Sizing the Energy: Why Four Hours and Not Five

Sizing the Energy: Why Four Hours and Not Five

September 18, 2026 · Dezzmond Team
Storage Sizing & Dispatching a Battery Battery Sizing

NREL’s 2024 ATB uses a four-hour utility-scale battery as its default capacity-factor case because four-hour systems were expected to be common, while also modeling 2-, 4-, 6-, 8- and 10-hour systems. That is a useful benchmark, not an economic optimum.

There is one, and it produces an uncomfortable result. On the same price shape, the same costs and the same cost of capital, the optimal duration is:

1 cycle per day        →  2 hours
1.4 cycles per day     →  3 hours
2 cycles per day       →  4 hours

Duration and cycling frequency are coupled decisions. A longer asset only earns the incremental energy-capacity investment when the market, contract or capacity construct creates enough valuable hours and the operating / warranty envelope allows the throughput.

This post builds the calculation: the marginal hour's revenue, the marginal hour's cost, where they cross, and the physical limit that caps the whole thing.

ℹ️ Note: The price shape is a single representative volatile day used to demonstrate method. Real analysis needs the full 8,760-hour price and market series for the specific node. The structure of the result generalises; the specific crossing point does not.

Why the optimal duration is a marginal-hour decision:

flowchart LR
  H1[Hour 1\nHighest-value spread] --> H2[Hour 2\nNext-best spread] --> H3[Hour 3] --> H4[Hour 4] --> H5[Hour 5+\nLower marginal value]
  C[Marginal cell cost per added MWh\nroughly flat assumption] -. compare each hour .-> H1
  C -.-> H3
  C -.-> H5

What Is the Duration Decision?

The choice of how many megawatt-hours to build behind a given megawatt of power. Expressed as hours — a 100 MW battery with 400 MWh is a four-hour battery — and decided by comparing what each additional hour earns against what it costs.

Both sides of that comparison have a specific shape, and the shapes are what make the answer non-obvious.

The Marginal Hour Earns Less Than the One Before It

Because the battery takes the best hours first.

A one-hour battery discharges into the single most valuable hour of the day and charges in the single cheapest. A two-hour battery adds the second best hour and the second cheapest — both worse than the first pair, by construction.

On the representative day, at 88% round-trip efficiency:

Hour added Marginal revenue per cycle
1st $85.91/MW
2nd $77.77
3rd $67.77
4th $50.36
5th $32.09
6th $26.82
7th $19.82
8th $13.68

The decline is not gentle. The eighth hour earns 16% of what the first does. And the shape is a property of the price distribution, not of the battery — a market with a flatter daily profile has a steeper decline, and a market with two distinct peaks has a flatter one.

This is why duration decisions do not transfer between markets. The same asset, the same cost, a different price shape, and a different answer.

The Marginal Hour Can Be Approximated With an Energy-Cost Curve

As a first-order approximation, adding duration behind a fixed MW rating adds energy-related equipment while much of the power-related balance of system is already installed.

As the next post sets out in detail, battery cost splits into a power component and an energy component:

Total cost ($/kW) = Pack cost ($/kWh) × duration (hours) + BOS cost ($/kW)

NREL’s utility-scale storage methodology explicitly separates a power component and an energy component, which is why duration can be modeled as energy cost × hours + power cost. But the marginal hour is not literally “cells only” or perfectly constant: containerization, HVAC, DC equipment, installation, augmentation design and procurement scale can move the incremental $/kWh. The worked example uses a flat $180/kWh energy component only to make the crossing visible:

Pack cost                                          $180/kWh
Per MW, per added hour                             $180,000
Annualised at 12% over 15 years (÷ 6.8109)      = $26,428/MW/yr

Under that simplifying cost assumption, the energy-cost line is flat while the marginal-revenue curve declines. The optimum is the last hour whose expected incremental value exceeds its incremental cost, subject to the physical and contractual constraints discussed below.

Where They Cross

Revenue per year is the marginal hour's revenue per cycle multiplied by the number of cycles. So the number of cycles moves the entire revenue curve up or down against a fixed cost line.

Hour At 365 cycles At 730 cycles Annual cost
1st $31,357 $62,714 $26,428
2nd $28,387 $56,774 $26,428
3rd $24,737 $49,474 $26,428
4th $18,383 $36,765 $26,428
5th $11,713 $23,426 $26,428
6th $9,789 $19,577 $26,428

Read down each column to the last row that clears the cost.

