Ancillary Services and the Battery Revenue Stack: The Inversion, and Why $/MW Is the Wrong Metric

Two numbers describe what happened to battery economics in ERCOT. Ancillary services fell from 84% of revenue to 48% in two years. Energy arbitrage went from 25% to 76% across a single year, June to June.

The revenue stack did not shrink so much as invert. And the volatility underneath it is severe: Modo Energy's ERCOT index went from $46,264/MW/yr in January 2026 to $15,306 in February, then back to $38,145 in April.

But the more useful observation is a different one. Two batteries can report the same revenue per megawatt and be consuming their asset at three times different rates, because of which products they sold. Revenue per megawatt is the number the market quotes. Revenue per cycle is the number that determines whether the asset was worth building.

This post covers the products, what saturated them, what replaced them, what RTC+B actually changed, and how to model a stack whose cost of earning varies by service.

ℹ️ Note: Market data here is point-in-time and moves fast. Every modelling figure is a labelled assumption, and the degradation economics in particular depend on warranty terms that are project-specific.

What Are the Products?

Four services, procured day-ahead through co-optimised offer curves, all of them keeping frequency near 60 Hz or holding capacity in reserve.

Service	What it does	Suits
Regulation Up / Down	Corrects small, continuous frequency deviations around 60 Hz	Fast assets — batteries
Responsive Reserve Service (RRS)	Very fast response to frequency events, short duration	Fast assets — batteries
ERCOT Contingency Reserve (ECRS)	Longer-duration contingency cover, introduced June 2023	Longer duration, longer ramp
Non-Spinning Reserve	Offline capacity brought on with notice	Longer duration, longer ramp

Modo's framing of the split is the one to carry: batteries "are particularly suited to providing Regulation and Response Reserve — because those services require very fast response, and have shorter maximum durations," while ECRS and Non-Spin favour "technologies that can provide power for longer durations, and are available to assets with longer ramp times."

That is a technical statement with a large commercial consequence, which the rest of this post is about.

Why Did the Ancillary Services Market Saturate?

Because it is small, and batteries are very good at it.

Ancillary service procurement is sized to the system's reliability need, not to the amount of storage wanting to provide it. It is a few thousand megawatts of requirement in a market that had 14.96 GW of installed battery capacity by the end of Q1 2026. Once enough batteries exist to cover the requirement several times over, the clearing price falls to whatever the marginal offer is willing to accept.

The collapse has been fast and large:

• Ancillary service revenues have fallen nearly 90% since 2023.
• Per-megawatt returns compressed 71% from 2023 to 2024.
• Average annual BESS revenue fell from $149/kWh in 2023 to a projected $17/kWh for 2025.
• October 2025 revenues settled at $2.03/kW for the month, 40% below October 2024.

This is the same mechanism the accreditation post described, arriving through a different market. A technology enters, solves the problem it is good at, and the payment for solving it falls because the problem is solved. Batteries did not fail at ancillary services. They finished them.

What Replaced It?

Energy arbitrage — buying low, selling high, in a market whose volatility is the product being sold.

The shift is the headline of the last two years. ERCOT batteries earned 76% of revenue from energy arbitrage in June 2025, up from 25% a year earlier, with other mid-2025 readings putting the mix nearer 60–65% arbitrage and 33% ancillary services.

ℹ️ Note: Those two readings differ because they are different months in a market that swings by two-thirds month to month. Take the direction, not the decimal. Any battery model calibrated on a single month's mix is calibrated on noise.

Arbitrage is a fundamentally different business from ancillary services, and the difference is not only where the money comes from.

An ancillary service award is paid for availability. The battery holds capacity, is paid whether or not it is called, and may do very little physical work. Arbitrage is paid for work. Every dollar requires a charge and a discharge — a cycle — and cycles are the consumable the asset is made of.

The revenue stack did not just rotate. It moved from a product that paid the battery for existing to one that pays it for wearing out.

The Metric That Matters: Revenue Per Cycle

This is where the analysis gets useful, and where most revenue models stop too early.

Modo's cycle-value work found that in the second half of 2023, ERCOT batteries averaged $1,963/MW per cycle — with enormous dispersion. Some earned less than $500/MW per cycle; one outlier reached $15,000/MW per cycle. Across the same period, batteries averaged 0.77 cycles per day, with some running nearly two cycles daily and others less than one cycle every ten days.

A twenty-fold spread in revenue per cycle, and a twenty-fold spread in cycling rate, inside the same market in the same six months.

Those are not differences in luck. They are differences in strategy, and the strategy determines how fast the asset is consumed.

