What a battery cost per kWh actually includes

"Battery storage costs about X dollars per kilowatt-hour" is one of the most repeated sentences in energy writing. It is repeated because it sounds like a price, the way gasoline has a price. It is not a price. It is the output of a specific model of a specific machine, and the National Renewable Energy Laboratory, which produces the most widely cited version of it, is unusually clear about that if you open the document.

The document is Cost Projections for Utility-Scale Battery Storage: 2025 Update, NREL/TP-6A40-93281, by Wesley Cole, Vignesh Ramasamy and Merve Turan, published June 2025. It is 26 pages, free, and the parts that matter most are in the methods section and one appendix table.

Qualifier one: a duration is baked in

The report's headline figures describe a four-hour lithium-ion system, and only that:

We only used projections for 4-hour lithium-ion storage systems for the mainland United States.

This matters more than it sounds, because the unit itself is a convention rather than a physical property. NREL says so directly:

We report our cost projections as a total system overnight capital cost expressed in units of $/kWh. However, not all components of the battery system cost scale directly with the rated energy capacity (i.e., kWh) of the system.

The inverter, for instance, scales with rated power in kilowatts, not with stored energy. Developer costs scale with both. The authors explain that they use dollars per kilowatt-hour anyway "because that is the most common way that battery system costs have been expressed in published material to date", which is a candid admission that the unit was chosen for convention rather than for fit.

The practical consequence is that a per-kilowatt-hour figure is meaningless without the duration attached, and the conversion is trivial once you know it. The report gives the arithmetic itself: "a $300/kWh, 4-hour battery would have a power capacity cost of $1200/kW." Change the duration and the same hardware produces a different headline number. A two-hour system and an eight-hour system quoted in the same units are not comparable, which is the same power-versus-energy confusion I wrote about in Report 095, arriving here in the cost column instead of the capacity column.

Qualifier two: a scope, and one notable exclusion

The bottom-up model builds the cost from ten subcomponents, four of them hardware and six of them soft costs. The hardware is battery cabinets (LFP cells, racks, fire suppression, thermal management, battery management), the bidirectional inverter, structural balance of system, and electrical balance of system. The soft costs are installation labor, permitting and interconnection, sales tax at an assumed 5.8 percent national average, contingency, developer overhead, and profit at a 5 percent markup on the hardware items.

That is a wide fence. Structural balance of system alone includes access roads, security fencing, geotechnical investigation, grading, compaction and erosion control. Electrical balance of system includes the transformer, switchgear, SCADA, substation equipment and transmission line.

But note what sits just outside the fence, in the permitting and interconnection line:

Cost of interconnection upgrades are not included in these costs.

Interconnection studies are in. The upgrades those studies require are out. Given that network upgrade costs are one of the main reasons projects die in the queue, a subject I covered in Report 126 three days ago, that exclusion sits directly on top of one of the largest and most variable real-world costs of building a battery. It is disclosed plainly. It just does not survive the trip into a headline.

The other thing this scope tells you is that these numbers describe a utility-scale project with a substation, a transmission line and a developer. A home battery is a different scope with different line items, so a residential quote and this figure are not two measurements of the same thing.

Qualifier three: the range is not a confidence interval

This is the one I would most like people to take away, because the low, mid and high cases get read as an uncertainty band around a best estimate. They are not that. They are order statistics over other people's published forecasts:

We defined our low, mid, and high projections as the minimum, median, and maximum point, respectively in 2026, 2035, and 2050.

The projections are collected from a survey of recent publications, normalized so that every one starts at 1.00 in 2024, and then the minimum, median and maximum of that collection become the three cases. So "the low case" means the single most optimistic cost decline anybody in the surveyed literature published. It does not mean a well-characterized favorable scenario, and it carries no probability.

The spread this produces is genuinely wide, and the report's own summary does not hide it. Near term: "some projections show costs increasing up to 10% while others show substantial declines of up to 23%." By 2035 the range runs from a 2 percent capital cost increase to a 56 percent reduction, and by 2050 from 8 to 68 percent reductions.

A field where credible published forecasts disagree about the sign of the change over the next year is a field where quoting a single number without its case label is not much better than guessing.

The numbers, and where they disagree with themselves

The bottom-up model puts the 2024 starting point at $334/kWh for a complete four-hour system, and everything else is that number scaled by the normalized trajectories.

Now the part I did not expect to find. There are three different sets of figures in circulation for this one report, and they do not agree.

The report's executive summary gives 2035 costs of $147/kWh, $243/kWh and $339/kWh for the low, mid and high cases, and 2050 costs of $108/kWh, $178/kWh and $307/kWh, in 2024 dollars. Appendix Table 2, which is the underlying data, gives exactly the same values in its 2035 and 2050 rows. Two independent places in the document agree, so those are the report's numbers.

