Reciprocating Engines:
The Honest Case For and Against

Nobody walks into a boardroom in 2026 eager to recommend a reciprocating engine. That is exactly why the machine deserves a careful file: on several dimensions that decide real projects it remains the hardest option to beat, and in California's strictest air districts it can be the hardest one to permit.

Section 01The machine nobody pitches

The reciprocating internal combustion engine is the least glamorous item on the on-site power menu. It burns fuel, it vibrates, it needs oil changes, and it has not been the subject of an enthusiastic keynote in decades. In a state whose energy conversation is organized around decarbonization, the engine is the technology that other proposals define themselves against.

It is also the most produced prime mover in industrial history. The U.S. Environmental Protection Agency's catalog of combined-heat-and-power technologies puts worldwide reciprocating engine production above 200 million units per year, with stationary generating sets available from tens of kilowatts to more than 18 megawatts per unit.¹ That production base is not trivia. It is the root of nearly every commercial advantage the engine still holds: low first cost, short delivery, and a service bench deeper than any alternative's.

This series prices every credible path to power for a specific site, with sourced and dated inputs, before any technology is chosen. This paper is the engine's file, with both columns filled in. The discipline matters here more than almost anywhere, because engines attract two lazy analyses: the seller's, which stops at the capital cost, and the fashionable one, which stops at the word combustion. A site with a deadline can afford neither.

Section 02The case for: capital, delivery, and the service bench

Start with first cost, because every engine proposal does. In the EPA's technology characterization, installed costs for natural gas engine CHP plants ranged from roughly $2,900 per kilowatt at the hundred-kilowatt scale down to about $1,433 per kilowatt for the largest single-unit class surveyed, in 2013 dollars.¹ Those dollars are dated and must be repriced for any current decision; the anatomy they describe is not. The same catalog is direct about the cross-technology comparison: below roughly 20 megawatts, engine generating sets generally carry lower upfront cost than gas turbine packages of comparable size, and the engine's electrical efficiency advantage at those sizes compounds the point.¹

Utility-scale surveys tell the same story from a different direction. The U.S. Energy Information Administration's February 2020 cost study estimated a four-engine, 21-megawatt natural gas plant at $1,810 per kilowatt overnight, in 2019 dollars, with a heat rate of 8,295 Btu per kilowatt-hour.² What survives the repricing of any dated survey is the ranking: among machines an owner can put behind a fence and dispatch on demand, the engine has generally been the cheapest ticket to a firm kilowatt. That ranking is a starting point for analysis, not a verdict. The rest of this paper is about what the ranking leaves out.

Delivery is the quieter advantage. Engines come off standing production lines that serve global markets for generating sets, marine propulsion, and industrial drives. They are not custom fabrications, and their order books have not stretched the way heavy-frame turbine slots have in the current cycle. In a market where the binding constraint is the calendar, equipment that ships on industrial timelines rather than multi-year ones is not a detail. It is often the reason the path exists at all.

Then there is the bench. Nearly every county in California has mechanics, dealers, parts inventories, and rebuild shops that know this machine, because the same architecture powers standby fleets, water systems, and off-road equipment everywhere. The EPA catalog prices full-service maintenance contracts for gas engines at roughly 1.0 to 2.5 cents per kilowatt-hour, a figure that exists only because a competitive service market exists.¹ Newer technology classes can be excellent machines and still leave an owner dependent on a single service organization. The engine's service risk is diversified by a century of installed base, and for an owner pricing twenty years of operation that difference belongs in the model, not the footnotes.

Section 03The case for: operating shape

Engines tolerate real load. In the EPA's characterization, a lean-burn natural gas engine at half load gives up roughly 8 to 10 percent of its full-load efficiency; gas turbines described in the same catalog give up 15 to 25 percent at the same point.¹ Loads that swing, batch processes, and campuses with strong day-night shape are the engine's natural habitat. A machine that is efficient only at 100 percent load is efficient only in the proposal.

Engine plants are usually built as fleets of identical units rather than one large machine, and the arithmetic of fleets is forgiving. One unit down for service takes a fraction of capacity offline instead of all of it. Survey data in the EPA catalog put natural gas engine availability near 96 to 98 percent across size classes, with multi-unit facilities configured above 99 percent.¹ Availability is not a brochure virtue; it is the design variable that decides whether a maintenance week is an event or a non-event.

