Revenue per Megawatt-Hour in Energy: Wholesale Power Market Performance in 2027

A merchant plant that looked profitable and wasn't

Picture a 600 MW combined-cycle gas plant in a deregulated market, roughly 45% annual capacity factor, delivering something on the order of 2.3 million MWh a year. The finance team reports a blended realized figure in the mid-$50s per MWh and calls the year fine. The trading desk looks at the same portfolio and sees three separate businesses that happen to share a turbine hall, and one of them is quietly losing money.

The first business is energy. The plant clears in the day-ahead market against a nodal locational marginal price, and its economics are entirely spark-spread driven: the price of power at its node minus the cost of the gas it burns, adjusted for heat rate. A plant with a 7,000 Btu/kWh heat rate burning gas at $3.50/MMBtu has variable fuel cost near $24.50/MWh before variable operations and maintenance. Whether the day cleared at $38 or $61 determines whether the unit ran at all, and the annual average hides the fact that most gross margin arrives in a few hundred hours.

The second business is capacity — availability payments for being there when the system needs the plant, whether or not it runs. This income is auction-cleared, lumpy across years, and set well before the delivery period. It arrives whether the spark spread is $2 or $40, and when spread into a per-MWh figure it can be enormous in a low-run-hours year and modest in a high-run-hours year. That inverse relationship is where naïve reporting breaks: a plant that ran less can show a *higher* capacity contribution per MWh while the overall business got worse.

The third business is ancillary services — regulation, spinning and non-spinning reserves, and in some markets reactive power or ramping products. A fast-start or fast-ramping unit can be economically dispatched to provide these even in hours where straight energy sales don't cover fuel. This is the line item most commonly under-optimized, because it requires bid submissions in products that plant operators don't think of as their day job.

The failure in the scenario is that nobody separates the three. The blended metric is stable, so no alarm fires. Underneath, energy margin has compressed with a warm shoulder season, capacity revenue is fixed and simply being divided by fewer MWh, and the ancillary desk has never bid regulation because the unit's automatic generation control certification lapsed. The blended number is a lagging average of three independently moving parts, and averaging destroyed the signal that would have prompted action.

The corrective move is structural, not analytical heroics: decompose the metric into its revenue streams before you report it, and hold each stream against a stream-specific benchmark. Energy revenue per MWh goes against the relevant hub or nodal index. Capacity revenue is reported both as dollars per kilowatt-year (the form the auction actually clears in) and as its per-MWh dilution, with generation volume shown beside it. Ancillary revenue goes against the theoretical maximum the unit could have captured given its ramp rate and certification status. Only then does the blended figure mean anything.

How the metric is actually assembled

Wholesale power revenue per MWh is a quotient, and both the numerator and the denominator have structure that most reporting flattens.

The denominator is metered energy delivered at the point of interconnection over the period. It is not nameplate capacity times hours. For a thermal unit it reflects forced outages, planned maintenance, economic non-dispatch, and derates from ambient temperature. For wind and solar it reflects the resource itself plus curtailment — both economic curtailment (the operator chose not to produce because the price was negative) and network curtailment (the grid operator instructed a reduction because of a transmission constraint). Two identical wind farms with identical resource can post materially different per-MWh figures purely because one sits behind a constrained interface.

The numerator stacks at least four components:

*Energy revenue* is settled at locational marginal price. LMP decomposes into a system energy component, a congestion component, and a marginal loss component. The system energy component is common to the whole market; congestion and losses are what make one node different from another. Two generators on the same day, in the same market, can settle tens of dollars apart per MWh entirely through the congestion term. Most generators settle across two sequential markets: a day-ahead financially binding schedule and a real-time balancing settlement on the deviation between the day-ahead position and actual output. Revenue is therefore day-ahead schedule times day-ahead LMP, plus (actual minus scheduled) times real-time LMP.

*Capacity revenue* is a payment for committed availability over a delivery period, cleared in a forward auction in markets that run one. It is priced per unit of capacity per unit of time and is typically subject to performance obligations: if the unit is unavailable during a declared scarcity event, it can face a non-performance charge that claws back part of the payment. Its per-MWh translation is capacity dollars divided by whatever the unit actually generated.

*Ancillary services revenue* comes from being paid to hold headroom or respond to control signals. Different markets structure these differently, but the economics rhyme: you're paid for a capability reservation, sometimes plus a mileage or performance payment for the actual movement. Batteries and fast hydro dominate the fastest products because response speed is the product.

