The Cost Curve: Why Dollars per Kilowatt-Hour Needs a Duration Attached

The Cost Curve: Why Dollars per Kilowatt-Hour Needs a Duration Attached

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

Start with a deliberately simplified comparison: two projects with the same assumed energy-component cost and the same assumed power-component cost, but different duration.

One-hour system                                  $560/kWh
Eight-hour system                                $228/kWh

The eight-hour worked case looks 59% cheaper on a $/kWh basis. Under the simplified linear cost model below, that difference comes from spreading the assumed power-linked cost over more MWh. Real vendor pricing can also change with duration, equipment count, containerization and scale.

Dollars per kilowatt-hour mixes energy-linked and power-linked costs. It is useful when comparing like-for-like duration and scope, but it can be misleading across different durations because the denominator itself changes.

This post takes the cost structure apart: the energy and power components, the curve they produce, when a marginal-hour approximation is valid, and how to compare vendor quotes without letting duration hide scope differences.

ℹ️ Note: Cost figures are labelled assumptions calibrated to published US ranges. Actual pricing varies enormously by geography, chemistry, scale and procurement timing.

The cost curve is fixed power cost plus variable energy cost:

flowchart TD
  A[Power-linked costs\nPCS, transformer, grid, civil] --> C[Fixed cost per MW]
  B[Energy-linked costs\nCells + containers] --> D[Variable cost per MWh]
  C --> E[Total installed cost]
  D --> E
  E --> F{Increase duration at same MW}
  F --> G[$/kW rises]
  F --> H[$/kWh falls as fixed cost is spread]

The Cost Equation

Battery cost has two components that scale on different things.

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

The NREL utility-scale methodology separates costs into an energy component and a power component so duration can be varied consistently. That is a useful reduced-form model, not a statement that every physical component belongs entirely in one bucket.

Cells and much of the DC storage equipment scale strongly with MWh. PCS, transformers and some interconnection equipment scale strongly with MW. Containers, HVAC, DC cabling, foundations, controls, installation and soft costs can contain both fixed and energy-dependent elements.

The two-component equation is therefore a sizing approximation that should be calibrated to the vendor scope.

Assumptions used throughout:

Pack cost                                        $180/kWh
BOS cost                                         $380/kW

At four hours those assumptions produce $275/kWh. Treat that as an illustrative scenario, not a current market benchmark; current project pricing varies by scope, geography, interconnection, tax treatment, procurement date and warranty package.

The Curve

Duration $/kW $/kWh Marginal $/kWh Pack share of cost
1 hour $560 $560 — 32%
2 hours $740 $370 $180 49%
3 hours $920 $307 $180 59%
4 hours $1,100 $275 $180 65%
6 hours $1,460 $243 $180 74%
8 hours $1,820 $228 $180 79%
10 hours $2,180 $218 $180 83%

Three columns worth reading carefully.

The $/kWh column falls steeply and then flattens. Most of the decline happens by four hours. Going from one to four hours cuts it by 51%; going from four to ten cuts it by a further 21%. The curve is a hyperbola, and the interesting part is at the short end.

The marginal column is flat at $180 only because the worked model assumes a constant energy-component cost. NREL uses a separable power/energy framework for utility-scale BESS, but actual incremental duration can change container count, HVAC, DC collection, installation, warranty and procurement pricing. Use the project-specific marginal energy cost curve where available.

The energy-component share rises with duration in the worked model. That is the key economic point: longer-duration systems are more sensitive to energy-component pricing, while short-duration systems are more sensitive to power-related BOS. The exact shares are scenario outputs.

Why $/kWh Can Mislead Across Durations

Because the BOS cost is being spread over a denominator that duration alone controls.

Take the eight-hour system's $228/kWh. Decompose it:

Pack, per kWh of the system                                  $180
BOS, per kWh of the system    ($380/kW ÷ 8 hours)          =  $48
                                                             ----
Total                                                        $228

Now the one-hour system's $560/kWh:

Pack, per kWh of the system                                  $180
BOS, per kWh of the system    ($380/kW ÷ 1 hour)           = $380
                                                             ----
Total                                                        $560

The worked energy component is $180/kWh in both. Under the simplified two-component model, the difference is the power-linked term spread over a different number of hours. Real vendor pricing can also vary with duration and scope.

