Why do battery storage cost forecasts disagree?

How far apart are published battery storage costs?

Nearly fivefold, before anyone forecasts anything. BloombergNEF's December 2025 survey put stationary storage battery packs at $70/kWh on global average. NREL's June 2025 update starts its projections from $334/kWh for a complete four-hour system in the US. Both numbers are correct. They answer different questions, and a client slide that sets them side by side as rival estimates of one cost has already made the error.

The forward curves widen the gap rather than close it. NREL built its battery storage cost forecast from the published projections of 17 organizations, from utility resource plans to the EIA, Lazard and Wood Mackenzie. Its own 2030 range for the same four-hour system runs from $207/kWh in the low case to $354/kWh in the high case, in 2024 dollars. The high case sits 71% above the low one on an identical boundary, duration and starting point.

This means the first question about any battery cost figure is not whether it is right, but what it counts, where, and as of when.

What does each figure actually count?

Cost boundary explains most of the spread. Between a cell leaving the factory and an energized project there are four commercial layers, and each published series stops at a different one. Cell prices matter for manufacturing competitiveness and rarely reach a client deck; the confusion starts at the pack.

The layers above the battery are not small, and they do not fall at the pace of cells. In NREL's bottom-up model, battery cabinets account for $210 of the $334/kWh. The other $124 is the inverter, structural and electrical balance of system, installation labor, permitting and interconnection studies, sales tax, contingency, developer overhead and profit. Ember's all-in figure for projects outside China and the US splits about $75/kWh of core equipment from about $50/kWh of installation and grid connection, and grid connection alone moves that second number from $30/kWh where it is cheap to $100/kWh in extreme cases.

This means a pack price is a floor under a project budget, not an estimate of it.

Units add a second definitional trap. NREL reports four-hour systems in $/kWh but notes that inverters scale with power, not energy; on its own conversion, a $300/kWh four-hour battery costs $1,200/kW. The same hardware looks cheaper per kWh and more expensive per kW as duration lengthens, which is why NREL calls it critical to cite the duration with every capital cost. BloombergNEF's 2025 benchmark of $78/MWh for a four-hour battery project is a third unit again: a levelized cost per unit of energy delivered, not a capital cost, and it cannot sit in the same column as either.

None of this is peculiar to batteries. Valuation work has the same problem with NTM against TTM revenue multiples, where two correct numbers describe one company. The fix in both cases is to name what was counted before comparing the result.

Which differences are genuine?

Some of the spread survives normalization, and that part is the analysis.

Geography comes first. On BloombergNEF's 2025 survey, the average battery pack cost $84/kWh in China, while packs in North America and Europe cost 44% and 56% more, roughly $121/kWh and $131/kWh. BloombergNEF attributes the premium to higher local production costs. The boundary is identical; the price is not. For a US client, the China figure marks the frontier, not the quote.

Trade policy compounds it in the US. Wood Mackenzie estimated in June 2025 that tariffs could raise US utility-scale storage project costs by anywhere from 12% to more than 50%, depending on the scenario, because nearly all battery cells in US projects in 2024 came from China. BloombergNEF's base case the same month, a blanket 54% import tariff, lifted the US four-hour turnkey system cost by 30% to $266/kWh. This means a global benchmark cannot go into a US business case without an explicit tariff assumption written beside it.

Vintage is the third. NREL completed its 2025 update in January and February 2025 and states that it excludes the tariff changes that followed. Its $334/kWh starting point reflects US installations in Q1 2024, before stationary pack prices fell 45% during 2025, and NREL itself describes it as "higher than many current estimates." A figure can be accurate for its date and stale for the decision.

Method sits underneath all three. NREL builds its current cost from components, using a modeled market price for 2024 and a minimum sustainable price for 2035 and 2050. BloombergNEF and Ember report what buyers paid or were quoted. It is the same split as top-down and bottom-up sizing, with the same consequence: engineering models move smoothly and lag a glut, while observed prices move fast and carry whatever discounting oversupply produces. BloombergNEF names cell manufacturing overcapacity and intense competition as causes of the 2025 decline.

