server room in a production plant, where an hour of downtime means inconvenience rather than lost output, N+1 with properly separated routes is the more sensible answer. The problem is paying for 2N and receiving N+1 with two labels on it.
A check that takes five minutes: take the single-line diagram and mark in one colour everything fed from path A, and in another everything from path B. The element that ends up in both colours is the common point. If there is none, the topology really is 2N. If there is one, it defines the facility’s real class, whatever the brief says.
What stays outside the diagram
Redundant power means nothing without redundant cooling. At densities above 10 kW per rack the hall temperature rises by several degrees a minute once cooling stops — servers trip on thermal protection long before the batteries run down. Precision cooling is made redundant by the same logic and with the same search for common points: a common chiller, a common pipe, a common pump.
Coming up in this series: how real battery autonomy is calculated, power density and the limits of air cooling, and per-circuit metering for reporting under Article 12 of the Energy Efficiency Directive.