The consequence for anyone reading a quote, benchmark or competitor comparison: duration must be normalized before a blended $/kWh comparison is interpreted as procurement performance. Two quotes at different durations can show different $/kWh even if their assumed energy and power components are identical.

How to Compare Properly

Split the quote into its two components and compare each on its own scale.

Given a quoted total for a known configuration:

Pack_$/kWh = (Total_$ − BOS_$/kW × MW × 1,000) ÷ (MWh × 1,000)
BOS_$/kW   = (Total_$ − Pack_$/kWh × MWh × 1,000) ÷ (MW × 1,000)

With two quotes for different durations from the same vendor, both components can be solved directly:

Quote A: 100 MW / 200 MWh  =  $74,000,000
Quote B: 100 MW / 400 MWh  = $110,000,000

Difference                 =  $36,000,000 for 200 MWh
→ Pack = 36,000,000 ÷ 200,000 kWh                    = $180/kWh
→ BOS  = (74,000,000 − 180 × 200,000) ÷ 100,000 kW   = $380/kW

That is a useful way to estimate the implied energy-linked slope and power-linked intercept when the two quotes have comparable scope and the cost relationship is approximately linear. It does not reveal the vendor's internal cost structure; it reveals what the quoted configurations imply.

Compare vendors on normalized implied components and scope: warranty, augmentation commitment, PCS / transformer supply, civil works, interconnection equipment, commissioning, taxes and owner-retained costs.

What Sits Outside the Cell / Pack Cost?

A substantial portion of installed BESS cost can sit outside the cells themselves: power conversion, transformers, controls, containers, thermal management, civil works, installation, engineering and other balance-of-system items.

NREL’s utility-scale BESS cost model explicitly separates the battery pack from inverter and BOS components, and notes that the pack is only one part of total system cost. The share changes with duration because some costs scale with MW and others with MWh.

For project finance, do not rely on a universal “40% non-battery” share. Build the actual sources-and-uses from vendor scope, EPC scope, owner costs, interconnection and construction assumptions.

Geography Changes Both Components

Installed cost differs by geography, but it is not safe to explain a country-to-country gap as “essentially all BOS.” Cell / pack procurement, tariffs, domestic-content requirements, labor, EPC structure, interconnection, land, permitting, taxes and logistics can all move.

The useful comparison is therefore not a headline country $/kWh. Normalize:

  • chemistry and usable energy;
  • duration and MW rating;
  • AC versus DC measurement point;
  • included augmentation / warranty scope;
  • power-conversion and transformer scope;
  • interconnection and owner costs;
  • taxes, tariffs and logistics.

Only then compare the energy and power cost components.

What Falls Over Time, and What Does Not

The two components have behaved very differently, and the divergence is the main long-run driver of configuration.

Battery energy-component costs have declined materially over time, but future decline is a scenario rather than a certainty. NREL’s 2024 ATB points to its 2025 utility-scale battery cost projections and models conservative, moderate and advanced cost paths rather than one deterministic learning curve.

Power-related and installation costs need their own forecast. They do not necessarily follow the same trajectory as battery energy costs, which is why NREL separates power and energy components for duration modeling.

The consequence is a steady shift in the mix, and it changes the answer to every question in this series:

If pack falls 33% and BOS is flat:
   Pack share at 4 hours          65% → 56%
   Marginal hour cost             $180 → $120
   4-hour system $/kWh            $275 → $215

The marginal cost falls by the full 33% while the system cost falls by 22%. Duration decisions therefore move faster than project economics, which is why a configuration standard ages badly even when the overall business case looks stable.

The two components should therefore be forecast separately where the data support it. A single escalator on blended $/kWh can hide changes in duration economics.

Augmentation Needs Its Own Scope and Cost Curve

Augmentation is often more energy-heavy than the original build because the project may reuse the existing point of interconnection and much of the AC power equipment. But “augmentation = pack only” is too strong.

Depending on the design, augmentation can require additional battery blocks or containers, racks, thermal management, DC cabling, controls, fire-safety systems, civil work, commissioning and integration. Some projects may also need power-equipment or transformer changes.