Why do battery storage cost forecasts diverge further out?

Forward curves inherit every difference above and add a fourth: what each forecaster assumes about learning, supply chains and policy. NREL's survey isolates that disagreement because it strips out the starting level. Each of the 17 organizations' projections is normalized to its own 2024 value; the low, mid and high cases are the minimum, median and maximum at 2026, 2035 and 2050, interpolated in between, then applied to a single $334/kWh starting cost.

Two readings matter for a client. The disagreement is about slope as much as level: in the high case, supply-chain constraints and tariffs push costs up 10% by 2026, after which they decline by less than 1% a year, while the low case falls 23% over the same two years. And the low and mid cases front-load their declines, because most publications expect larger reductions near term that slow over time. By 2050 the cases sit 8% to 68% below 2024.

The starting point carries as much weight as the slope. NREL notes that a higher or lower starting point shifts every projection up or down, and its own 2025 update landed close to the low case of its 2023 report mainly because current system costs were estimated lower. Start the same mid-case curve from Ember's $125/kWh instead of $334/kWh and the 2030 figure moves by more than NREL's entire low-to-high range.

This means two forecasts can agree on the rate of decline and still sit far apart on the 2030 number a client will quote.

Which battery storage cost forecast should a consultant use?

The one whose boundary matches the client's decision. Reconciling conflicting sources is usually a multi-hour job with no guaranteed resolution; most of this conflict, though, resolves the moment each figure's boundary is named. A procurement question and a project finance question can cite the same year and the same source and still need different figures.

1. State the boundary of every figure: pack, cabinet, turnkey system or installed project.
2. Convert each to one duration, and show $/kWh and $/kW together for the four-hour case.
3. Put every figure in one dollar year and record the date of the prices, not only the publication date.
4. Tag the market and the tariff regime: China, the US with tariffs, or elsewhere.
5. Bridge from pack to installed cost with explicit adders for power conversion, balance of system, labor, interconnection and soft costs.
6. Carry low, base and high cases as named scenarios, with the driver of each written beside it.

What survives those steps is the genuine disagreement, and it belongs on the slide as a range with its drivers named. A client can act on $207 to $354/kWh by 2030 depending on tariffs and learning; it cannot act on a number no source published. The discipline is the one used to reconcile market size estimates that don't match, and a range built this way is far easier to defend when the client challenges it, because every gap in it has a stated cause.

The published figures

Why the figures differ

• definitional BloombergNEF $108/kWh global pack and BloombergNEF $70/kWh stationary pack, both 2025 Same survey and year, different segment: the $108 covers packs across electric vehicles, storage and other uses; the $70 counts stationary storage packs only.
• definitional BloombergNEF $70/kWh stationary pack and Ember $125/kWh all-in project Different boundary: Ember adds installation and grid connection, about $50/kWh, to roughly $75/kWh of core equipment; a pack price excludes both.
• genuine BloombergNEF $84/kWh China pack and North American packs 44% higher Same boundary and survey; the gap is a real price difference that BloombergNEF attributes to higher local production costs.
• genuine Ember $125/kWh (outside China and the US, October 2025) and NREL $334/kWh (US, Q1 2024) Both price an installed project of four hours or more. The gap is real: a different market (the US, with tariffs), an earlier price date (Q1 2024, before the 2025 fall in pack prices) and a different method (a bottom-up model against observed prices). NREL also itemizes sales tax, contingency and developer profit.
• genuine NREL 2030 low case $207/kWh and high case $354/kWh Identical boundary, duration, dollar year and starting cost; the published forecasts disagree on learning, supply-chain constraints and tariffs.

Frequently asked

Keep reading

• How to reconcile market size estimates that don't match
• SaaS revenue multiples: why 8.5x and 3.3x can both be right