Engines also start fast and start dark. They need little more than batteries or compressed air to come up, which is why they have anchored black-start and emergency service for decades.¹ In hybrid designs, from engine-plus-storage portfolios to islandable campuses, that trait does work no efficiency number captures: it is the difference between a plant that rides through a grid event and one that waits for the grid to come back.

Efficiency closes the operating case. Natural gas engine electrical efficiency runs from about 30 percent on a lower-heating-value basis for the smallest units to over 46 percent for large lean-burn machines, and the heat the engine rejects arrives in two convenient streams, jacket water and exhaust, that make hot water and low-pressure steam.¹ Hosts with year-round low-grade thermal demand, such as food processors, institutional campuses, and treatment facilities, can push total fuel utilization into the 70-percent range in well-matched service.¹ The qualifier does the work in that sentence: recovered heat has value only where a real thermal load displaces real purchased fuel.

Section 04The case against: the permit is the project

Now the other column. Everything the engine earns on capital and calendar, it can spend at the air district, and in California's strictest districts it can spend more than it earns. The mechanism deserves to be stated precisely, because it is the single most decision-relevant fact in this file.

The federal floor comes from the new-source performance standards for stationary spark-ignition engines. A new non-emergency natural gas engine of five hundred horsepower or more must meet 1.0 gram of NOx per horsepower-hour, roughly 82 parts per million by volume at 15 percent oxygen, alongside CO and VOC limits.⁴ Much of the country permits engines against something close to that floor.

California's strict districts are not close to that floor. The South Coast district's engine rule applies to every stationary engine above 50 brake horsepower and caps NOx at 11 parts per million, VOC at 30, and CO at 250, corrected to 15 percent oxygen.³ The rule's 2019 amendment goes further in a way its own structure makes plain: the tightest compliance table in the rule is titled "Alternative to Electrification."³ The direction of travel is not subtext. It is a heading.

Meeting numbers like those on a lean-burn engine means selective catalytic reduction plus an oxidation catalyst, and in sensitive areas a continuous emissions monitoring system; the EPA catalog notes that engine systems offered into Southern California and the northeastern United States are configured exactly this way.¹ Each element is proven, and each adds capital, urea logistics, catalyst change-outs, monitoring overhead, and new failure modes. Aftertreatment narrows, without erasing, the capital advantage the engine walked in with. A study that prices the bare generating set in a strict district has priced a machine that cannot be permitted there.

The engine's advantage is bought at the equipment counter and spent at the permit counter. Which balance survives depends on the district.

California has also long maintained certification standards for distributed generation at levels near 0.07 pounds of NOx per megawatt-hour, set in 2007. Engines have been certified at that level, carrying full aftertreatment; non-combustion technologies such as fuel cells sit near such levels by architecture rather than by cleanup.¹ That asymmetry, cited without drama, is the honest core of the permitting case: the engine can usually get to the required number, but it pays for the trip in equipment, in operating burden, and above all in permit calendar.

The calendar is the deeper cost. In a strict district, a new engine permit is an engineering exercise, a modeling exercise, and sometimes a public one; near sensitive receptors it can trigger health-risk screening for air toxics such as formaldehyde. None of that is exotic, but all of it takes months the owner thought the engine was saving. District treatment is also the least portable fact in this file: the same machine that permits routinely in one California district faces explicit electrification pressure in another. That is why a defensible study names the district before it prices the machine.

One boundary keeps sales conversations honest. The emergency-engine regime familiar from standby fleets is not a prime-power lane: under the federal rules, an emergency engine may run maintenance checks and readiness testing up to 100 hours per calendar year, with unlimited operation only in genuine emergencies.⁴ An engine intended to run for revenue, bridging, or scheduled service is a non-emergency engine, and it carries the full permitting weight described above.

Section 05The case against: neighbors, calendars, and carbon

The against column continues past the stack. Engines make low-frequency noise and structure-borne vibration, the kind that travels through slabs and reaches neighbors at frequencies a fence does not stop. The remedies are standard practice, acoustic enclosures, exhaust silencers, isolation mounts, setbacks, and none is free. Near housing, noise conditions can shape the site plan as much as the air permit does. Turbines emit higher-frequency noise that attenuates more readily; fuel cells and batteries hum. On this dimension the engine is the loudest class on the menu and should be modeled as such.

