Combined Heat and Power
in 2026: When Waste Heat
Changes the Answer

Every cogeneration proposal contains the same seductive arithmetic: one fuel purchase, two useful products, and a total efficiency no separate arrangement can match. The arithmetic is real, and it is conditional. It holds exactly as far as the site's thermal demand carries it, and not one degree further.

Section 01The arithmetic, and where the fleet actually lives

Combined heat and power is the oldest idea in distributed energy. Generate electricity at the point of use, and instead of rejecting the heat that generation inevitably produces, capture it and put it to work: process heating, steam, hot water, or, through absorption chillers, chilled water.¹ The U.S. Environmental Protection Agency states the headline result plainly: CHP can achieve total efficiencies of more than 80 percent, against roughly 50 percent for the conventional arrangement of grid electricity plus an on-site boiler, an arrangement in which nearly two thirds of the input energy is wasted before it does anything useful.¹

That efficiency delta is the entire commercial case for cogeneration, and it explains where the American fleet was built. When the U.S. Department of Energy released its updated CHP Installation Database, it counted 82.6 gigawatts of capacity across roughly 4,400 sites nationally as of the end of 2016.⁴ The composition of that fleet is the real lesson: cogeneration's natural habitat has always been refineries, chemical plants, paper mills, food processors, hospitals, and district systems, hosts whose appetite for heat runs around the clock and around the calendar. The technology did not spread to those industries by coincidence. It spread there because that is where the second product has a buyer every hour of the year.

Two honest caveats belong in the opening section rather than the fine print. First, most CHP classes burn fuel on site, which commits the owner to combustion permitting, fuel-price exposure, and rotating-machinery maintenance for the life of the plant; the efficiency premium is compensation for taking on those obligations, not a free good. Second, the headline efficiency is an accounting identity, not an operating fact. A machine is credited with 80 percent only when the heat it recovers is actually absorbed by a real load. The rest of this paper is about that condition.

Section 02Efficiency you can bank versus efficiency on the nameplate

The accounting rule that separates real cogeneration from brochure cogeneration is short: recovered heat counts when, and only when, it displaces energy the site would otherwise have purchased. Recoverable heat is a property of the machine. Used heat is a property of the site. Total efficiency belongs to the pair, never to the equipment alone.

The valuation discipline that follows is what we call the boiler-displacement test. A unit of recovered heat is worth the fuel the site's boiler would have burned to produce that same unit, at the boiler's actual seasonal efficiency, at the site's actual delivered gas price. As an illustrative figure only: a boiler operating at 80 percent seasonal efficiency means each unit of recovered heat absorbed by a real load displaces 1.25 units of purchased boiler fuel. That multiplier is the entire thermal side of the business case. And it has a hard zero in it: the same unit of heat rejected through a dump radiator displaces nothing, is worth nothing, and drags the plant's true efficiency back toward that of a simple generator.

This is why identical machines produce wildly different economics at different addresses, and why a cogeneration model that was not built from this site's thermal records was not built for this site at all. The electrical side of a CHP model travels well from project to project. The thermal side never does.

Recovered heat that displaces no purchased fuel is not efficiency. It is exhaust routed through more expensive pipe.

Section 03Coincidence: the same hour, not the same year

The most misleading number in cogeneration analysis is the annual thermal total. A site that uses a large amount of heat across a year can still be a poor host if the heat arrives in the wrong hours, because heat, unlike money, does not bank well. Commercial thermal storage exists and occasionally rescues a marginal profile, but for most projects the working rule stands: the heat must be wanted in roughly the same hour the machine makes it.

The analytical tool is a thermal load duration curve: every hour of the year's heat demand, sorted from the hungriest hour to the emptiest. A generator running baseload draws a nearly flat recoverable-heat line across that chart. The value of cogeneration lives in the region where the demand curve sits above the recovery line. Everything below the line is heat with nowhere to go. Everything above it is boiler fuel the site still buys. Sizing a CHP plant is, at bottom, the art of placing that one line honestly.

