Microgrids and Single Assets:
Resilience, Honestly Priced

Resilience is the easiest line item in an energy proposal to oversell, because the buyer is usually asked to price fear instead of hours. The honest method is shorter than the sales deck: measure the site's actual outage record, price what an hour of darkness costs this specific operation, and buy exactly as much ride-through as that arithmetic defends.

Section 01Two different products share one word

A standby generator behind an automatic transfer switch and a campus that can island from the grid are routinely sold under the same word. They are not the same product. They do not cost the same, they do not buy the same thing during an outage, and they do not impose the same obligations for the twenty years after commissioning. Most resilience mistakes begin by conflating them.

The single standby asset is the mature product. It waits. When utility service fails, it starts; a transfer switch moves a bounded set of critical circuits onto it, and the site accepts a gap measured in seconds to minutes. When the feeder returns, the switch transfers back and the machine returns to sleep. Every fire marshal, insurer, and service shop in the state understands it. Its failure modes are known, its testing rituals are codified, and its price is set by a competitive market that has built it for decades.

A microgrid is a different commitment. The governing standard for microgrid controllers, IEEE 2030.7, describes the product precisely: a bounded electrical system that operates in two steady states, grid-connected and islanded, under a control system that manages dispatch in both and manages four transitions between them: planned islanding, unplanned islanding, reconnection, and black start.¹ That taxonomy rewards slow reading, because the engineering difficulty, and most of the money, lives in the transitions. Holding an island steady is work. Entering one without dropping the loads that matter, and rejoining the utility without a fault, is where the controls budget, the protection studies, and the switchgear line go.

The commercial distinction follows from the technical one: a standby asset buys recovery; a microgrid buys continuity. Recovery means the critical panel comes back inside a minute and the site limps with dignity. Continuity means the process never notices. Between those two words sits, at most sites, the single largest cost step in the entire resilience conversation, and the correct choice between them is a property of the site's record and economics, not of anyone's catalog.

Section 02The measured record, not the imagined one

Start with what interruptions actually look like. In 2024, the average electricity customer in the United States lost service for about eleven hours, the most in a decade, and roughly eighty percent of that time traced to major events, hurricanes above all.² Strip out the major events and the routine average has held near two hours a year for the past decade; major-event time itself averaged closer to four hours a year across 2014 through 2023.²

Two lessons sit in those numbers. First, the routine grid is, on average, fairly reliable: two hours a year does not justify much capital on its own. Second, the average is a national blend of calm years and catastrophe, and the difference between them is the tail. Resilience spending is almost never about the routine hours. It is about whether the site can absorb the rare long event, and averages are the wrong instrument for tails.

Which is why the only record that matters is the site's own, assembled in three layers. The utility's reliability reporting for the surrounding system comes first, filed annually with state regulators, supplemented wherever possible with circuit-level history from the utility itself. The site's own operating log comes second: interval-meter gaps, control-system alarms, and incident reports preserve every event the institution has already survived, usually with timestamps the utility's indices smooth away. Third come the forward exposures no history can show: location in a wildfire de-energization zone, where planned shutoffs during high-risk weather can extend across days; service from a single radial feeder with no alternate path; a substation in a floodplain.

Then price the hour, because hours are only half the arithmetic. For a manufacturer, a dark hour costs contribution margin plus scrap plus restart, and restart is routinely the largest term: a four-hour interruption that forces an eight-hour requalification is a twelve-hour event. For cold storage, the honest number is often small for the first several hours, because thermal mass rides through, and then grows steeply. For a data floor, the cost may be contractual and reputational at once. And for some facility classes, healthcare above all, emergency power is a code obligation that exists regardless of economics; this paper concerns the discretionary increment above obligation, not the obligation itself.

Both errors are common and both are expensive. Sites in mild territory buy islanding campuses against a record showing ninety routine minutes a year. Sites on radial feeders in wind corridors, whose margin dies in hour two, buy nothing and call it thrift. The cure for both is the same sentence: price resilience against the measured record and the priced hour, and write both numbers down before any equipment is named.

