Fuel Cells:
The Honest Case For and Against

No machine in distributed energy attracts more motivated reasoning than the fuel cell: the companies that build them describe the future, and the companies that sell competing equipment describe an overpriced experiment. Both descriptions are sales documents, so this paper prices the honest case instead, in both directions.

Section 01The machine, stated plainly

A fuel cell makes electricity by oxidizing fuel electrochemically, at an electrode, with no flame anywhere in the power block. That single fact drives nearly everything else in this paper, on both sides of the ledger. The pollutants that regulators police most aggressively in on-site generation, oxides of nitrogen above all, are creatures of flame temperature; remove the flame and they mostly never form. The U.S. Environmental Protection Agency's catalog of combined-heat-and-power technologies characterizes exhaust signatures for these systems at under two parts per million of carbon monoxide and under one part per million of nitrogen oxides, with negligible sulfur oxides.¹

The family divides into two broad branches, and the branch matters more than most buyers realize. The low-temperature classes, built on polymer-membrane and phosphoric-acid chemistries, start relatively quickly, tolerate load changes better, and return their recovered heat as hot water. The high-temperature classes, built on molten-carbonate and solid-oxide chemistries, reform pipeline gas internally, occupy the top of the electrical-efficiency range, and exhaust heat hot enough to matter thermally, around 700 degrees Fahrenheit in the catalog's representative systems. The price of those virtues is operational: long start-ups and a strong preference for running flat and continuously.¹

One honesty item belongs in the first section rather than the footnotes. On pipeline natural gas, which is what nearly every fielded system in California runs on today, a fuel cell is still a gas appliance. The reforming step that turns methane into hydrogen-rich gas emits carbon dioxide in proportion to fuel consumed, and the federal catalog says so plainly.¹ The non-combustion claim is a claim about criteria pollutants and the permit code. It is not a carbon claim, and a buyer who hears it presented as one is listening to marketing.

Section 02The case for: a permit that never gets filed

The strongest argument for fuel cells in California is not on the spec sheet. It is in the rulebooks of the state's most demanding air districts. The South Coast district's Rule 219, as amended June 5, 2026, lists fuel cells that produce electricity in an electrochemical reaction, naming the phosphoric-acid, molten-carbonate, polymer-membrane, and solid-oxide chemistries, among equipment not requiring a written permit.² The Bay Area district's permitting regulation carries a parallel exemption, in nearly the same words, at Section 2-1-128.22.³ The posture is statewide by design: a 2000 statute directed the state air board to adopt a certification program, with uniform emission standards, for generation technologies that are exempt from district permitting, and stationary fuel cells are the archetypal equipment class under it.⁴

To feel the weight of that, look at the workstream on the combustion side of the aisle. A new engine or turbine in a strict district is a permitted source: control-technology determinations, possible emission-offset obligations, public process, source testing, and permit conditions that follow the machine for its operating life. In a congested district that chapter is commonly the critical path of the whole project, and its duration is not fully within the owner's control. Exempt equipment converts the air-permitting chapter into a checklist item. In the time-to-power era, when the deadline rather than the price is often the binding constraint, that is not a soft benefit. At some sites it is the decisive one.

Three caveats keep the point honest. First, exemption from a written air permit is not exemption from law: registration programs can still apply, building and fire approvals still run, interconnection for parallel operation is its own workstream, and environmental review attaches to projects, not just machines. Second, the advantage is jurisdictional; in a district where a combustion permit is quick and lightly conditioned, the differential shrinks toward zero. Third, rules move. The June 2026 amendment date on the South Coast rule is its own reminder that the only text that counts is the current text, pulled the week of the decision.

Section 03The case for, continued: efficiency, silence, and heat

The second pillar is electrical efficiency. Because the machine converts chemical energy to electricity directly, without a heat engine in the middle, its ceiling is not set by combustion thermodynamics. The U.S. Department of Energy characterizes fuel cells as able to convert fuel directly to electricity at efficiencies capable of exceeding 60 percent.⁵ Fielded natural-gas systems in the EPA catalog run from the mid-thirties to the mid-fifties of percent on a higher-heating-value basis, with the high-temperature classes at the top of the band.¹ Combustion machines at distributed scale reach numbers like those only by adding bottoming cycles. The reason this matters commercially is arithmetic, not engineering: fuel is the largest lifetime cost line of any gas-fired machine, and electrical efficiency is a permanent multiplier on it, every operating hour, for decades.

Two practitioner notes keep the efficiency pillar honest. Vendor literature usually quotes efficiency on the fuel's lower heating value, which reads several points higher than the higher-heating-value basis on which gas is actually billed; comparing a brochure number against a utility tariff without converting the basis is one of the most common modeling errors in this field. And efficiency is not a static property. Stacks degrade in service, so the number that belongs in a model is the expected average across the stack's life at the site's real load shape, not the day-one figure at rated output.