The strong hosts are boringly consistent, which is itself useful information. Hospitals draw sterilization steam, domestic hot water, laundry, and reheat every month of the year, day and night, in heating season and cooling season alike. Food and beverage plants run clean-in-place cycles, wash-down, cooking, and pasteurization on production schedules rather than weather. Process industries dry, distill, render, and pasteurize around the clock. Campus district loops aggregate dozens of small loads into one steadier one, subject to the seasonal caveat in Section 06. The weak hosts are equally consistent: buildings whose thermal demand is comfort heating alone, which in most of California means a modest load concentrated in winter mornings; facilities that go dark for the summer; any site whose heat demand is chiefly an artifact of the coldest few hundred hours of the year.

The evidence standard is measurement, not recollection. Monthly boiler fuel logs are the minimum. Metered steam or hot-water profiles by hour are the standard where the decision is close. An annual gas bill split between process and comfort by someone's estimate is a screening input, not a basis for capital. In our study work, the thermal record is requested on day one, and the depth of what comes back is itself diagnostic: a site that cannot document its heat load yet is not ready to size a machine against it.

Section 04Grade: heat is not one commodity

Coincidence in time is necessary and not sufficient, because heat is not fungible across temperature. Water at 180 degrees Fahrenheit, steam at 15 pounds of pressure, and steam at 150 pounds are three different products. A machine that produces one cannot serve a process that requires another, no matter what the efficiency slide claims. Every prime-mover class has a thermal signature, and the signatures differ from one another far more than the electrical specifications do.²

The mismatches run in both directions. Selling a hot-water machine into a high-pressure steam site strands the load. Buying turbine-grade exhaust and degrading it into warm water serves the load but strands the capital that paid for the grade. Neither error survives a study that lists the site's thermal services by required temperature and pressure before any equipment class is discussed, which is the order of operations that matters.

One further comparison belongs in every modern study, and sellers of heat-recovery equipment rarely volunteer it: where the thermal demand is low-temperature, an industrial heat pump served by grid or on-site electricity is a direct competitor to recovered heat, and in California's mild climates it can be a strong one. A site with a credible electrification path for its low-grade heat should see that path priced on the same page as cogeneration. A thermal case that cannot beat the electrified alternative in writing was assembled, not analyzed.

Section 05Absorption cooling: turning heat into cold, carefully

For summer-peaking sites, the maneuver that can rescue coincidence is absorption cooling: thermally driven chillers that produce chilled water from heat rather than from a compressor motor. The federal fact-sheet numbers frame the physics of the match.³ Single-effect machines are driven by hot water at roughly 200 to 240 degrees Fahrenheit or low-pressure steam near 15 pounds, which is precisely the grade a reciprocating engine's jacket circuit produces, and deliver on the order of 0.7 units of cooling per unit of driving heat. Double-effect machines require hotter input, roughly 350-degree water or steam near 115 pounds, turbine territory, and convert at better than one to one. Direct exhaust-fired designs typically require exhaust above roughly 750 degrees.³

The case for it is precise. A California site whose heating demand collapses in May can keep its machine thermally honest by converting surplus heat into chilled water during exactly the months when electric cooling is the site's largest and most expensive load. Absorption converts a seasonal liability into displaced chiller demand, extends the thermal duration curve across the calendar, and pairs naturally by temperature grade: single-effect with engine hot water, double-effect with turbine steam.

The case against it is equally precise. The chiller, its larger cooling tower, the water treatment, and the tower's real water consumption are real capital and real operating burden, in a state where water is never free. A coefficient of performance near 0.7 means a unit of driving heat produces less than a unit of cooling, where a modern electric chiller produces multiples of its input with far less of the site's attention. The arithmetic works when the driving heat is genuinely surplus, priced near zero at the margin. It fails the moment anyone proposes running the generator harder in order to feed the chiller, at which point the site is burning purchased fuel to make cooling at a fraction of the efficiency of the electric machine it already owns. The sizing rule that keeps projects honest is one sentence: size the absorption machine to the heat you cannot otherwise use, and never size the generator to the chiller.

Section 06The failure cases, named

Cogeneration's failure modes are not exotic. They are the same five, project after project, and every one of them is visible during the study phase if anyone is paid to look.

The summer dump. A plant sized to January's heating load spends May through October rejecting heat through dump radiators. For those months the site owns a generator with cogeneration capital costs and simple-generation efficiency. If the summer has no credible heat sink, whether process load or a soberly priced absorption case, the honest design is a smaller machine, and the honest study says so.