The body text in Section 3 does not. It reads "$147/kWh, $234/kWh, and $339/kWh in 2035", putting the mid case at 234 where the summary and the table both say 243. Two transposed digits, and 234 happens to be the mid-case value for 2037 in the same table, which is the kind of coincidence that makes a typo hard to spot.

The larger discrepancy is external. The Department of Energy's OSTI catalog record for this report carries an abstract quoting "$152/kWh, $247/kWh, and $349/kWh in 2035 and $111/kWh, $184/kWh, and $333/kWh in 2050". Not one of those six figures appears anywhere in the published PDF. I searched the extracted text for each of them individually to be sure.

I do not know why, and I want to be careful here. The most ordinary explanation is a version mismatch, an abstract written against an earlier draft and never updated when the final numbers moved. I did not contact NREL or OSTI, and I am not asserting which record is stale. What I can say is procedural and useful: the abstract that search engines and aggregators surface for this report does not match the report's own table, so if you need the number, take it from Appendix Table 2 and cite the page.

What is frozen into the starting point

One more caveat that has aged into significance. The report states:

This work was completed in January and February 2025. It does not include impacts from changes in tariffs that have occurred since that time.

Read carefully, that is not a claim that the model is tariff-free. It is a claim that the model is frozen at early-2025 trade policy. The cost breakdown in Table 3 shows the assumption explicitly: the lithium-ion cells are "LFP cells from China" with the noted distortion "301 tariff and 3.4% general duty", and a passthrough line records "Half of 45X used to decrease price to improve competitiveness", 45X being the U.S. advanced manufacturing production credit.

So trade policy and domestic manufacturing incentives are both inside the $334/kWh starting point, at their early-2025 settings. Anyone quoting this report in late 2026 is quoting a number with a policy snapshot embedded in it, and the report told them so on page iv.

Why I care about this one

I help design the AI battery-cycling systems for a veteran-owned (HUBZone) energy-storage integrator. I do not own that company and earn nothing from this link. Full policy here.

The reason cost scope is not an academic issue on that side of the work is that the software has to know which cost it is optimizing against. A cycling strategy that maximizes throughput is trading capacity fade against revenue, and the exchange rate between those two is a replacement cost. Use a headline per-kilowatt-hour figure that excludes interconnection, assumes a four-hour system and encodes last year's tariffs, and you have built a controller that is very precisely optimizing the wrong quantity. My own background is chemistry and materials, so I came to this expecting the interesting uncertainty to be in the cells. Much more of it is in the fence line.

What I could not confirm

I read the report; I did not audit it. I did not reproduce the bottom-up cost model, verify the PVSCM framework it adapts, check the literature survey against the underlying publications, or inspect any accompanying dataset. Every figure quoted here is quoted from the report's own text and Appendix Table 2, and no value has been read off a chart.

On the discrepancies: I have established what the PDF says and what the OSTI record says, and that the OSTI figures do not appear in the PDF. I have not established which is authoritative or why they differ, and I did not contact either institution. Treat the version-mismatch explanation as my most likely reading, not as a finding. If NREL issues a revised version or OSTI updates the record, this report should be corrected in place.

I obtained the PDF through a mirror of the NREL document server rather than nrel.gov directly, and extracted its text locally. The file self-identifies as NREL/TP-6A40-93281, June 2025, with the correct authors and NREL's own suggested citation, and the DOI and publication date match the OSTI record, so I am confident it is the report. I have not compared it byte for byte against the copy served from nrel.gov.

These are utility-scale projections for the mainland United States and four-hour lithium-ion systems only. Nothing here supports a claim about residential battery pricing, other durations, other chemistries, or other countries. I deliberately quote no residential price figures, because the ones I found came from vendor and content-marketing pages with no primary source behind them.

None of this is my research. My published work is in microwave spectroscopy. The projections, the model and the caveats all belong to the NREL authors, who documented them clearly enough that this entire report is really just an argument for reading their methods section.

The signal

A dollars-per-kilowatt-hour battery figure is not a price, it is a result. It arrives with a duration, a scope and a derivation attached, and the transformation from "result" to "price" happens when those three get dropped, which is usually at the first citation.

So when you meet one of these numbers, ask the three questions. What duration does it assume, since the unit does not carry one. What is inside the fence and what is outside it, because interconnection upgrades are outside this one. And is the range a confidence interval or the spread of a literature survey, because here it is the second and that means the low case is somebody's most optimistic paper rather than a good outcome.

Then check that the number you are about to quote is actually in the report. That last one I did not expect to have to say, and I would not have found it if I had stopped at the abstract.