Availability figures are quoted everywhere in the data centre industry, often without the context needed to compare them. The difference between "four nines" and "five nines" sounds small. In practice it is the difference between most of an hour of outage each year and about five minutes.
The arithmetic of nines
Availability is the share of time a system delivers its service. A year has 525,600 minutes, so each additional nine divides the permitted downtime by ten:
| Availability | Permitted downtime per year |
|---|---|
| 99.9% ("three nines") | about 8.8 hours |
| 99.99% ("four nines") | about 53 minutes |
| 99.999% ("five nines") | about 5.3 minutes |
For comparison, the Tier IV facility standard is commonly associated with around 99.995% availability for the facility as a whole. A power supply that targets five nines at the point of connection is designed so that power is very rarely the reason a facility misses its own target.
No single machine delivers five nines
Every generator, engine, turbine, transformer and breaker needs maintenance and will eventually fail. A single unit — however well built — might reach 95–98% availability once planned maintenance is counted. Five nines is not a property of equipment. It is a property of architecture.
The design principles are well established:
- Redundancy. More generating capacity than the load needs, so any one unit (N+1) or a whole set of units (2N) can be out of service without dropping load.
- Isolation. Faults are contained so a problem in one block cannot propagate to others. Separate fuel paths, separate controls, separate protection zones.
- Fast transfer. Uninterruptible power supplies and batteries hold the load for the seconds it takes to bring standby capacity online.
- Concurrent maintainability. Any unit can be taken offline for service without reducing supply below the load.
- Monitoring and response. Continuous condition monitoring, spares on site and a maintenance team that can act before a fault becomes an outage.
How CNP designs for data centres
CNP builds generation in modular blocks from 1 to 20 MW per site, co-located next to the data centre. For data centre supply, the plant is designed with full backup and redundancy rather than sized to the load alone:
- Generating blocks are specified with spare capacity so maintenance and faults do not reduce supply.
- Battery and UPS integration covers transfer between sources.
- The plant can operate fully off-grid, or alongside a grid connection where one exists.
- CNP operates and maintains the plant under a long-term agreement, with remote monitoring and site staff.
On that basis CNP offers contractual availability of up to 99.999% at the point of supply to the facility.
Why redundancy multiplies availability
A useful way to see why architecture matters: if two independent units can each carry the load, supply is lost only when both are unavailable at the same time. If each unit is available 98% of the time and failures are independent, the chance of both being down together is 2% × 2% = 0.04%, giving 99.96% availability from two ordinary units. Add a third, and the combined figure rises again.
Real systems are not perfectly independent — shared switchgear, controls or fuel can fail together — which is why good designs separate those too. But the principle holds: availability comes from how units are combined, not from any single machine. See N, N+1, 2N: power redundancy explained.
Six questions to ask any supplier quoting availability
An availability number is only useful with a definition attached. Before comparing offers, ask:
- Where is it measured? At the generator terminals, the plant switchboard or the facility's incoming supply?
- What counts as an outage? A full loss of supply only, or also voltage and frequency excursions outside an agreed band?
- Over what period? Annual availability can hide a single long outage; monthly measurement is stricter.
- Are planned outages excluded? If maintenance windows do not count, the design is not concurrently maintainable.
- What happens if the target is missed? A commitment without a remedy is an estimate.
- What redundancy sits behind it? Ask for the single-line diagram and the N+X configuration.
Where the grid fits
A grid connection, where one is available, is a useful additional source. It is not a substitute for redundancy onsite: grid availability in many regions is well below four nines, and a connection brings its own fault exposure. Many operators treat the grid as one more redundant path rather than the primary supply.
Next step
If you are planning a facility in the 1–20 MW range, we can take you through a redundancy design for your load profile. Assess your site.