The oversize trap. Plants sized to the electric bill, or to the next frame size in a catalog, rather than to the thermal base. The reliable sizing anchor is the slice of thermal demand present in most hours of the year, not the winter peak and not the electric peak. Oversized machines cycle, run at part load where every technology class gives up efficiency, or dump heat. In cogeneration, the smaller machine running flat out is very often the better investment than the larger machine running embarrassed.

The grade mismatch. Section 04's error, contracted for. It is discovered at commissioning, when the process still needs the boiler at full fire because the recovered heat arrives a full temperature grade below what the process accepts, and the pro forma quietly stops being circulated.

The phantom baseline. Displaced boiler fuel is only worth claiming if the boiler was truly going to burn. Crediting heat against a boiler scheduled for retirement, against a process slated for electrification, or against a load the process cannot actually accept at the delivered temperature inflates the thermal value with fuel that was never going to be purchased. The baseline is the site's documented forward plan, not its past.

The resilience conflation. A cogeneration plant is not backup power unless it was engineered to island: paralleling switchgear, load shedding, black-start provision, and testing, all specified and paid for. Many CHP installations trip offline with the grid by design. Outage service is a legitimate requirement and a real engineering scope; it is not a complimentary feature of efficiency. A proposal that implies otherwise is borrowing value from a product it has not included.

One operational truth underlies all five: the boiler does not leave. It remains for machine outages, thermal peaks, and turndown floors. Cogeneration displaces boiler fuel, never the boiler, and a model that deletes the boiler's standby costs has overstated the project before the first hour of operation.

The screening discipline compresses into six questions, and they are answerable from records the site either has or does not:

1. Is there a documented thermal baseload?The emptiest month sets the honest machine size, and it must come from fuel logs or meters, not an annual bill divided by twelve.
2. Do heat and power coincide by the hour?A thermal duration curve against a flat recovery line, with the stranded area priced at zero, not at hope.
3. Does the machine's thermal grade match the load's?List the site's thermal services by required temperature and pressure first; admit equipment classes second.
4. Where does the summer heat go?A named process sink, a soberly priced absorption case, or an admitted dump. One of the three, in writing.
5. Is the displaced boiler real and staying?Thermal value is claimed against the site's documented forward plan, including any electrification intent, not against last year.
6. Is islanding specified or merely implied?If outage service matters, the switchgear, shedding scheme, and testing are in the scope and the price. If they are not, the plant rides through nothing.

Section 07When waste heat changes the answer

The 2026 context sharpens all of this. Organizations across California are studying on-site generation for a reason earlier papers in this series document at length: the interval between requesting grid capacity and receiving it has become a planning constraint in its own right, and self-supply is increasingly evaluated on speed as much as on price. Waste heat bears on that evaluation in one direction only. A genuine thermal host, one that clears the six questions above, converts a marginal generation case into a strong one, because the same fuel purchase is suddenly doing two jobs with documented buyers for both. A site without real thermal demand should evaluate generation as generation, compare it honestly against storage, grid options, and waiting, and let the cogeneration premium go. Bolting heat-recovery hardware onto a project whose heat has no home buys efficiency optics, not efficiency.

On incentives, the discipline is the same as everywhere in this series. Under current federal law as of August 2026, the investment tax credit for qualifying property is 30 percent (26 U.S.C. §48 and §48E, as applicable to the specific property and its construction timing); statutory adders exist but each must be individually qualified for the specific project, and none should be assumed in a screening model.⁵ How thermal-side equipment is classified for credit purposes is itself a technical question. Engage qualified tax counsel before a single incentive dollar enters the pro forma. Combustion-based CHP also permits through the local air district, and district treatment varies enough across California that it is a site fact to be established early, not a footnote; non-combustion classes face a materially different and generally lighter air-permitting path, which is part of their honest case.

The conclusion is the paper's opening sentence run in reverse. A machine that makes two products is only better than a machine that makes one if somebody at the site is standing there, every hour, wanting the second product at the temperature it arrives. Whether that somebody exists is knowable in advance, from records, for a bounded fee, before capital moves. That is the study. We prepare it from the owner's side, at a fixed price, with no equipment margin behind the answer, because the question of where the heat goes deserves an answer from a party with nothing riding on it.