*Attribute and contract revenue* covers renewable energy certificates, any applicable tax-credit monetization, and the settlement of bilateral hedges — a power purchase agreement, a fixed-for-floating swap, or a heat-rate call option. A hedged generator's realized figure is a blend of market settlement and contract settlement, and the contract can dominate.

The single most useful derived metric from this stack is capture rate: realized energy price divided by the simple time-weighted average index price over the same period, expressed as a percentage. It isolates the timing-and-location question from the level question. A market where average prices fell 20% will drag everyone's revenue per MWh down together; capture rate tells you whether you fell more or less than the market did, which is the only part you actually control.

Capture rate behaves very differently by technology, and it is a structural property, not a management scorecard, until you compare like with like. Solar produces concentrated in midday hours; in high-solar-penetration regions those are precisely the lowest-priced hours, so solar's capture rate sits structurally below 100% and declines as more solar is added — the well-documented cannibalization effect. Wind faces the same dynamic in high-wind regions, with the added feature that heavy wind hours can push nodal prices negative, at which point continued production destroys value unless a production-based subsidy makes negative-price generation still worth it. Dispatchable thermal, by contrast, can exceed 100% capture because it chooses its hours: it runs when prices are high and sits out when they aren't, so its realized average beats the flat time-weighted average by construction.

That last point deserves emphasis, because it inverts the intuition. For a peaker, a *low* capacity factor with a *high* capture rate is the intended outcome. Judging a peaking unit by revenue per MWh alone and concluding it should run more is one of the most expensive misreadings of this metric.

Real numbers, ranges, and the shape of the distribution

Concrete figures matter here, but wholesale power is one of the most volatile markets in the economy, so treat every number below as an order-of-magnitude anchor rather than a forecast — actual values differ by market, year, fuel price, and weather, and 2027 outcomes depend on gas prices and load growth that nobody can pin down in advance.

The distribution is the story. Wholesale energy prices are not normally distributed; they have a long right tail. In a typical year a large share of a merchant generator's annual gross energy margin arrives in a small fraction of hours — often single-digit percentages of the year producing a majority of the margin. In scarcity conditions, prices can move from tens of dollars per MWh to hundreds or, in markets with high offer caps during genuine shortage, to thousands. The practical consequence: annual average revenue per MWh is a summary statistic of a distribution whose tail dominates. Managing to the average is managing to the wrong object.

Negative prices are routine, not exotic. In regions with high renewable penetration and limited export capability, oversupply during low-load, high-resource periods drives LMP below zero. A generator with a per-MWh production subsidy may rationally keep producing down to the negative of that subsidy before curtailing. A generator without one should curtail as soon as price plus any avoided variable cost goes negative. The frequency of negative-price hours has been rising in high-renewable zones, and it is the single largest structural drag on wind and solar capture rates.

The spark spread governs thermal dispatch, not the power price. Implied heat rate — power price divided by gas price — is the number a gas trader actually watches. A unit with a 7,000 Btu/kWh heat rate is in the money whenever the implied market heat rate exceeds roughly 7.0 (plus variable O&M in heat-rate terms). This is why a year of high power prices can be a *bad* year: if gas rose faster than power, spreads compressed even as revenue per MWh rose. Reporting revenue per MWh without reporting gross margin per MWh alongside it will produce exactly this false positive.

Capacity revenue translation. A capacity payment quoted per kilowatt-year converts to a per-MWh figure by dividing by (capacity factor × 8,760). At a 50% capacity factor, one dollar per kilowatt-year becomes roughly $0.23/MWh; at a 10% capacity factor, it becomes roughly $1.14/MWh. The same auction result therefore produces a five-fold difference in per-MWh capacity contribution depending only on how much the unit ran. This is the arithmetic behind the earlier warning: never compare capacity-per-MWh across units with different run profiles without normalizing.

Ancillary services are capacity-constrained, not demand-constrained. The total requirement for regulation or reserves in any market is a small percentage of system load — these are thin markets. That has two consequences. First, the total revenue pool is finite, so a rush of new fast-responding resources into a regulation market drives clearing prices down quickly; early entrants earn far more per MW than late ones. Second, because the pool is thin, per-MWh ancillary revenue for a resource that provides them can be a surprisingly large share of total revenue for low-energy-throughput assets like batteries, precisely because the denominator is small.