Section 03The bill of materials

What resilience money actually buys, in descending order of visibility:

• Generation or storage. The machinery itself, sized to the protected load. This is the only line most proposals price carefully.
• Islanding capability. The engineering that lets a site separate cleanly and hold itself. Protective relays coordinated for utility fault current can sit silent in an island, where available fault current may be a small fraction of the utility-fed value, particularly on inverter-based sources. Grounding arrangements that were correct grid-connected can become incorrect islanded. Synchronizing equipment must prove the island can rejoin without a bump. This work is invisible in a rendering and unforgiving at commissioning.
• Controls. The microgrid controller and its integration. In the National Renewable Energy Laboratory's multi-project cost study, reported controller costs ranged from $6,200 to $470,000 per megawatt of generation, with a mean near $155,000.³ A range spanning nearly two orders of magnitude is not a market being coy; it is a signal that the word controller describes scopes of wildly different ambition, and that the buyer must specify functions, not a noun.
• Switchgear and infrastructure. Paralleling gear where multiple sets run as one plant, and, at campus scale, the medium-voltage switchgear and supervisory systems that let several buildings island together. The same study found the heaviest additional-infrastructure costs in exactly the segments that island at medium voltage.³
• Fuel systems. Tanks, treatment, pipe, and contracts. Section 04 is devoted to them.
• Load management. The shed schedule and the wiring that enforces it. The cheapest kilowatt in any island is the one deliberately not served; subtraction, engineered in advance, stretches every other line on this list.
• The testing calendar. A permanent operating cost, regulated in both directions, taken up in Section 05.

For scale, the same national study, built on a project database assembled through roughly 2016, found mean installed costs near $2.1 million per megawatt for community microgrids, $2.5 million for utility projects, $3.3 million for campuses, and $4.1 million per megawatt for commercial and industrial sites. Generation itself accounted for about 76 percent of cost in the campus segment, while in commercial projects soft costs alone ran to roughly 43 percent of the total.³ Treat those figures as anatomy rather than quotation: they are dated, the sample is heterogeneous, and current pricing belongs to current bids. The durable lesson is the structure. In simple networked-generator projects the money tilts toward controls and retrofit integration; in complex campuses, toward the machinery; in small commercial projects, toward the soft costs. The premium over a plain standby asset is not one number. It is a stack of separable decisions, each of which can be taken or declined.

Section 04Fuel is the honest constraint

Islanding switchgear is worthless on day three beside a dry tank. Duration, stated in hours at a stated load, is the only honest unit of resilience, and duration is a fuel property before it is an equipment property. Each fuel posture deserves its case for and its case against, stated in the same breath.

Stored diesel is the incumbent for a reason: the energy sits on site under the owner's control, at typically the lowest capital cost per kilowatt among these options, with a service ecosystem that reaches every county. The honest limits: the tank is finite and must be sized in days, not adjectives; stored fuel degrades and needs a sampling and polishing program; and the refueling contract that looks firm in a calm month is a shared promise during a regional event, when every counterparty holds the same clause. Priority language is worth reading before the storm. Start failure, meanwhile, clusters at exactly the moment of demand, which is why the testing regime of Section 05 exists at all.

Pipeline gas machines, whether reciprocating engines, turbines, microturbines, fuel cells, or linear generators, replace the tank with a pipe. The case for: fuel arrives continuously, storage never expires, and gas delivery is a separately pressurized system that has historically kept operating through most electric interruptions. The case against: the pipe is a dependency the owner does not control, it is not invulnerable to earthquake or coordinated regional disruption, and a gas machine permitted strictly for emergency service inherits the same annual-hour arithmetic as diesel while giving up diesel's stored-energy autonomy. Gas machines are strongest in the opposite posture: running daily as the site's primary power with islanding designed in, so that resilience arrives as a byproduct of an asset that already justifies itself.

