Who Runs the Plant
at 2 a.m.

Most energy studies end at commissioning. The plant then runs for twenty years. An owner who priced the asset carefully and the operation casually has bought the smaller half of the decision.

Section 01The second decision

The capital decision gets the attention. It is discrete, it has a number, and it ends. The operating decision is continuous, it has no single number, and it does not end. It is also the decision that determines whether the asset performs the way the model said it would.

This is not an argument against owning generation. Customer ownership is often the right structure, and the reasons are sound: no take-or-pay obligation, no margin stacked on the energy, direct control of the asset. But ownership is a package. The capacity and the maintenance arrive together, and only one of them appears on the capital line.

The useful reframing is this. An owner is not deciding whether to buy equipment. An owner is deciding whether to add a function to the organization, and then deciding how much of that function to perform in house. Those are two separate questions and they deserve separate answers.

Section 02What a maintenance number actually covers

Non-fuel operations and maintenance is usually quoted in cents per kilowatt-hour. The figure looks precise. It is rarely comparable across proposals, because the scope boundary moves, and the boundary matters more than the number.

The clearest published illustration comes from the U.S. Department of Energy technology fact sheets. For fuel cell systems in combined heat and power service, contracted maintenance is estimated at 0.7 to 2.0 cents per kilowatt-hour when stack replacement and refurbishment are excluded. For the same representative systems, maintenance is 3.6 to 4.5 cents per kilowatt-hour once stack replacement and refurbishment are included.¹

Nothing dishonest has happened. Both figures are correct, and the same document sets out the overhaul events behind the difference: shift catalyst replacement every three to five years, reformer catalyst replacement around five years, and stack replacement every five to ten years.¹ The problem is that a proposal quoting the lower figure and a proposal quoting the higher one are describing different products, and an owner comparing them side by side will read a difference in quality that is really a difference in scope.

The same discipline applies to every technology. The DOE fact sheet for reciprocating engines reports non-fuel maintenance falling from 3.0 cents per kilowatt-hour at the smallest packaged sizes in its table to 1.3 cents at the largest, with selective catalytic reduction adding roughly 0.25 cents where it is fitted.² The gas turbine fact sheet reports 1.5 cents falling to 0.9 cents across its range, with catalytic reduction adding 0.20 to 0.25 cents.³ The microturbine fact sheet reports 1.9 to 1.6 cents.⁴

Two cautions on those numbers, both of which a study should state rather than bury. First, they are dated. The engine, turbine, and microturbine sheets were refreshed between 2023 and 2024, while the fuel cell sheet dates from 2016.^1,2,3,4 A cost comparison drawn from documents years apart in vintage is a starting point for questions, not a conclusion. Second, the emission-control line is not decoration. Where a control device is required, its maintenance is a permanent addition, and in California it frequently is required.

A third published benchmark is worth knowing and worth handling carefully. The Energy Information Administration publishes fixed maintenance costs in dollars per kilowatt-year for utility-scale generating technologies. In the current edition, simple-cycle combustion turbines carry $6.87 to $9.56 per kilowatt-year, combined cycle $12.12 to $15.51, battery storage $40.00, solar photovoltaic with tracking $20.23, and geothermal $150.60.⁵ Those figures describe central-station plants built and staffed as merchant power projects. A site-scale plant behind a customer meter does not inherit those economies, and an owner who applies them directly will understate the cost. They are useful for the shape of the differences between technologies, not for the level.

Two proposals quoting different maintenance rates are usually describing different scopes. Read the exclusions before you read the price.

Section 03The arithmetic of a continuously manned post

Before choosing a staffing model, it helps to know what the most demanding one costs in headcount, because that number surprises people.

A year contains 8,760 hours. A single position staffed at all times therefore requires 8,760 person-hours of coverage. A full-time employee is nominally available for about 2,080 hours a year, and materially fewer after vacation, holidays, illness, and training. On the nominal figure, one continuously manned post requires about 4.2 employees. On a realistic availability figure the requirement is closer to five, and that is for one person on site at a time, with no allowance for a second person during work that safe practice will not permit alone.

This is arithmetic, and it is offered here as arithmetic rather than as a cost estimate for any particular site. But it reframes the question usefully. The question is not whether an organization can hire an operator. It is whether the organization intends to build a small department, with a supervisor, a call rotation, a training budget, and a succession problem when the one person who understands the plant retires.

Many sites conclude that they do not, and that conclusion is entirely respectable. The point of running the arithmetic is to reach it deliberately rather than to discover it in year two.