A practical benchmarking protocol. Rather than chasing published industry averages that may not match your market, technology, or vintage, build internal benchmarks that are reproducible:

1. Index baseline. For each asset, compute the simple time-weighted average LMP at its settlement node for the period. This is the "do nothing but exist" price.
2. Shape baseline. Compute the generation-weighted average price using the asset's actual output profile but the index price — this is what a perfectly passive asset with your production shape would have earned.
3. Realized. Compute actual energy revenue per MWh.
4. Attribute the gap. Realized minus shape baseline is the value your bidding, unit commitment, and real-time decisions added or destroyed. Shape baseline minus index baseline is the value your technology's inherent production profile added or destroyed — a structural fact, not an operating result.

That four-line decomposition is more useful than any external benchmark, because it separates what you control from what you inherited. Run it monthly per asset and the conversation shifts from "our number is low" to "our shape cost us X and our execution added Y."

Set thresholds on the decomposition, not the blend. Reasonable alarm conditions: execution gap negative two months running; ancillary participation below the unit's certified capability for any product; forced outage rate during the top-100 priced hours materially exceeding the annual forced outage rate (this is the single most expensive availability failure mode, because unavailability correlates with exactly the hours that carry the margin).

Trade-offs: hedged certainty versus merchant upside

Every decision that raises expected revenue per MWh generally raises its variance too, and the right point on that curve depends on the balance sheet, not on the metric.

Merchant exposure versus long-term contract. A fully merchant asset captures the entire tail. In a scarcity year it earns extraordinary revenue; in a mild, oversupplied year it may not cover fixed costs. A fully contracted asset under a long-term power purchase agreement earns a known, usually lower, price and gives up the tail entirely. The contracted asset is financeable at lower cost of capital, which is often the actual reason for the contract — the PPA isn't primarily a revenue decision, it's a financing decision. The trade-off is real and asymmetric: cheaper debt is worth accepting a lower expected price, up to the point where the contract price no longer covers debt service plus a return.

A structural warning on shaped contracts. A PPA that obligates delivery of a fixed volume in each hour, written against an intermittent resource, converts a production-risk asset into a production-*and*-price-risk asset. If the resource underproduces in an hour when prices are high, the seller must buy replacement power at the high price and deliver it at the contract price. This "shape risk" or "volumetric risk" has produced losses far exceeding the value of the contract itself for sellers who priced it as if it were a simple as-generated deal. As-generated PPAs, where the buyer takes whatever is produced, carry no such exposure. The distinction is worth more than a few dollars per megawatt-hour of contract price.

Congestion: hedge or absorb. A generator at a node that consistently prices below the trading hub faces a persistent basis drag. The financial instrument for this is a transmission right that pays the congestion difference between two points. Buying them removes basis volatility but costs an auction premium, and — critically — they are financial instruments whose payout depends on the grid operator's congestion revenue being sufficient. In periods of unusual grid topology (major outages, new lines energized), rights can under-fund and pay out less than the congestion actually incurred. They reduce basis risk; they do not eliminate it. The alternative is to absorb the basis and price it into the asset's expected revenue, which is defensible for a node with stable historical basis and indefensible for one whose basis is driven by a single transmission element.

Ancillary participation versus energy availability. Holding capacity in reserve to sell a regulation or reserve product means not selling that capacity as energy. When energy prices spike, the reserved capacity is earning a reservation payment while the energy market is paying multiples more. Co-optimized markets handle this automatically by pricing reserves against the energy opportunity cost, but the resource owner still faces a decision about which products to offer into at all, and about the wear-and-tear cost of following a fast control signal. For a battery, cycling against a regulation signal consumes warranty cycles; the correct comparison is regulation revenue per MWh *net of amortized degradation* against energy arbitrage revenue per MWh.

Day-ahead certainty versus real-time optionality. Selling the full expected output day-ahead locks in a known price and removes real-time exposure. Holding back and settling in real time captures intraday scarcity but exposes you to the reverse. The systematic answer is not one or the other but a policy: schedule day-ahead to a confidence level of your forecast (for a thermal unit that is close to full output; for wind it might be a lower percentile), and let the deviation settle in real time. Over-scheduling an intermittent resource day-ahead means buying back at real-time prices in exactly the hours the resource underperformed — which, for wind, tends to be the calm hot hours when prices are highest.