It is worth sitting with the 0.77 cycles per day figure, because it contradicts the way batteries are usually modelled. A financial model typically assumes one full cycle per day, sometimes two, as a reasonable operating baseline. The observed fleet average was well under one — and the dispersion around it means the average describes almost none of the individual assets. Some batteries were running at nearly two cycles daily and consuming warranty throughput at roughly twenty-five times the rate of those cycling once every ten days.

Two conclusions follow for anyone building a model. A cycling assumption is a strategy assumption, not a technical parameter, and it should be stated and defended as one. And because the dispersion is so wide, a fleet average is close to useless as an input — the right approach is to model the strategy the asset will actually run, and then check that the warranty permits it.

Regulation Versus Reserve: The Trade Nobody Prices

The single most important operational finding, and it is stark.

Batteries pursuing Regulation contracts were 13 times more likely to have to physically respond than those on Reserve services. Regulation corrects small, continuous deviations, so it calls constantly. Reserve services — ECRS and RRS — sit and wait for events that mostly do not happen.

The result at the portfolio level: batteries with Regulation-dominant strategies cycled approximately three times more often than those prioritising Reserve.

Now put that next to the revenue. Regulation and Reserve can clear at similar enough prices that a battery reports comparable revenue per megawatt on either strategy. But one of them has cycled three times as hard to get there.

Modo is explicit about what that costs: "more frequent cycling leads to faster degradation and lower maximum charging capacity, and can have warranty implications with cell manufacturers. This ultimately leads to higher maintenance costs, increased downtime, and decreased enterprise value of the battery."

There is a structural driver worth noting too. Shorter-duration batteries — under an hour — and smaller assets under 10 MW tended to pursue Regulation-heavy strategies precisely because it let them compete on a dollars-per-megawatt basis with longer-duration systems. They bought headline revenue with asset life. On the metric the market quotes, they kept up. On the metric that matters, they did not.

What Does Duration Do to the Stack?

It determines which products the asset can sell, and the answer has moved.

A sub-one-hour battery is a Regulation and RRS asset almost by construction. It can hold a fast, short-duration obligation and very little else, and as the Modo data shows, that is exactly the strategy those assets adopted — competing on dollars per megawatt by cycling hard.

A two-hour battery can do arbitrage on a daily spread and sit in the reserve products. It is the configuration most of the ERCOT fleet was built at, and it is the one most exposed to the inversion described above: it lost the ancillary service revenue that justified it and has to earn the replacement from a spread it can only capture twice a day.

A four-hour and longer battery reaches products the shorter assets cannot, captures wider spreads, and — in a capacity market — earns a materially better accreditation. The ladder from the previous post prices this explicitly: 55% at four hours, 65% at six, 67% at eight, 75% at ten.

The economics have therefore moved in the same direction as the accreditation rules, for a related reason. When ancillary services paid well, short duration was efficient: the product had a short maximum duration, so paying for energy capacity the product could not use was waste. Now that most revenue comes from arbitrage and capacity accreditation, energy capacity is the thing being sold and duration is the constraint.

That is a hard lesson for an existing fleet. A one-hour battery built in 2022 against a market that no longer exists cannot be lengthened cheaply, and its strategic options are narrower than they were on the day it was financed — which is the clearest illustration in this series of what merchant revenue risk actually means.

Is a Tolling Agreement the Answer?

For bankability, usually yes. For value, it depends what is given away.

Under a tolling agreement an offtaker pays a fixed capacity payment — typically in dollars per kilowatt-month — and takes the right to dispatch the asset, together with the market revenue and the market risk. The owner becomes, in effect, a provider of a service rather than a merchant trader.

The case for it is straightforward given everything above. The merchant stack swings by two-thirds month to month, has no floor, and is exposed to a market redesign that arrived in December 2025 and will arrive again. A fixed payment is bankable in a way that none of that is: it supports leverage, it survives a lender's downside case, and it converts an unfinanceable asset into a financeable one.

The case against is that the toller captures the upside and, more subtly, controls the cycling. This is where the cycle analysis becomes a contract question rather than an operating one. If the toller dispatches the asset three times harder than the owner would have, the owner bears degradation it did not choose, against a payment that did not price it.

The market's answer is a throughput limit in the tolling agreement — a cap on cycles per year or on energy throughput, with a payment or penalty above it. That clause is doing the same work as the breakeven cycle value in the model above, expressed contractually rather than analytically, and it deserves the same attention an availability guarantee or an LD cap would get.