On the worked day and cost assumptions, one full cycle per day supports two hours before the next marginal hour falls below the assumed annualized cost. At two full cycles per day, the same simplified arithmetic supports four hours. Those are scenario outputs, not duration recommendations.

Same battery, same market, same cost of capital. The only thing that changed was how often it runs.

Four Hours Is a Statement About the Whole Operating Case

That is the substantive finding, and it reframes the question.

In the worked example, four hours only clears because the assumed second daily cycle lets the deeper hours earn enough revenue. In a real model, some days may support two cycles, some one, and some none. Duration should therefore be derived from the hour-by-hour marginal value distribution and feasible dispatch, not from a fixed average cycle count chosen first.

A Time-Budget Feasibility Check

For a simplified full-power, full-depth cycle there is also a time budget: charging and discharging both consume hours.

Discharging for D hours requires charging for D / RTE hours, because round-trip efficiency means more energy goes in than comes out. So each cycle commits:

Hours per cycle = D + D / RTE = D × (1 + 1/0.88) = D × 2.136

Run that against the possibilities:

Cycles/day Duration Hours committed of 24
1 2 hours 4.3
1 4 hours 8.5
1 6 hours 12.8
2 2 hours 8.5
2 4 hours 17.1 tight
2 6 hours 25.6 impossible

The last row is not a commercial judgement. Twenty-five point six hours do not exist in a day.

Maximum duration at 2 cycles/day = 24 ÷ (2 × 2.136) = 5.62 hours

Under the deliberately simple assumptions above — same charge/discharge MW, full-depth cycles and all round-trip loss allocated to charging — two complete cycles create a theoretical time-budget limit near 5.6 hours. That is not a physical law or an explanation for industry duration. Partial cycles, asymmetric charge/discharge power, ancillary-service commitments, SOC bounds, variable efficiency and multi-day operation change the feasible schedule. Use the chronological dispatch model as the real test.

What Would Move the Answer?

Four things, and none of them is the battery.

A wider price shape. Larger spreads lift the whole revenue curve and push the crossing point later. This is the mechanism by which growing renewable penetration lengthens optimal duration — more midday depression and steeper evening ramps mean more valuable hours to fill.

A second daily peak. A market with distinct morning and evening peaks supports two cycles naturally. One with a single evening peak does not, and no amount of willingness to cycle creates a second spread that is not there.

Lower energy-capacity cost. The marginal cost line falls, so more hours can clear it. Whether this is the dominant driver depends on the market’s price shape, capacity value, charging availability and contract structure.

Capacity accreditation. Where duration changes accreditation, the capacity value can justify energy capacity that arbitrage alone would not. PJM’s published 2026/27 3rd Incremental Auction class ratings, for example, are 54% for 4-hour storage, 61% for 6-hour, 65% for 8-hour and 73% for 10-hour. Those ratings are delivery-year specific and should be modeled directly rather than converted into a permanent ladder.

So Why Not Five?

Because the fifth hour has to earn its own incremental cost.

In the worked arbitrage-only example, the fifth hour at two assumed full cycles per day earns less than the assumed annualized energy cost and therefore does not clear. A capacity market can change that answer, but do not spread an accreditation step evenly across the hours between two classes. If PJM publishes a 4-hour class rating and a 6-hour class rating, a 5-hour project must be modeled under the applicable tariff / accreditation methodology rather than by giving “half” of the 4-to-6-hour uplift to the fifth hour.

That changes the practical process:

  1. solve the energy-only marginal value from the hourly dispatch;
  2. apply the actual capacity-accreditation treatment for the proposed duration;
  3. add any contracted duration payment or penalty;
  4. compare the resulting incremental value with the incremental energy-capacity cost.

The fifth hour is justified only if that combined project-specific value clears its cost.

Why the Answer Has Been Lengthening

Because lower energy-capacity cost can move later marginal hours above the value threshold even when the revenue curve is unchanged.