Batteries invert the metric. For storage, revenue per MWh discharged is the natural framing, but it is gross of the cost of charging. A battery that buys at $18/MWh and sells at $74/MWh shows $74 of revenue per MWh discharged and roughly $56 of gross spread before round-trip efficiency losses — and at 85% round-trip efficiency it bought about 1.18 MWh for every MWh sold, so the true charging cost per MWh discharged is closer to $21. Revenue per MWh, used naively on storage, systematically flatters the asset. Report spread capture per MWh discharged net of charging and efficiency, or the metric misleads.

Pitfalls that reliably destroy the number

Averaging across hours, nodes, and products. Already covered, but it earns its place at the top of the list because it is nearly universal. Any single blended figure covering a year, a fleet, and every revenue stream will be stable enough to look healthy while any of its components collapses. The fix is decomposition before aggregation, always.

Confusing revenue with margin. Revenue per MWh rising while gas prices rise faster is margin compression presented as performance. Every thermal asset report should carry gross margin per MWh — revenue minus fuel and variable O&M — immediately beside revenue per MWh. If only one number survives to the executive summary, it should be the margin one.

Outages that correlate with high prices. Forced outages are not randomly distributed. Heat drives both peak prices and thermal derates; extreme cold drives both peak prices and fuel-supply and freeze-related failures. A unit with a respectable annual availability figure can still miss a disproportionate share of the year's margin if its unavailable hours cluster in the priced hours. Track availability weighted by price, not just by count of hours — the two numbers can differ dramatically and only one of them is about money.

Treating curtailment as a technical statistic. Curtailment shows up in operations reporting as a percentage of potential generation and rarely gets converted into dollars. It should be, and with the correct price: curtailed MWh valued at the price that prevailed during the curtailed hours. Economic curtailment during negative prices is value-*creating* and should never be reported as a loss. Network curtailment during high-priced hours is expensive and may justify a transmission investment or a siting decision. Reporting both as one percentage conflates a good decision with a bad outcome.

Ignoring performance obligations attached to capacity payments. In markets where capacity payments carry a performance requirement, non-performance during a declared event can trigger charges that materially exceed the period's capacity revenue. The exposure is asymmetric and back-loaded: you collect steadily and can lose a multiple of it in a single event. Any pro forma that books capacity revenue as certain and ignores the penalty tail is mispricing the asset.

Letting product certifications lapse. Participating in regulation or fast reserve products typically requires telemetry, control-system certification, and periodic performance testing. These lapse quietly. A unit that has technically been capable of earning ancillary revenue for two years but was not certified to bid has been leaving a real, recoverable stream on the table with no line item anywhere showing the absence. Audit certification status against unit capability annually; the gap is pure recoverable revenue.

Over-scheduling intermittent output day-ahead. Covered above, but the failure mode deserves naming: a wind operator schedules a forecast-mean volume day-ahead, the forecast busts low on a hot still afternoon, and the operator buys back at a real-time price several times the day-ahead price. The loss on those few hours can exceed the entire year's incremental gain from day-ahead scheduling. Schedule to a conservative percentile for resources with meaningful forecast error.

Building the metric on unreconciled settlement data. Grid operator settlements are resettled — initial statements are revised, sometimes months later. A revenue-per-MWh dashboard built on initial settlement data will disagree with the general ledger, and the first time finance notices, the operational metric loses credibility permanently. Build the operational view on initial data for speed if you must, but reconcile to final settlement monthly and show the variance. A metric that finance and operations compute differently is a metric nobody trusts.

Assuming stationarity. Historical capture rates are a poor guide to future ones in a market whose generation mix is changing quickly. Every additional gigawatt of solar in a region lowers midday prices and therefore lowers solar capture; every retirement of a dispatchable unit raises evening scarcity and therefore raises the value of dispatchability. A 2027 pro forma built on backward-looking capture rates will overstate intermittent revenue and understate flexible revenue. Model the shape forward, or at minimum sensitivity-test it against a materially different mix.

Related questions

Why does revenue per MWh differ so much between two identical plants?

Location and dispatch. Different settlement nodes carry different congestion and loss components of LMP, and different commitment decisions produce different hour-mixes. Two identical units can differ by tens of dollars per MWh purely on nodal basis and which hours they chose to run.