Sponsors who negotiate the capacity payment carefully and the throughput limit loosely have optimised the visible number and given away the one that determines the asset's condition at the end of the term.

What Did RTC+B Actually Change?

Less than the attention it received, and it is worth being precise because the previous post in this series was looser than it should have been about this.

Real-Time Co-optimisation Plus Batteries, live since 5 December 2025, co-optimises energy and ancillary service awards across day-ahead and real-time, adds state-of-charge accounting for batteries, and introduces separate day-ahead products — ECRS_DAM, RRS_DAM — visible in the revenue stack.

Modo's assessment is the honest one: RTC+B "reworked the mechanics for only a small, ancillary slice of revenue. Batteries still earn most of their money from energy arbitrage, which RTC+B left largely unchanged beyond a marginal efficiency gain."

So the correct modelling response is narrower than a rebuild. Re-model the ancillary services leg under the new co-optimised products; leave the arbitrage leg largely as it was, with a modest efficiency uplift and an expectation of tighter day-ahead to real-time convergence over time. A wholesale reconstruction of the arbitrage model on RTC+B grounds is work that the change does not justify.

The state-of-charge accounting is the piece with the longest tail. A dispatch engine that understands a battery's energy limitation should, over time, stop awarding it obligations it cannot physically meet — which improves both the quality of the awards and the accuracy of any model that assumes awards are deliverable.

Does the Same Pattern Hold Outside ERCOT?

The direction does; the shape of the stack does not.

ERCOT is the extreme case because it is energy-only. With no capacity market, a battery's entire revenue comes from products that are either saturating or volatile, which is why the inversion was so abrupt and why the merchant risk is so raw.

In PJM the same asset has a third leg. Capacity revenue on the duration ladder is a separate, forward-procured line — and as the previous post set out, a four-hour battery accredited at 55% of nameplate at $325/MW-day earns roughly $65,000 per megawatt-year before it sells a single megawatt-hour of energy. That is comparable to, and currently larger than, the entire annualised ERCOT merchant index. A capacity construct does not remove volatility from the arbitrage leg; it puts a substantial, contracted-for-three-years floor underneath it.

The general lesson is one the ERCOT numbers make vivid but which applies anywhere: ancillary service requirements are fixed quantities and will saturate wherever enough storage is built. The question is what else the market offers when they do. In an energy-only market the answer is arbitrage alone. In a capacity market it is arbitrage plus an accredited capacity payment that rewards duration — the same duration that improves the arbitrage.

That alignment is not accidental, and it is a reasonable argument for siting long-duration storage in capacity markets and shorter-duration, faster assets where ancillary services are still scarce. The mistake is assuming an ERCOT revenue history transfers to a market with a different stack, or the reverse.

How Do You Model the Stack in Excel?

With a cost of earning attached to each revenue line. A stack model that sums revenue without charging cycles against the asset will rank strategies in the wrong order.

The inputs

Assumptions, labelled as such:

Power                                           100 MW
Duration                                        2 hours  (200 MWh)
Installed cost                                  $250/kWh
Capex per MW        2 MWh × $250 × 1,000       = $500,000/MW
Warranty throughput                          6,000 cycles
Throughput-attributable cost  500,000 ÷ 6,000  = $83.33/MW/cycle
Gross revenue (2026 index, annualised)        = $33,000/MW/yr

The two strategies, at identical gross revenue

RESERVE-HEAVY (ECRS / RRS)
   Cycling rate            0.4 /day → 146 cycles/yr
   Gross revenue                            = $33,000/MW/yr
   Throughput cost   146 × $83.33           = ($12,167)/MW/yr
   Net                                      = $20,833/MW/yr

REGULATION-HEAVY
   Cycling rate            1.2 /day → 438 cycles/yr
   Gross revenue                            = $33,000/MW/yr
   Throughput cost   438 × $83.33           = ($36,500)/MW/yr
   Net                                      = ($3,500)/MW/yr

Same revenue. One strategy earns $20,833 per megawatt-year and the other loses $3,500. The difference is entirely the cost of earning, and it is invisible in every revenue figure the market publishes.

Revenue per cycle as the operating target

Reserve-heavy     $33,000 ÷ 146              = $226/MW/cycle
Regulation-heavy  $33,000 ÷ 438              =  $75/MW/cycle
Breakeven                                     =  $83/MW/cycle

That breakeven line is the number an operator should have on the wall. Below roughly $83/MW/cycle on these assumptions, a cycle destroys more value than it creates, and the correct action is to stand down rather than to chase volume.