Section 04Four staffing models, honestly compared

There are four common structures. Each is a real answer for some sites and a poor answer for others.

One structural note applies to all four. Whoever performs the work, the regulatory obligations sit with the owner or operator of the equipment. A contract can transfer the labor. It does not transfer the duty.

Section 05The compliance work is a job, not a checkbox

Owners with standby generators often assume their existing experience transfers to a plant that runs continuously. It does not, and the gap is larger than most expect.

Federal air rules for stationary reciprocating internal combustion engines set explicit work practices with explicit intervals. Under the relevant table in 40 C.F.R. Part 63, Subpart ZZZZ, emergency compression-ignition engines must have oil and filter changed every 500 hours of operation or within one year plus 30 days of the previous change, whichever comes first; the air cleaner inspected every 1,000 hours or on the same annual cycle; and all hoses and belts inspected every 500 hours or annually, replaced as necessary. For spark-ignition units in the same table, spark plug inspection runs on a 1,000-hour or annual cycle, and other categories in the table carry 1,440-hour and 4,320-hour intervals depending on engine type.⁶

The rule also provides a route that a well-run operation should know about. An owner may use an oil analysis program to extend the specified oil and filter change interval, performed at the same frequency as the change it replaces, analyzing at minimum total base number, viscosity, and water content. The condemning limits are specified: total base number below 30 percent of the value when the oil was new, viscosity changed by more than 20 percent, or water content above 0.5 percent by volume. If any limit is exceeded, the oil and filter must be changed within two business days.⁶ This is a genuine cost lever. It is also a records obligation, because the analysis program must form part of the maintenance plan and the results must be kept.

Which is the general pattern. The same subpart requires the engine and any after-treatment device to be operated and maintained according to the emission-related written instructions of the manufacturer or under a maintenance plan the owner develops, requires a non-resettable hour meter in defined cases, and sets out what records must be kept.⁶ None of that is difficult. All of it is somebody's job every week, and if that person is not named, it does not happen.

California adds a second layer, and the threshold is hours

Local air district rules are where continuous operation separates sharply from standby operation. Under San Joaquin Valley Rule 4702, an engine rated at 1,000 brake horsepower or greater that is permitted to operate more than 2,000 hours per calendar year, or that carries an external emission control device, must install, operate, and maintain continuous monitoring equipment for oxides of nitrogen, carbon monoxide, and oxygen, or an alternate monitoring arrangement approved by the district. The rule contemplates a continuous emissions monitoring system, a parametric system, or an approved alternative.⁷

The same rule requires an engine operating log maintained monthly, covering total hours of operation, fuel type, maintenance or modifications performed, monitoring data, and compliance source test results, retained for at least five years and made available to the district on request. Compliance source testing runs at least once every 24 months for the engines it covers.⁷

Two observations follow. First, 2,000 hours is roughly 23 percent of the year. A plant intended to carry real load crosses that line early, which means an owner comparing a standby installation to a continuously operating one is comparing two regulatory regimes, not two sizes of the same one. Second, a continuous monitoring system is itself equipment. It is calibrated, it drifts, it fails, and it generates data that someone must review before the district does. It belongs in the maintenance plan and in the maintenance budget.

Rule 4702 is cited here because it is public, specific, and representative of how California districts approach the question. It is not the rule everywhere. Each district writes its own, thresholds differ, and the requirements applicable to any particular site must be confirmed with that district before they are relied upon.

For emergency and standby systems specifically, NFPA 110 is the standard that typically governs periodic inspection, exercise, and record retention, with applicability determined by the authority having jurisdiction.⁸ It is a useful reference and a poor analogue. A standby system is tested so that it will start. A continuously operating plant is monitored so that it will keep running within permitted limits. The obligations differ in kind, not in degree.

Section 06Who is permitted to touch the equipment

A staffing plan that ignores the electrical safety rules will not survive contact with the site.

In California, the high-voltage electrical safety orders define high voltage as a sustained voltage of more than 600 volts. A qualified electrical worker is defined as a qualified person who, by reason of a minimum of two years of training and experience with high-voltage circuits and equipment, has demonstrated by performance familiarity with the work to be performed and the hazards involved. A qualified person, more broadly, is one who by reason of experience or instruction is familiar with the operation to be performed and the hazards involved.⁹

The practical consequence is direct. A capable maintenance technician who keeps the mechanical systems of a plant in excellent order may not be permitted to work on the switchgear that connects it. Whether the interconnection sits above or below 600 volts is therefore not only an engineering detail. It shapes who the owner must employ or contract, what training must be funded, and how long a fault takes to clear at night.