Should a peaking plant try to raise its revenue per MWh?

Usually not directly. A peaker's high revenue per MWh comes from running only in expensive hours. Pushing it to run more lowers the metric while possibly raising total margin — or lowering it. Judge peakers on gross margin dollars and on availability during high-priced hours instead.

How does curtailment show up in the metric?

It shrinks the denominator. Curtailed output isn't delivered, so it isn't counted. Economic curtailment during negative prices actually raises revenue per MWh and is the correct decision. Network curtailment during high-priced hours costs real money and should be valued at prevailing prices.

What's the right reporting cadence?

Daily for realized price against index at the desk level, weekly for capture rate by asset, monthly for the full decomposition reconciled to final settlement, and annually for structural review of contract position and capture-rate assumptions against the changing generation mix.

Does revenue per MWh work for battery storage?

Only with adjustment. Revenue per MWh discharged ignores charging cost and round-trip efficiency losses, flattering the asset. Use spread capture per MWh discharged, net of charging cost grossed up for efficiency, and subtract amortized cycle degradation before comparing against alternatives.

FAQ

What exactly counts as revenue in the numerator?

Energy settlement across day-ahead and real-time, capacity payments net of any non-performance charges, ancillary services reservation and performance payments, renewable or other environmental certificate sales, and the net settlement of any bilateral contracts or financial hedges attributable to the period. Excluding hedge settlement is the most common omission and it can be the largest single term for a contracted asset.

Why is capture rate more useful than the raw metric?

Because it separates market level from operator skill. If regional prices fall broadly, every generator's revenue per megawatt-hour falls together and the metric tells you nothing about your own decisions. Capture rate compares your realized price to the index over the same period, isolating the timing, location, and availability choices you actually control from the price environment you don't.

Can capture rate legitimately exceed 100%?

Yes, and for dispatchable resources it should. A unit that runs only in high-priced hours realizes an average above the flat time-weighted index by construction. Intermittent resources generally sit below 100% because their output concentrates in hours their own technology has made abundant and therefore cheap. Comparing the two figures across technologies without adjusting for that structural difference is meaningless.

How should capacity revenue be allocated per MWh?

Report it both ways. The per-kilowatt-year form is how the auction cleared and how it should be compared year over year. The per-MWh form is arithmetic: divide by capacity factor times hours in the period. Always show generation volume beside the per-MWh figure, because a lower-running unit shows a higher capacity contribution per MWh for reasons that have nothing to do with performance.

What single change most reliably improves realized revenue?

Aligning availability with price. Moving planned maintenance out of historically high-priced windows, and reducing forced outages specifically in the highest-priced hours, moves the number more than most bidding refinements — because such a large share of annual margin concentrates in so few hours. A price-weighted availability metric will surface this; a plain availability percentage will not.

Is the metric still meaningful in a regulated, cost-of-service market?

Less so as a performance measure. In a regulated context, allowed revenue is set by tariff against approved costs and an authorized return, so revenue per megawatt-hour largely reflects the rate case rather than market execution. It remains useful for cost benchmarking across a fleet and for comparing regulated economics against what the same asset might earn merchant, but it is not measuring trading performance.

Sources

• PJM — Markets and Operations
• ERCOT — Market Prices
• CAISO — Today's Outlook and Market Data
• U.S. Energy Information Administration — Electricity Data
• FERC — Electric Power Markets
• NYISO — Energy Market and Operational Data
• MISO — Markets and Operations
• Lawrence Berkeley National Laboratory — Electricity Markets and Policy
• NREL — Energy Analysis
• ISO New England — Markets

Related on PULSE

• [Revenue Per Megawatt-Hour (MWh) for Independent Power Producers](/knowledge/ik0497)
• [Top 10 Energy Production Cost per MWh and Revenue per Barrel Metrics](/knowledge/ik0557)
• [Top 10 Airlines Revenue per Available Seat Mile Performance Indicators](/knowledge/ik0512)
• [Top 10 Hospitality Revenue per Available Room Performance KPIs](/knowledge/ik0490)
• [What are the key sales KPIs for the Commercial Building Energy Management Systems industry in 2027?](/knowledge/ik0296)
• [What are the key sales KPIs for the Commercial Solar Battery Energy Storage System (BESS) Integration industry in 2027?](/knowledge/ik0289)