The third pillar is silence. The EPA catalog characterizes representative systems as quiet enough for conversational level at close range, roughly 60 dBA at thirty feet, and acceptable for indoor installation; the Department of Energy notes the mechanism, few moving parts.^1,5 For hospitals, laboratories, campuses, and dense urban buildings, where property-line noise ordinances and neighbor relations decide what can actually be built, that is a siting characteristic no combustion machine of comparable scale offers without a serious acoustic-enclosure budget. The near-zero criteria emissions compound the same advantage: air intakes, operable windows, and sensitive neighbors stop being fatal constraints.

The fourth pillar is heat. The high-temperature classes exhaust at temperatures useful for genuine heat recovery, and representative systems in the catalog post total combined-heat-and-power efficiencies in the seventies and eighties of percent on a higher-heating-value basis.¹ There is an honest trade inside this trade: the better the machine is at making electricity, the less heat it has left to give, so the power-to-heat ratio rises exactly as the electrical efficiency does. A site with a heavy, high-grade thermal load can rationally prefer a combustion cogeneration unit that makes more and hotter heat at lower electrical efficiency. The machine should be matched to the site's measured power-to-heat ratio, not to the largest number on any one line of a spec sheet.

Section 04The case against: capital, fuel, the stack, and the maker

Now the other ledger, with the same discipline. The first entry is capital. The same federal catalog that documents the efficiency band prices its representative commercial-and-industrial systems between 4,600 and 10,000 dollars per kilowatt installed, in 2014 dollars, and is blunt about the consequence: high installed cost is the reason fuel-cell economics have concentrated in premium applications.¹ Equipment prices have moved since those estimates were compiled, and only current quotes belong in a real decision, but the structural causes of the premium have not disappeared: specialized stack materials, production volumes far below the engine industry's, and a full fuel-processing and power-electronics train around the stack. A buyer should walk in expecting the fuel cell to be the most capital-intensive machine on the shortlist, and should demand that its permitting, efficiency, and siting advantages visibly pay for that premium at the specific site. Sometimes they do. Often they do not.

The second entry is the fuel line. A fuel cell buys gas every hour it runs, and in California that exposure is unusually expensive. The state's industrial gas price averaged 11.48 dollars per thousand cubic feet in 2024 against a national average of 4.07, and its commercial price 14.01 against 10.07, per the U.S. Energy Information Administration.⁶ The same state that grants the permitting advantage charges nearly triple the national industrial price for the fuel. As an illustrative frame only: at a mid-band electrical efficiency of 45 percent on a higher-heating-value basis, the heat rate is roughly 7,600 Btu per kilowatt-hour, so each dollar per million Btu of delivered gas adds about three-quarters of a cent to each kilowatt-hour produced; at delivered prices near California's 2024 industrial average, the fuel line alone runs above eight cents per kilowatt-hour, before capital, maintenance, or the stack accrual. High efficiency softens this exposure relative to every other gas-burning machine. It does nothing to soften it relative to the tariff, to solar, or to storage, and a study that hides that distinction is advocating, not analyzing.

The third entry deserves its correct name. The stack, the electrochemical core of the machine, wears out in service. The catalog's characterization is direct: recommended stack rebuilds every five to ten years, and expensive.¹ Its maintenance figures make the same point in dollars: contracted maintenance excluding the stack-replacement sinking fund is estimated at 0.7 to 2.0 cents per kilowatt-hour, while all-in operating and maintenance costs for the representative commercial systems run 3.6 to 4.5 cents in 2014 dollars; most of the spread between those two lines is the stack.¹

The stack rebuild is not a contingency. It is an appointment on the calendar, and the only question is whose balance sheet has priced it.

Treating the rebuild as a risk to be discounted is the buyer's error; treating it as a line item the seller will address later is the seller's. The disciplined posture is contractual: a full-term service agreement that carries the stack at a stated fee with stated escalators, written performance commitments across stack life, and clarity about output and efficiency in the years just before a rebuild. An owner who cannot get those terms in writing has learned something important about the offer.

The fourth entry is concentration. Each fuel-cell class is manufactured by a small number of firms worldwide, the stacks are proprietary, and there is no third-party rebuild ecosystem remotely comparable to the parts and service channels behind reciprocating engines. Buying the machine is closer to entering a fifteen-year service relationship than to purchasing equipment: stack supply, remote monitoring, and service pricing at renewal all live with the maker. The diligence is unglamorous and non-optional. Read the supplier's financial condition from public filings where they exist. Establish the depth of its service organization in the region. Ask, in the contract, what happens if the maker stumbles while the stack calendar keeps running, because the calendar does not pause for anyone's restructuring.

Three smaller entries complete the column. The machine is single-fuel in practice, so a gas-supply interruption is an outage unless the site pairs storage or a second source. The high-temperature classes cycle poorly, with long start-ups noted in the catalog, which makes them baseload machines regardless of the duty cycle in the pitch deck.¹ And on pipeline gas the carbon dioxide continues, which matters twice: once in the owner's own accounting, and once as policy risk attached to any gas-consuming asset in California across a multi-decade life.