Model augmentation from a vendor / EPC scope for the augmentation year:

Augmentation cost
= added usable MWh × energy-equipment cost
+ incremental container / BOS
+ installation / integration
+ owner costs
+ any required power-system upgrades

Use the expected future cost curve only as an explicit scenario. A future pack-price decline is not guaranteed and does not remove the non-pack augmentation scope.

What the Split Means for a Constrained Site

The cost structure interacts with the binding constraint in a way that is worth stating explicitly.

On a power-constrained site, much of the power-related infrastructure may already be fixed, so incremental duration is more energy-heavy. It is still not automatically “pack only”: verify the incremental DC/BOS, layout and integration cost.

On an energy-constrained site, the pack is capped by land and fire code. Additional spend can only buy power, at $380/kW, earning capacity and ancillary revenue. The duration question does not arise; the power question does.

On a budget-constrained site, the trade-off is live in both directions and the frontier from the constraint-stack post applies:

Budget = MW × $380,000 + MWh × $180,000

Every dollar moved from power to energy buys 380/180 = 2.11 MWh per MW given up. That exchange rate is the entire configuration decision on a budget-constrained project, and it is computable before any revenue modelling.

What the Curve Says About Long-Duration Storage

A useful implication falls out of the pack-share column, and it explains why long-duration storage is a different investment proposition rather than simply a longer one.

In the worked two-component model, the energy component rises from 32% of total cost at one hour to 83% at ten hours. That makes longer-duration systems more sensitive to energy-component pricing, but it does not make power conversion, interconnection, installation or site costs disappear.

Three things follow.

Longer duration increases sensitivity to the energy component, but long-duration economics are not “almost entirely” a cell-cost story. Installation, land, thermal management, power conversion, interconnection and technology-specific BOS can remain material, especially for non-lithium technologies.

Energy-cost changes pass through more strongly as duration increases in the two-component framework. The exact pass-through depends on the project’s energy/power cost split and should be calculated rather than inferred from duration alone.

It also explains why alternative chemistries target long duration first. A technology with a lower energy cost and a worse power cost — flow batteries being the obvious example — is uncompetitive at one hour, where BOS dominates, and can be competitive at ten, where it does not. The crossover is a function of the same two-component structure this post has been describing, with different coefficients.

None of that makes long duration economic on its own; the previous post's marginal-value calculation still governs. Longer-duration value can depend on energy shifting, capacity, transmission / congestion, resilience, contract structure and technology-specific cost — not only on cell prices.

Chemistry Changes Several Inputs at Once

Chemistry and system architecture can change more than the cell price:

  • usable energy density and footprint;
  • round-trip efficiency;
  • thermal-management design;
  • cycle and calendar degradation;
  • allowable C-rate;
  • safety / fire-test basis;
  • warranty and augmentation strategy;
  • energy-linked and power-linked capex.

NREL's ATB uses LFP as the representative current stationary-storage lithium-ion technology. That does not justify a universal table saying LFP always has lower BOS than NMC or that NMC always wins on constrained land.

For a live project, use the OEM's actual:

usable MWh per block
PCS / power architecture
efficiency curve
warranty throughput / retention
fire-test basis
temperature limits
footprint
quoted capex scope

Chemistry is therefore a multi-variable design choice inside the constraint stack, not merely a $/kWh switch.

How Do You Build the Cost Curve in Excel?

As two inputs and a function, never as a single $/kWh assumption.

The core

Cost_per_kW(d)   = Pack_$kWh × d + BOS_$kW
Cost_per_kWh(d)  = Cost_per_kW(d) / d
Marginal_$kWh    = EnergyLinked_$kWh              (constant only in this linear worked model)
Total_Capex      = Cost_per_kW(d) × MW × 1,000

The two outputs to publish

PF_EnergyLinkedCost_$kWh                          $180   worked assumption
PF_PowerLinkedCost_$kW                             $380   worked assumption

Those two worked drivers are enough for the simplified sizing curve. A live project may need additional fixed, site-specific and non-linear cost terms, but one blended $/kWh still cannot answer the marginal-duration question.