Batteries are the only entry that makes the transition disappear: ride-through is effectively instantaneous, there is no combustion permit, no stack, no fuel logistics, and siting is quiet. The honest limit is arithmetic. Duration is bought in hours and paid in capital, recharge requires a surviving source, and carrying a multi-day tail on storage alone is generally the most expensive way to do it. Storage is superb in the first minutes of an event and merciless about the last days.

Solar paired with storage makes the recharge indigenous, and for shallow loads in fair weather the pairing can run for a long time. It must also be priced against its weather correlation: the storms that cause the longest outages often suppress production at the same time, and winter output is materially lower than summer's. A strong layer; a risky sole provider.

Two class-specific notes complete the honesty. Fuel cells are electrochemical rather than combustion, quiet, and in strict air districts materially easier to permit; they are also capital-intensive and generally suited to continuous duty rather than cold standby cycling, which makes them a poor pure-standby purchase and a credible everyday-power foundation for a site that wants islanding on top. Combustion machines carry the mirrored trade: mature, dispatchable, and inexpensive per kilowatt, but in California's stricter districts new combustion capacity faces demanding control-technology requirements, and any machine that runs daily needs its own maintenance outages covered inside the island plan.

The practitioner's answer to fuel security is rarely one fuel. It is diversity: a pipe plus a tank, or a tank plus stored electrons, sized so the island survives the failure of any single source. Price diversity deliberately, as a design decision, rather than discovering it later as change orders.

Section 05The testing calendar nobody prices

Readiness is not a property of equipment. It is manufactured, monthly, by people, and the manufacture is regulated from both directions. On the minimum side, NFPA 110, the standard broadly adopted for emergency and standby power systems, requires the machine to be exercised under load at least monthly for at least thirty continuous minutes, with loading criteria attached, and prescribes supplemental load-bank testing where building load cannot exercise the machine properly, because a lightly loaded diesel accumulates unburned fuel and soot until the day it is asked to perform.⁴

An untested standby asset is not resilience. It is a hypothesis with a fuel tank.

On the maximum side, air regulation caps the very same activity. Federal hazardous-air-pollutant rules for stationary engines allow an emergency engine at most one hundred hours a year for maintenance checks and readiness testing, with non-emergency operation counting inside the cap.⁵ California goes further: under the state's airborne toxic control measure, a new emergency standby diesel engine above fifty brake horsepower is limited to fifty hours a year of maintenance and testing, with district discretion to allow up to one hundred only at very low particulate emission rates.⁶ Emergency operation itself is not the constrained quantity. Practice is.

Run the arithmetic on the corridor this creates. Twelve monthly half-hour exercises consume six engine-hours; an annual load-bank protocol adds roughly another hour and a half; call it seven and a half hours of mandatory practice against a fifty-hour California allowance, before a single diagnostic run, repair verification, or training hour is logged. The corridor is workable, but it is a corridor, and it carries a commercial consequence that outweighs the hours themselves: an emergency-classified machine is barred from routine economic operation. Its capital works a handful of hours a year and earns nothing the rest. Permitting a machine for non-emergency service instead buys the right to run daily, at the price of stricter emission controls and a longer permitting path. That single classification choice moves more money over an asset's life than most equipment choices do.

Microgrids multiply the calendar rather than escaping it. A credible islanding site schedules planned island exercises, at full transition, with the loads that matter connected, at a time chosen deliberately rather than delivered by weather. It patches and revalidates controller software. It re-studies protection coordination when loads or sources change. It trains successors, because a microgrid that lives in one retiring engineer's head is a single point of failure with excellent documentation of everything else. Every one of those lines is a recurring cost. A resilience price that omits them is not a price; it is a down payment.