This is a question to settle during the study, not after commissioning, because the answer can change the preferred configuration.

Section 07Each technology asks for something different

Operating burden is a real axis of comparison, and every technology has an honest case on it.

Reciprocating engines. The case for: the highest density of qualified service labor of any option, well-understood overhaul cadences, parts available from many suppliers, and a failure mode that is usually gradual and visible. The case against: the highest routine touch of the group, the highest published non-fuel maintenance cost per kilowatt-hour at small sizes, consumables that must be scheduled and recorded, and emission-control equipment that adds both cost and monitoring obligations.

Gas turbines and microturbines. The case for: lower published non-fuel maintenance cost per kilowatt-hour, fewer routine interventions, and a compact operating envelope a small team can supervise. The case against: overhauls are specialist events rather than routine work, hot-section work generally leaves the site, the pool of qualified providers is narrower, and lead time on a major component is a scheduling risk that a site with one unit feels acutely.

Fuel cells. The case for: few moving parts, low local emissions, quiet operation, and a maintenance profile that does not depend on a mechanic being present during routine running. The case against: the major maintenance events are scheduled replacements of catalysts and stacks on multi-year clocks, they are capital in scale rather than operational, and as Section 02 showed, they are exactly what a low quoted maintenance rate is most likely to exclude.¹

Solar and storage. The case for: the lowest operating labor of any generating option, no fuel supply to manage, no combustion permit, and nobody required on site at 2 a.m. The case against: inverter replacement, storage capacity maintenance or augmentation, warranty administration, and performance monitoring are real recurring obligations, and the published fixed maintenance cost for storage is not trivial relative to other technologies.⁵

Continuing to buy from the grid. The case for: no plant staff, no permits, no overhauls, and no operating risk transferred to the organization. A paper about maintenance has to state this plainly, because for a site whose real constraint is cost rather than capacity or timing, the operating burden of ownership can exceed the savings that motivated it. The case against: it leaves the owner exposed to the capacity and timing constraints that started the conversation, and it forfeits control over the schedule.

None of these is a verdict. They are the trade-offs a site-specific study exists to resolve against that site's load, labor market, permit posture, and tolerance for operating risk.

Section 08Ten questions to answer before capital moves

1. What does the quoted maintenance rate exclude?Ask for the exclusion list in writing. Major overhauls, catalyst and stack replacement, emission-control consumables, and monitoring-system upkeep are the usual omissions.
2. Which staffing model are we actually choosing?Continuous, day shift with on-call, full contract, or hybrid. Name it before signing, and price the one named.
3. How many people is that, honestly?Run the coverage arithmetic against realistic availability rather than nominal hours, and include the supervisor.
4. Who responds at 2 a.m., and how long do they take?Get the response commitment and the escalation path in writing, and confirm the staffed hours of the provider rather than assuming them.
5. How many hours a year will we operate?This number determines which air-district monitoring and testing obligations attach. It is a compliance input, not only an economic one.
6. What monitoring equipment is required, and who maintains it?A continuous or parametric monitoring system is an asset with its own calibration, failure, and data-review burden.
7. Who keeps the records, and where?Retention periods run for years. Assign the task to a role rather than a person, and confirm the records survive that person's departure.
8. Is the interconnection above 600 volts?If so, identify now who is qualified to work on it and what that costs, whether employed or contracted.
9. What is the parts and lead-time exposure on a single-unit site?Ask what a major component failure means in weeks, and what inventory or pooling arrangement shortens it.
10. What happens at contract renewal?Model the position the organization will be in when the incumbent provider is the only party that knows the plant.

Section 09The operating plan belongs in the decision

An energy study that compares capital costs and stops has answered the easier question. The harder one is what the organization looks like after the plant is running: who is on the payroll, what is on the calendar, which records must exist, and what happens on the night something fails.

Those answers change the comparison. A technology with a higher installed cost and a lower operating burden can be the better decision for an organization with no appetite to build an operating function. A technology with the lowest headline maintenance rate can be the worst decision if that rate excludes the events that dominate lifetime cost. And for some sites, the honest conclusion of the operating analysis is that the plant should not be owned at all.

The right time to reach any of those conclusions is before the capital commitment, on paper, with the exclusions written down and the headcount counted. That is what a bounded, technology-neutral study is for.