That study is the cleanest demonstration we have found of the problem this page exists to solve. The authors ran seven published tools on one project and attributed the dispersion to "differences in modeling depth, treatment of critical parameters (e.g., electrolyzer efficiency, capacity factor, storage, and inflation), and the tools' temporal resolution".[1] Not to differing views about the project. To differing views about what the calculation is.
There is no standard, and that is a finding
It is easy to assume that a metric this widely quoted must be defined somewhere. We went looking.
ISO has published a methodology standard for hydrogen — ISO/TS 19870:2023 — and it covers the determination of greenhouse gas emissions from production, conditioning and transport.[16] It does not address cost. CEN and CENELEC's hydrogen standardisation roadmap covers gas quality, purification and compression, grid connection, material compatibility and a greenhouse-gas footprint methodology. Levelised cost does not appear in it.[17]
What exists instead is one voluntary harmonisation proposal, from Agora Industry, which is careful and specific and worth reading.[2] It observes the problem in its own words: that "inconsistent use of the term LCOH, or inconsistent system boundaries, can lead to policy decisions being made based on incorrect assumptions", and that in high-level studies "a sufficiently transparent description of the LCOH calculation in terms of system boundaries and cost drivers is not given".[2]
And then there is the disclaimer on Agora's own accompanying calculator, which we quote in full because it is the most honest sentence in the entire literature:[3]
This Excel tool contains numerous simplifications and is not intended for use in projects. No liability is assumed for the applications of the tool nor for any errors in the tool. The tool does not replace a proper consulting project.
The publishers know. Lazard's hydrogen analysis carried its own: "This analysis is illustrative in nature and should not be considered as a benchmark for any given project."[12] The European Hydrogen Observatory says its defaults do "not show actual hydrogen production costs from operational water electrolysis plants in Europe".[8]
The figures get quoted as benchmarks anyway.
The four choices that move the number most
1. Lower or higher heating value
This one is mechanical, it happens before any cost assumption, and the sides are cleanly drawn.
The IEA reports efficiency on an LHV basis. IRENA did in 2020. NREL's H2A guide states that "unless otherwise specified, efficiencies are reported on an LHV basis". Lazard states a heating value of 33 kWh/kg, which is LHV. Agora recommends LHV explicitly, precisely "to avoid errors resulting from inconsistent use of higher heating value (HHV) and LHV".[2]
The United Kingdom is on the other side, and not just for reporting. BEIS states that "when quoting MW or MWh, they refer to MW H2 (HHV) and MWh H2 (HHV), unless otherwise stated".[9] The UK's hydrogen contracts are denominated the same way — the HAR1 strike price is expressed in £ per MWh H₂ on an HHV basis.[15] A convention that reaches the strike price of a real contract is not an academic preference.
IRENA changed sides between publications: LHV in 2020, HHV in 2022.[6][7] And the European Hydrogen Observatory's calculator manual — the EU's own tool — does not state anywhere which basis its kWh/kg consumption figures use.[8]
Figure 1
The same 65%-efficient electrolyser, on the two conventions
Electricity consumed per kilogram, when the same efficiency percentage is read against the lower or the higher heating value.
IRENA's own stated pairing.
GaugeH₂ arithmetic on the same percentage. Not a published figure.
Only the first bar is published. IRENA states 65% efficiency at 51.2 kWh/kg on an LHV basis. The second is our arithmetic: on a higher heating value of 39.4 kWh/kg — derived from BEIS's own conversion table, where 1 MWh H₂ (HHV) = 0.85 MWh H₂ (LHV) = 25.4 kg H₂ — the same 65% implies 60.6 kWh/kg. An 18% difference in the largest cost line, before a single price assumption is made.
Source: IRENA for the LHV figure; GaugeH₂ arithmetic for the HHV figure, on UK BEIS's published conversion · Data as of 2026-08-30
Show the data ▸Hide the data ▾
| Basis | Energy per kg (kWh) | At 65% efficiency (kWh/kg) |
|---|---|---|
| Lower heating value | 33.5 | 51.2 |
| Higher heating value | 39.4 | 60.6 |
| Difference | 17.6% | 18.4% |
An 18% difference in the electricity term, which is the largest line in most green hydrogen cost stacks, from a choice that a published figure often does not disclose.
2. Stack replacement
The divergence here is not a range of percentages. It is three incompatible accounting treatments.
BEIS annuitises it: "for electrolysis, variable OPEX refers to annuitised stack replacement costs", with stack costs at "60% (PEM), 50% (Alkaline), 60% (SOE) of plant CAPEX" replaced every eleven, nine and seven years.[9] The US DOE treats it as a discounted capital event: "electrolyzer replacement costs of 11% of the total installed capital costs are assumed to be incurred every 40,000 hours of operation".[10] Agora recommends folding it into CAPEX, and cites a literature range of 25% to 35% of system CAPEX.[2]
The IEA models no replacement event at all. It sets a stack lifetime of 50,000 hours and states that "the selected value (50000 hr) comes from the IEA's analysis of the optimum economic lifetime, considering degradation issues".[4] Degradation is absorbed into a lifetime choice rather than modelled.
Figure 2
Stack replacement: three published answers, and three different denominators
The replacement cost each methodology assumes, as a percentage of capital cost.
60% (PEM), 50% (Alkaline), 60% (SOE) of plant CAPEX.
25% from BNEF (2022), 35% from Agora's own 2021 work. Of system CAPEX.
Of total installed capital cost.