Section 06When the single asset wins, and when it does not

The single standby asset, duplicated for redundancy where the load justifies it, is the correct purchase more often than the industry selling alternatives would suggest. It wins when the measured record shows short and rare interruptions; when the loads that truly cannot fail fit a bounded panel: life safety, controls, refrigeration, communications; when a gap of seconds is survivable, as it is for far more processes than their owners first claim; when nobody intends the asset to run on ordinary days; and when the organization is honest about its appetite for operating complexity. Its virtues are underrated precisely because they are boring. Every authority having jurisdiction understands it, every insurer can rate it, every service shop can fix it, and its testing ritual fits in a morning.

Between the extremes sits networked standby: several sets, often including retrofitted existing units, paralleled to behave as one plant. It buys redundancy and staged load pickup at a modest premium, with the cost anatomy tilted toward integration rather than iron, and for many campuses it is the quietly correct answer.³

The full microgrid earns its premium when specific conditions hold, and the discipline is to name which one is doing the work. Continuity rather than recovery: the process genuinely cannot accept the gap. Tail exposure: the record and the geography argue for multi-day events whose hourly cost is severe, where duration and orchestration decide outcomes. Existing assets worth conducting: the site already hosts, or independently justifies, continuous generation and storage, so islanding is an increment on an economic decision rather than the entire reason. Scale: multiple buildings need to island together at medium voltage, with eyes open that this is exactly where the infrastructure money concentrates.³ Flexibility: engineered load shedding real enough to stretch fuel and shrink machinery.

The anti-case deserves equal print. A campus microgrid whose fuel plan dies at twenty-four hours is a standby system with better marketing; the brochure says islanding and the tank says Tuesday. State the duration at stated load or the resilience claim is not a claim. The mirrored failure is under-buying by sites whose two measured facts, a radial feed and a margin that dies in hour two, already justify the spend. Both errors trace to the same root: nobody priced the hour and nobody pulled the record.

Section 07The arithmetic, stated plainly

The frame fits on one page, and every number in it should be labeled illustrative until a specific site fills it in. Annualize the resilience increment: capital recovery, maintenance, fuel, controls upkeep, and the testing calendar of the proposed design, minus the same figures for the baseline the site would buy anyway. Set against it the exposure: routine measured hours multiplied by the priced hour, plus the tail, long events weighted by honest probabilities and valued at their true and often nonlinear cost, plus the obligations that exist regardless: code, contract, custody of anything that cannot be allowed to spoil or stop.

Then read the comparison the way an underwriter would. If routine hours alone carry the increment, buy it without ceremony. If only the tail carries it, the purchase is insurance, and it should be forced to compete with the other ways institutions buy survival: process hardening, larger tanks on a simpler asset, contractual remedies, a second site. Insurance that never faced its alternatives was not purchased. It was absorbed.

Six questions settle most of these decisions. They fit on one page and require no vendor's cooperation.

1. State the protected load and the duration.Kilowatts served and hours sustained, in writing. A resilience proposal that cannot fill in both blanks is a brochure.
2. Pull the measured record.Utility reliability history for the serving circuit plus the site's own log, several years deep, alongside the forward exposures the record cannot show.
3. Price the hour three times.A routine hour, a full day, a full week: margin, spoilage, restart, penalties, obligations. The three answers are usually three different arguments.
4. Name the fuel plan by day.What runs, what sheds, what refuels, under whose signed obligation, through day one, day three, day seven. Adjectives are not a fuel plan.
5. Cost the calendar.Testing hours, fuel burned practicing, staff time, permitted-hour consumption, island exercises, controller upkeep, every year of the asset's life.
6. Price the simpler rival.An honest microgrid proposal contains the standby asset it must beat, priced with the same care. If that page is missing, someone chose the answer before the analysis.

Resilience, honestly priced, is a respectable purchase, and some sites should buy far more of it than they have. The dishonest version is the one bought against imagined weeks of darkness, or refused against a record that already shows the loss. Our part in this is narrow and deliberate: an independent study, on the owner's side, that prices the credible paths against the record and the hour, with sourced numbers and no equipment margin waiting behind the recommendation. The site does not need the impressive answer. It needs the true one.