Maintenance is a calendar, not a contingency. The EPA characterization puts routine service, meaning oil, filters, and spark plugs, at every 500 to 2,000 operating hours; a top-end overhaul at 8,000 to 30,000 hours; and a major overhaul at 30,000 to 72,000 hours.¹ Run an engine near-continuously, on the order of 8,000 hours a year, and the top-end lands within the first few years of service and a major rebuild in mid-life. Fleet design absorbs this and the mature service market prices it, but the owner should see the outage calendar and the rebuild reserve in the model on day one. A proposal showing twenty years of uninterrupted output has not shown an engine; it has shown a brochure.

The carbon ledger is the heaviest line. At the heat rate in EIA's 2020 survey, 8,295 Btu per kilowatt-hour, and EIA's natural gas factor of 117 pounds of CO2 per million Btu, simple arithmetic gives roughly 0.97 pounds of CO2 per kilowatt-hour before any credit for recovered heat displacing boiler fuel.² That is a derived, illustrative figure, and site math will differ; the point is its order of magnitude. An engine is a combustion commitment for its service life, and a board with public climate targets should reconcile that commitment in writing before it signs, not after.

Two further postures deserve daylight. First, lean-burn engines pass a fraction of fuel methane through unburned, and the oxidation catalysts that scrub CO and organic compounds convert perhaps 60 to 70 percent of methane at best.¹ The pollutant ledgers engines are permitted against count non-methane organics, while the methane itself is a potent greenhouse gas; an honest emissions posture says so. Second, the regulatory trajectory in California's urban districts points one way. Rules tighten. An owner underwriting a long engine life should price the possibility of retrofit obligations arriving mid-life, and renewable gas attributes, where offered, are a procurement choice with cost and diligence of their own, never a default assumption.

Incentives rarely rescue this ledger. Under current federal law the investment tax credit is 30 percent for qualifying clean-energy property, and conventional gas-fired engine plants generally sit outside the qualifying categories; statutory adders, where relevant at all, must be individually qualified, never assumed.⁵ Treat any incentive line in an engine proposal as a question for qualified tax counsel rather than a modeling default.

Section 06The scorecard, and where engines genuinely win

Held honestly, both columns produce a decision profile rather than a verdict. The table below is the scorecard we defend.

Engines genuinely win, despite the fashion, in five recurring profiles:

• The deadline site with gas in the ground. Where firm gas service exists, the district screen prices in, and the alternative is years of waiting for wires, the engine is often the only path that meets the date at reasonable capital.
• The thermal host. Steady hot-water or low-pressure steam demand converts the engine's waste streams into avoided fuel, and the economics stop depending on the electric line alone.
• The swing load. Shifts, batches, and seasonal duty fit the engine's part-load tolerance and multi-unit staging better than they fit machines that want to run flat.
• The resilience anchor. In islandable portfolios the engine's black-start and fast-start behavior is the keystone even when another technology carries the energy; a small engine fleet is frequently what lets an elegant portfolio actually stand up dark.
• The service-risk pricer. Owners who weight twenty-year serviceability, second sources, and local labor find in the engine the only class where every one of those markets is already deep.

And engines should lose, on the merits, in four:

• The strictest-district urban site near sensitive receptors, where permit calendar, operating conditions, and community process erode the advantages one by one.
• The flat, around-the-clock load with no thermal host, where a higher-efficiency or non-combustion path compounds its advantage over every operating hour.
• The decarbonization-led decision, where the binding constraint is the owner's emissions commitment; a combustion machine at any price is the wrong tool for that job.
• The site without firm gas. An engine without gas service trades the electric queue for the gas queue, and sometimes that trade is worse.

Section 07The decision discipline

The engine file closes the way it opened: this is a machine that rewards honest accounting and punishes fashionable accounting in both directions. In California the screen order matters. District treatment first, because it is the constraint that moves the whole answer; then thermal fit; then the outage calendar and redundancy design; then lifecycle operating cost at honest hours; then the carbon ledger the owner actually carries, written down where the board can see it.

Run that order and the reciprocating engine will sometimes be the right answer, stated without apology, and sometimes the wrong one, stated without euphemism. Both findings are worth exactly what the analysis behind them is worth. Ours is independent, prepared on the owner's side of the table, and paid the same fixed fee whichever machine wins. That is the only posture from which a paper like this one can be written.