The volatility the annual number hides

January 2026    $46,264/MW/yr
February 2026   $15,306/MW/yr    (−67%)
April 2026      $38,145/MW/yr    (+149%)

A model running on an annual average will show a stable asset. The asset is not stable; it is an average of months that differ by a factor of three. Size any debt against the distribution, not the mean — and note that a battery has no contracted floor unless a tolling agreement provides one.

ℹ️ Note: The throughput cost above charges capex fully against cycles, which overstates it — batteries also age calendrically. The refinement is to split installed cost between cycle-limited and calendar-limited components and charge only the first per cycle. Doing so lowers the breakeven but does not change the ranking, which is the result that matters.

To build the product-by-product stack with cycle costing and the breakeven cycle value, prompt Dezzmond with your power, duration, warranty terms and market.

What Do Developers and Lenders Actually Check?

• What is the revenue per cycle, not just per megawatt?
• What is the breakeven cycle value at which a cycle stops being worth running?
• What cycling rate does the operating strategy imply, and does the warranty permit it?
• Is the ancillary service revenue modelled post-RTC+B with the new day-ahead products?
• What share of revenue is arbitrage — and has it been calibrated on more than one month?
• Is there a tolling agreement, or is the entire stack merchant?
• Does the model charge degradation against revenue at all? Many do not.

Frequently Asked Questions

What are ERCOT's ancillary services?

Regulation Up and Down, Responsive Reserve Service, ERCOT Contingency Reserve Service and Non-Spinning Reserve. Batteries are best suited to Regulation and RRS, which require very fast response over short durations.

Why did battery ancillary service revenue collapse?

Because the requirement is fixed and the supply is not. Ancillary service revenues have fallen nearly 90% since 2023 as installed storage reached 14.96 GW against a requirement of a few thousand megawatts.

What replaced ancillary services in the stack?

Energy arbitrage, which went from about 25% of ERCOT battery revenue to roughly 76% in a year. It is a different business: ancillary services pay for availability, arbitrage pays for physical work.

Why does Regulation cost more than Reserve to provide?

Because it is called far more often. Batteries on Regulation were 13 times more likely to have to respond physically, and Regulation-dominant strategies cycled about three times more than Reserve-dominant ones — consuming warranty throughput to earn the same headline revenue.

Does a tolling agreement solve the volatility?

It makes the asset financeable by converting merchant revenue into a fixed capacity payment, but the toller takes the upside and controls dispatch. The clause that matters as much as the payment is the throughput limit, which caps how hard the asset can be cycled.

Does duration still matter now that ancillary services have collapsed?

More than before. Ancillary services had short maximum durations, so short assets were efficient. Arbitrage and capacity accreditation both reward energy capacity, and PJM's ladder runs from 55% at four hours to 75% at ten.

Did RTC+B change battery economics?

It changed the ancillary services leg, which is now under half the stack. Energy arbitrage — most of the revenue — was left largely unchanged beyond a marginal efficiency gain, so re-modelling should be targeted rather than wholesale.

Closing: A Stack That Pays for Work, Not for Waiting

The battery investment case of 2022 and 2023 was largely an ancillary services case: a small, high-priced market that storage could dominate on technical merit. That case is over, and it ended the way these things usually do — not through failure but through success, with prices falling roughly ninety percent as enough capacity arrived to meet a fixed requirement several times over.

What replaced it is a harder business to underwrite and a more durable one. Energy arbitrage is not capacity-limited in the way an ancillary service requirement is; it scales with volatility, and volatility in a market with growing renewable penetration and a halved real-time offer cap is its own forecasting problem.

The analytical shift that matters is smaller and more immediate. Once most revenue comes from work rather than from waiting, the asset has a marginal cost of earning, and a revenue model without one cannot distinguish a good strategy from a value-destroying one. Two ERCOT batteries reporting the same revenue per megawatt — one at $226 per cycle and one at $75 — are not in the same condition, and only one of them will be worth what the model says in year eight.

Series C now turns from what a project earns to whether it can connect at all. The next two posts cover interconnection: what FERC Order 2023 replaced the serial queue with, and what network upgrades and withdrawal penalties actually cost.

Sources: Modo Energy — ERCOT's Ancillary Services: A Beginner's Guide · Modo Energy — ERCOT: What Is the Value of a Cycle for Battery Energy Storage Systems? · pv magazine USA — Battery Energy Storage Revenues for Ancillary Services Fall Nearly 90% in ERCOT · Modo Energy — ERCOT Battery Storage in 2026: 7 Things to Watch