Run the same calculation at a pack cost of $120/kWh instead of $180:

Marginal cost at $180/kWh                          = $26,428/MW/yr
Marginal cost at $120/kWh                          = $17,619/MW/yr
Arbitrage only With accreditation
1 cycle/day 2 → 4 hours 2 → 5 hours
2 cycles/day 4 → 6 hours 5 → 7 hours

In this worked example, reducing the assumed energy-component cost from $180/kWh to $120/kWh moves the crossing materially outward. That illustrates the mechanism; it does not imply that a one-third pack-cost decline generally doubles optimal duration. The result must be recomputed on the actual cost curve and hourly revenue case. A specification of four hours written when cells were expensive is not a technical conclusion — it is a price observation that has since been overtaken.

Two implications worth carrying.

Duration decisions should be dated. A configuration standard adopted three years ago encodes a cost that no longer applies, and re-running the calculation is cheap.

Procurement timing interacts with configuration. A project procuring cells in a falling market should be asking whether its duration specification was set against the price it will actually pay, or against the price when the specification was written.

What If the Offtake Specifies It?

If the tolling agreement fixes MW, usable MWh or duration, some of the configuration becomes a contractual requirement rather than an unconstrained design output. The calculation in this post does not disappear — it becomes a negotiating input instead of a design one.

Three uses for it in that setting.

Pricing the toll. If the required configuration differs from the owner’s merchant optimum, the toll price should compensate for the contracted capability and risk allocation actually being provided.

Testing the throughput limit. The toll's cycle cap and the duration together determine the hours committed per day. A four-hour toll permitting two cycles commits 17.1 hours of every 24 — which is tight, and worth confirming the toller understands before it is agreed rather than after.

Establishing who benefits from cheaper cells. A long toll signed at today's configuration fixes the duration for its term while cell prices continue to fall. Whether the owner may augment to a longer duration, and who captures the resulting value, is a clause worth having.

How Do You Build the Duration Calculation in Excel?

As two curves and a crossing, with the cycling assumption stated as an input rather than buried.

The marginal revenue curve

Sort the day's prices ascending:           s = SORT(prices)
For duration d:
   Sell(d) = SUM(LARGE(s, 1:d))
   Buy(d)  = SUM(SMALL(s, 1:d)) / RTE
   Margin(d) = Sell(d) − Buy(d)

Marginal_Hour(d) = Margin(d) − Margin(d−1)

The marginal cost line

Marginal_Cost = Energy_Linked_$/kWh × 1,000 ÷ Annuity(Cost_of_Capital, Life)
              = 180,000 ÷ 6.8109                    = $26,428/MW/yr

The optimum

Optimal_Duration = MAX d such that Marginal_Hour(d) × Cycles ≥ Marginal_Cost

The feasibility check that must sit alongside it

Hours_Committed = Duration × Cycles × (1 + 1/RTE)
IF Hours_Committed > 24 THEN infeasible

Max_Duration = 24 ÷ (Cycles × (1 + 1/RTE))

Under the simplified full-depth, equal-power assumptions above, a six-hour asset cannot complete two full cycles inside 24 hours. A chronological dispatch model should be the final feasibility test.

The output that matters

PF_OptimalDuration              (hours)
PF_ImpliedCyclesPerDay          the assumption it depends on
PF_MaxFeasibleDuration          the physical ceiling
PF_MarginalHourValue            at the chosen duration

The second line is the one to publish alongside the first. A duration recommendation without its cycling assumption is half a statement, and the half that is missing is the one that moved the answer from two hours to four.

ℹ️ Note: Run this on the full annual price series, not a representative day. A single day fixes the shape and removes the seasonal variation that determines how many days actually support two cycles — which is the input the whole calculation turns on.

Why the annual series matters more than the method

The representative day above makes the mechanism visible and overstates the precision of the answer, for a specific reason worth naming.

Cycling frequency is not a choice made once. It is a daily outcome of whether that day's price shape contains a spread wide enough to clear the round-trip efficiency loss and the degradation cost. Some days offer two such spreads, many offer one, and a meaningful number offer none worth taking.

So "two cycles a day" is not an operating policy — it is an annual average of a highly variable daily count, and the distribution behind it matters as much as the mean. A year averaging two cycles made up of 150 two-cycle days, 150 one-cycle days and 65 idle days is a different asset from one cycling steadily twice every day, even though both produce 730 cycles.

Two consequences for the duration calculation.