The exchange rate on a budget constraint

PF_MWh_per_MW_traded = BOS_$kW × 1,000 ÷ (Pack_$kWh × 1,000)
                     = 380 ÷ 180                        = 2.11

The sensitivity that actually matters

Pack cost −33% ($180 → $120)
   4-hour $/kWh    $275 → $215        (−22%)
   1-hour $/kWh    $560 → $500        (−11%)
   Marginal hour   $180 → $120        (−33%)

Note the three different percentages from one input change. The headline project cost moves least on a short-duration system, most on a long one, and the marginal cost moves by the full amount — which is why cell price declines change duration decisions more than they change project costs.

ℹ️ Note: Never carry a single blended $/kWh cost assumption into a sizing model. It cannot produce a marginal cost, so it cannot answer the duration question, and it silently embeds the duration it was derived from.

To build the cost curve, decompose vendor quotes into pack and BOS, and run the budget frontier, prompt Dezzmond with your quotes and configuration options.

What Do Developers and Lenders Actually Check?

  • Is the cost model two components or one blended $/kWh? One cannot answer the marginal question.
  • What are the pack and BOS figures separately, and how were they derived?
  • Have vendors been asked to quote two durations so the structure can be solved?
  • Is a $/kWh comparison being made across different durations? It is meaningless if so.
  • What share of cost is not the battery, and how does that change the pass-through of cell price moves?
  • On a budget constraint, what is the MWh-per-MW exchange rate?
  • How does the grid connection cost scale with MW and MWh? Treat the actual connection scope as fixed, power-linked or energy-linked rather than assuming one behavior.

Frequently Asked Questions

Why does $/kWh fall as duration rises?

Because balance-of-system cost is incurred per megawatt and is then divided by more megawatt-hours. The cells cost the same per kilowatt-hour at every duration — only the BOS term per kilowatt-hour shrinks.

Is a lower $/kWh always better?

Not by itself. Across different durations, a lower blended $/kWh can simply reflect the same power-related cost spread over more MWh. Normalize duration and scope, then compare the underlying energy and power cost components.

What does an extra hour of duration cost?

In the worked two-component model it is $180/kWh, but real incremental duration can include more than cells and need not be constant. Use the project-specific marginal energy-capacity cost.

How do you find a vendor's real cost structure?

Ask for comparable-scope quotes at two or more durations. Their differences give an implied marginal energy cost and power-cost intercept under a linear approximation. More quotes reveal whether the relationship is actually linear.

How should augmentation be costed?

From the expected augmentation scope: battery blocks / containers plus incremental BOS, installation, integration and any required upgrades. Do not use the original blended system $/kWh, but do not assume augmentation is pack-only either.

Do pack and BOS costs fall at the same rate?

Not necessarily. Battery energy costs and power/BOS costs have different drivers and should be forecast separately. Their divergence changes duration economics, but it is only one of several drivers of optimal duration.

How much of the cost sits outside the cells?

There is no durable universal percentage. It depends on duration, scope, geography, interconnection, tariffs, safety design and vendor architecture. Decompose the actual quote rather than applying a generic split.

Closing: Two Numbers, Not One

A common way to describe battery cost — blended $/kWh — makes comparisons across durations easy to misuse. It is not wrong as a description — a four-hour system genuinely does cost $275 per kilowatt-hour — but it collapses two independent quantities into one ratio, and the ratio moves with a variable that has nothing to do with procurement.

The fix is small: carry an energy-cost curve and a power-cost curve separately, with explicit scope. That gives the model a defensible marginal cost for another MWh and another MW instead of asking one blended $/kWh number to answer both questions.

It also makes vendor comparison honest. Multiple comparable-scope duration quotes let the model estimate the marginal cost curve instead of inheriting one blended ratio. If the implied slope changes by duration, keep the curve non-linear rather than forcing it into two constants.

That completes the sizing half of this series: what limits the build, what the megawatts are for, how many hours, and what those hours cost. The second half is about operating what you built. the next post starts it with the decision the whole asset exists to make: when to charge.

Sources: NREL 2024 ATB — utility-scale battery storage; 2–10 hour cost modeling and separate power / energy components · NREL 2025 Cost Projections for Utility-Scale Battery Storage · DOE — 2022 Grid Energy Storage Technology Cost and Performance Assessment · NREL SAM — battery replacement, dispatch and lifetime modeling · Dezzmond analysis; $180/kWh and $380/kW values are illustrative worked assumptions.