Section 05The ledger, side by side

Set out as a single table, the two cases look like this. No row is decisive alone; the site decides which rows are load-bearing.

Section 06Where fuel cells genuinely win

Four site archetypes recur, and in each the winning mechanism is specific enough to name.

• The permit-bound site. A strict district, a real deadline, and a combustion path facing control-technology review and offset exposure. Here the exemption converts the schedule's largest uncertainty into a non-event, and schedule certainty is worth real money when the alternative is a year or more of process risk. This is the cleanest fuel-cell win in California, and it is entirely jurisdictional: move the same project to a lenient district and the advantage largely evaporates.
• The neighbor-bound site. Hospitals, laboratories, campuses, and dense urban buildings, where noise ordinances, air intakes, and community relations govern what can be built at all. A machine at conversational sound levels with near-zero criteria emissions can be sited where a combustion hall cannot, and that siting freedom is sometimes the entire project.
• The flat, continuous load with hot-water-grade thermal demand. Continuous operation amortizes heavy capital and collects the efficiency edge every hour of the year, and a matched thermal load absorbs the heat the power block gives up. The machine's preferred duty cycle and the buyer's best economics coincide, which is rarer than it sounds.
• The expensive-gas site choosing among gas machines. Where delivered gas is dear, as it is in California, the highest-efficiency gas machine loses the least per therm. This is a relative win against other gas-burning equipment only. Dear gas argues just as loudly for paths that burn nothing, and an honest study says both things in the same breath.

Note what the list requires: the win case is conjunctive. District, duty cycle, thermal fit, and a contractually carried stack, at the same time. When those align, the machine wins on the merits, without leaning on a single unqualified incentive assumption.

Section 07Where they clearly lose

• Standby, peaking, or strongly cyclic duty. Capital-heavy machines earn by running, so an idle fuel cell is expensive insurance, and the high-temperature classes tolerate cycling poorly besides. Engines own this duty, and honesty concedes it without a fight.
• Steam-heavy sites. Where the thermal load is large and high-grade, combustion cogeneration delivers more and hotter heat per dollar of capital. Forcing the highest-electrical-efficiency machine into a heat-dominated site mismatches the one ratio that matters.
• Short holding periods. An owner whose horizon is shorter than the first stack rebuild is buying an event that someone must price. If the service agreement does not carry it, the exit negotiation will, at a discount set by the buyer's advisor rather than the owner's.
• Sites the grid can serve on time, or gaps storage can bridge. Documented, timely utility service beats the entire self-generation family on operating burden, and short-duration constraints are storage problems, not generation problems. The fuel cell loses these cases the same way every other machine does, and a neutral study prices the losing rows anyway.
• Zero-fossil mandates without a verified renewable-fuel path. On pipeline gas the carbon continues. Where policy or a corporate commitment requires zero on-site fossil fuel, the machine stays on the list only with a real, dated, priced renewable-gas supply, or a fuel-switch path supported by evidence rather than a future-fuels slide. Readiness claims about tomorrow's fuel are not today's capability, from any equipment class.

Section 08Pricing the decision

For an owner whose shortlist includes a fuel cell, six questions do most of the work, roughly in the order they kill deals:

• Rule text, current week. Pull the district's exemption and registration language the week of the decision, not from memory and not from a vendor summary. The rules carry amendment dates for a reason.
• Efficiency at your load. Higher-heating-value basis, part-load behavior, and the degradation curve across stack life, applied to measured interval data. The brochure point at rated output is the least informative number in the file.
• The stack, inside the contract. A full-term service agreement with the rebuild priced, escalators stated, and written performance commitments through the years before and after the rebuild.
• Gas sensitivity, stated plainly. Model the delivered-price band, and name the gas price at which each alternative path overtakes the machine. If the analysis cannot name that number, it is not finished.
• The vendor file. Financial condition from public filings where available, regional service depth, and the stack supply chain, examined rather than assumed in either direction.
• Exit alignment. The holding period against the stack calendar, and the assignability of the service agreement to a successor owner.

On incentives, one discipline: under current federal law as of August 2026, the investment tax credit for qualifying property is 30 percent under 26 U.S.C. §48 and §48E; statutory adders exist but each must be individually qualified before it is underwritten, and qualified tax counsel belongs in the transaction before any credit is booked.⁷ California's Self-Generation Incentive Program, historically material to fuel-cell purchases, has a generation budget closed to new applications; it belongs in models as history, not as revenue.

The honest verdict is that the fuel cell is a specialist's machine, and the specialty is real. In a strict district, on a continuous load, next to sensitive neighbors, with the stack carried contractually and the fuel line priced across its band, it wins on the merits against every combustion alternative, and sometimes against waiting for the wire. Deployed as a default answer everywhere else, it loses: usually to a cheaper machine, sometimes to the grid, occasionally to doing nothing at all. Which side of that ledger a specific site occupies is not a matter of conviction. It is a finding, and the only analysis worth paying for is the one that was equally prepared to reach either answer.