The marginal hour is used less often than the average implies. The fifth hour of a five-hour battery is only used on days with an exceptionally wide spread. Its effective utilisation is therefore well below the headline cycle count, which pushes the optimum shorter than a flat average suggests.

Seasonality concentrates the value. In most markets the widest spreads cluster in a few months. A duration justified by summer economics is carrying capital through a winter that does not use it, and the annual calculation is what reveals whether that trade works.

The practical fix is to compute the marginal hour's utilisation directly from the price series — the count of days on which the battery would actually have reached that depth — rather than multiplying a marginal value by an assumed cycle count. It is more work and it is the difference between a defensible duration and a plausible one.

To build the marginal revenue curve, the cost line, the crossing and the feasibility ceiling, prompt Dezzmond with your price series, pack cost and cost of capital.

What Do Developers and Lenders Actually Check?

  • What cycling assumption is the duration based on? It determines the answer.
  • Is that cycling rate achievable given the price shape — are there two distinct spreads a day?
  • Does the warranty permit it? Throughput limits bind independently of economics.
  • Is the configuration chronologically feasible? Use the full dispatch schedule; the Duration × cycles × 2.136 check is only the simplified 88%-RTE example.
  • What is the marginal hour's value at the chosen duration, and how close is it to the cost line?
  • Is any hour justified by accreditation rather than arbitrage? Say which.
  • Has the calculation been run on the full year or on a representative day?

Frequently Asked Questions

Why is four hours the industry standard?

It is a common reference duration in today’s lithium-ion market, but there is no universal economic rule that makes four hours optimal. The project-specific answer depends on hourly spreads, cycling, cost, capacity rules, warranty, interconnection and contract.

Why does each additional hour earn less?

Because the battery discharges into the best hours first. The second hour is by construction less valuable than the first, and on the worked day the eighth hour earns 16% of what the first does.

Why is the marginal cost constant?

Because much of the power-related BOS is fixed per MW while additional duration adds energy-related equipment. NREL explicitly models utility-scale BESS cost with separate power and energy components, but the incremental $/kWh can still vary with duration and configuration.

Is there a physical limit on duration?

There is a chronological feasibility limit, but not one universal duration ceiling. Under the worked full-power/full-cycle assumptions, two complete cycles of a 6-hour asset would require more than 24 hours. Real feasibility depends on partial cycling, charge/discharge MW, SOC bounds, efficiency and multi-day scheduling.

So should it be four hours or five?

On the worked arbitrage-only case, four hours is the last duration clearing the assumed cost at two full cycles per day. A capacity market may justify more duration, but a five-hour asset must use the actual accreditation methodology; it is not valid to interpolate a published 4-hour and 6-hour class rating mechanically.

How much does a cheaper cell change the answer?

Materially in the worked case. Dropping the assumed energy-component cost from $180 to $120/kWh moves several marginal hours above the cost line. The magnitude is scenario-specific and should be recomputed from the project cost curve.

What lengthens optimal duration over time?

Lower energy-component cost, wider or more persistent spreads, capacity-accreditation value, longer-duration contracts and other products that pay for sustained output can all lengthen the economic optimum. Changes in saturation, warranty or charge availability can shorten it.

Closing: State the Cycling Assumption

The answer to “why four hours” should be a reproducible bridge: because, under this hourly price series, dispatch strategy, market-product treatment, warranty, energy-cost curve and financing assumptions, the fourth hour clears its marginal cost and the fifth does not.

Change any one of those and the answer can change. In the worked case, lowering utilization shortens the crossing and lowering the energy-linked cost lengthens it; a different price shape can change both.

None of that is difficult to compute. It is two curves, a crossing point and a feasibility check — a morning's work against a real price series, and it produces a number that can be defended rather than inherited.

A four-hour answer may survive the analysis, but the value is in knowing which assumptions make it clear. When cost, price shape, accreditation or throughput terms move, the configuration should be re-tested rather than inherited.

the next post takes the cost side of that crossing properly: why the cost per kilowatt-hour falls as duration rises, and what that does to the comparison.

Sources: NREL — Utility-Scale Battery Storage ATB; separate power and energy cost components for 2–10 hour systems · PJM — Final 2028/29 BRA ELCC class ratings for 4-/6-/8-/10-hour storage · NREL SAM — battery dispatch and lifetime modeling · Dezzmond analysis; marginal-hour, cycle-count and cost examples are illustrative.