This article walks through that calculation in plain arithmetic. The figures used are illustrative, not guarantees for any particular machine, but the method is the same one engineers use to compare N+1 and N+2 layouts and to test whether a proposed plant matches the uptime the campus needs.
Availability and reliability are different numbers
Availability is the share of time a unit is able to run when needed. It is reduced by forced outages, when something breaks, and by planned outages, when the unit is taken down for inspection. Reliability describes how likely a unit is to keep running for a set period without failing, and starting reliability is the chance it starts successfully on demand.
For a prime power plant that runs all year, availability is the main input, because the units are already running. For a plant that starts on a grid event, starting reliability matters just as much. Our post on starting reliability for campus turbines covers that side; this article concentrates on running availability.
Gather the unit inputs
Start with an availability figure for one turbine generator package, including its auxiliaries. Manufacturer statistics, fleet data from industry databases and the operator’s own records are the usual sources. Separate forced outage time from planned maintenance time, because planned work can be scheduled when redundancy is available, while forced outages arrive at random.
Be honest about the auxiliaries. A turbine may be mechanically healthy and still unavailable because of a failed lube oil pump, a control fault or a blocked inlet filter. A package-level figure that includes these systems is far more useful than a core engine figure that ignores them.
The arithmetic for N, N+1 and N+2
Suppose the campus needs four turbines running to carry its load, and each unit has a forced availability of 97 percent. With only four installed, all four must be available, and the chance of that is 0.97 to the fourth power, about 88.5 percent. That leaves roughly 1,000 hours a year when the plant cannot carry full load, which is why nobody designs a prime power plant without spare units.
With five installed (N+1), the plant fails only when two or more units are out at once. The binomial calculation gives about 99.15 percent, or roughly 74 hours a year short of full load. With six installed (N+2), it rises to about 99.95 percent, near 4 hours a year. If unit availability drops to 95 percent, the same layouts give about 97.7 and 99.8 percent. The lesson is that the spare units and the quality of maintenance both move the result a long way.
Planned maintenance eats into redundancy
The arithmetic above assumes that every installed unit is available apart from random failures. In practice, each turbine also spends time in planned maintenance, from short inspections to major overhauls or engine exchanges. While one unit is down for planned work in an N+1 plant, the remaining units are running at N with no spare, and any forced outage during that window leaves the plant short.
That is why many prime power campuses choose N+2, or N+1 with a short engine exchange process so planned windows stay brief. The maintenance contract has a direct effect here: response times, spare engine pools and parts holdings set how long a unit is out. Our guide to maintenance contracts for data center turbines and the explanation of what a long-term service agreement covers both bear on this input.
Common-mode failures break the model
The binomial method treats each turbine as independent. Real plants have shared systems that can take out several units at once: a single fuel gas supply line, one gas metering station, a common control network, a shared busbar or a single step-up transformer. One failure in any of these can remove the whole plant, however many turbines are installed.
Estimate the availability of each shared system separately and multiply it into the result. A plant at 99.95 percent on turbines alone can fall well below that if its fuel supply has only one path. Designs that split busbars, duplicate gas trains and use redundant switchgear arrangements keep common-mode risk from overwhelming the benefit of spare turbines.
Partial capacity still counts
A plant that is one unit short is not the same as a plant that is dark. With battery storage, demand response from cooling plant or a controlled load shed of non-critical IT, the campus can ride through a partial shortfall for some time. The calculation should therefore report the probability of each capacity state, such as full load, full load minus one unit and so on, not just pass or fail.
Spinning reserve affects this too. If the plant runs one unit lightly loaded as a reserve, a trip can be absorbed without load loss. Our post on spinning reserve for campus turbines explains the trade between reserve margin and fuel efficiency.
Turning the result into decisions
Use the calculation to compare options on equal terms. Run it for N+1 and N+2, for different unit sizes and for different shared-system layouts. Smaller units mean more of them for the same load, which usually improves plant availability at a given redundancy level, at the cost of more equipment to maintain. Modular platforms such as the New USP&E FX-45 45MW Modular Gas Turbine for the Hyperscale Data Center sector in the USA are planned around this kind of block-by-block redundancy, with slots from 2028 to 2030.
Finally, write the assumptions into the contract. An availability guarantee is only meaningful if it states the unit figures, the maintenance windows allowed, how partial capacity is counted and which shared systems sit inside the contractor’s scope. Agreeing those definitions before signing prevents arguments later about whether a shortfall counts.
About the authors
This guide comes from the USP&E data center power and reliability engineering team. USP&E is an EPC and O&M contractor, not a broker, with 150+ power projects in 45+ countries, 25,000 MW contracted since 2002 and 24+ years in the field. Our overview of turbine plants built for data centers gives more context on plant layouts.
Offices: Colorado Springs (1755 Telstar Dr, Suite 300), Knoxville (800 S Gay St, Suite 700), London (162-168 Regent Street), Johannesburg (22 Kildoon Rd, Bryanston), Cape Town (31 Brickfield Road, Woodstock) and China (No. 1 Ningzhou Road, Wuxi). Regions: USA | AFRICA | UK | UAE | CHINA. ISO 9001 and ISO 45001 certified.
Gas turbine plant availability questions
How do you calculate gas turbine plant availability?
Take the availability of each unit, then use binomial arithmetic to find the chance that at least the required number of units is available, and multiply by the availability of shared systems.
What availability does an N+1 gas turbine plant achieve?
As an illustration, five units of 97 percent availability where four are needed give about 99.15 percent. Real results depend on unit data, maintenance and shared systems.
Is N+2 worth it for a data center plant?
Often yes for prime power, because planned maintenance on one unit leaves an N+1 plant with no spare. N+2 keeps a spare during maintenance windows.
What is a common-mode failure?
A single fault that removes several units at once, such as loss of the only fuel gas line or a shared busbar. It is not reduced by adding turbines.
What is the difference between availability and reliability?
Availability is the share of time a unit can run. Reliability is the chance it keeps running, or starts, without failing over a given period.
Should availability be guaranteed in the contract?
It can be, but only with clear definitions of unit data, allowed maintenance, partial capacity and which systems are in the contractor’s scope.
Test your plant layout
Send the campus load, the layout you are considering and the redundancy target, and our engineers will run the availability numbers for each option.
If a document set is easier than a conversation, send it to info@uspeglobal.com or upload a summary through the data center power enquiry page; to talk it over, dial +27 10 822 2324. Units that can form or extend a plant are listed in our gas turbine stock list.