These three percentages are NOT comparable, and the chart is here to show that they are usually treated as though they were. BEIS's percentage is of plant CAPEX and is annuitised into variable OPEX. The DOE's is of total installed capital cost and is a discounted capital event every 40,000 hours. The 25–35% range Agora cites is of system CAPEX and it recommends folding it into CAPEX outright. Three denominators, three accounting treatments, one label. The IEA models no replacement at all: it sets a 50,000-hour stack life and states that the value comes from its analysis of the optimum economic lifetime considering degradation.
Source: UK BEIS; US DOE Hydrogen Program Record 24005; Agora Industry, citing BNEF and its own earlier work · Data as of 2026-08-30
Show the data ▸Hide the data ▾
| Source | Treatment | Trigger | % of capital cost | Of what |
|---|---|---|---|---|
| UK BEIS | Annuitised variable OPEX | Every 11 / 9 / 7 years | 60 / 50 / 60 | Plant CAPEX |
| Agora (citing BNEF and itself) | Folded into CAPEX | — | 25–35 | System CAPEX |
| US DOE | Discounted capital event | Every 40,000 hours | 11 | Installed capital cost |
| IEA | Not modelled | 50,000-hour stack life | Not stated | — |
Note what the chart cannot show, and what the note beneath it says: the three percentages have three different denominators. A figure of 11% of installed capital cost and a figure of 55% of plant CAPEX are not two views of the same quantity, and the gap between them is much smaller than the numbers suggest — but nobody can tell you how much smaller without both boundaries stated.
3. After-tax discounted cash flow, or pre-tax annuity
This is the deepest divide, and it is geographic.
The American models are after-tax DCF. NREL's H2A "estimates the hydrogen selling price (or levelized cost) using a standard discounted-cash-flow methodology and a specified after-tax internal rate of return".[11] It exposes fields for depreciation schedule and type, and computes a total tax rate. Lazard's hydrogen analysis operates the same way, with a corporate tax rate and a seven-year MACRS depreciation schedule.[13] The DOE's record states a nominal weighted average cost of capital of about 10%, built from a real return on equity, a debt-to-equity ratio and an interest rate.[10]
The European methods are pre-tax annuities. Agora's published formula applies a capital recovery factor to CAPEX, adds OPEX as a percentage of CAPEX, and adds the electricity term.[2] There is no corporate tax in it, no depreciation, no capital structure and no interest during construction. The European Hydrogen Observatory sets a default cost of capital of 6% and a default economic lifetime of 25 years, and does the same.[8] So do the IEA and both IRENA reports.
These are not the same calculation with different parameters. They are two different financial objects that share a name. One of them can answer "what price must this project sell at to earn its equity holders their required after-tax return". The other cannot, because it does not know what tax is.
4. Oxygen credits
Electrolysis produces eight kilograms of oxygen for every kilogram of hydrogen.[8][18] Whether that oxygen is worth anything in the LCOH is a question on which two European institutions have taken opposite public positions.
The IEA excludes it, in the same sentence across three publications: "for water electrolysis, water costs and possible revenues from oxygen sales have not been considered in the cost analysis".[4] Agora recommends exclusion, reasoning that by-product revenues and public funding alike "are strongly contingent on individual project setups".[2]
The Clean Hydrogen Partnership's calculator — the EU's own — puts oxygen revenues in the LCOH equation as an explicit term, alongside CAPEX, electricity, other OPEX, grid fees, taxes and subsidies.[8] It defaults to zero, so a user who never touches it gets an exclusion; a user who fills it in gets an inclusion. Both are labelled LCOH.
We found exactly one peer-reviewed quantification, and it is a net present value effect rather than a euro-per-kilogram one: an improvement of up to 13% in a specific wastewater-treatment configuration.[18] No major methodology publishes an oxygen credit in €/kg or $/kg.
The other eight, in one table
| Choice | Where publishers diverge |
|---|---|
| Capacity factor and full load hours | Expressed as full load hours by region (IEA), as an hourly capacity-factor profile (IRENA 2022), as "the 4,000 cheapest operating hours" (European Hydrogen Observatory), as a load factor of 98% for grid-connected alkaline and PEM (BEIS), or as a free variable with no prescribed value (Agora). |
| Degradation | No methodology examined applies it to output. Two apply it to efficiency: the EHO models a degradation rate per 1,000 hours and assumes replacement "resets the energy consumption back to its initial value"; Agora recommends an average specific energy requirement instead. The IEA absorbs it into the stack-life choice. IRENA 2020, Lazard, BEIS: not stated. |
| CAPEX boundary | Stated explicitly by the JRC, the European Hydrogen Observatory and Agora. Not stated by most others. Covered separately in our piece on the €/kW problem. |
| Construction period and interest during construction | Almost universally ignored. BEIS states a three-year build but not the financial treatment. NREL's H2A parameterises the construction period; whether interest is capitalised is not stated in the guide. The IEA, IRENA, Agora, the EHO, Lazard and the Hydrogen Council do not mention it. |
| Electricity price basis | A dedicated asset's LCOE (IEA); a fully off-grid system with the renewable CAPEX inside the boundary (IRENA 2022); the 4,000 cheapest wholesale hours, with grid fees and electricity taxes included only for grid-connected cases (EHO); industrial retail price including policy costs (BEIS); or a flat exogenous price (Lazard, IRENA 2020, DOE). Agora's recommendation is one line: include all charges. No source examined uses a stated PPA price. |
| Water and desalination | The IEA excludes water costs entirely, and says so. IRENA 2022 is the only source with full figures — 9 litres per kilogram, desalination at USD 3/m³, an LCOH increment of 3.8% — and then does not compute it, "given the marginal contribution". Agora makes it conditional on whether a plant has to be built. Lazard states a cost without a consumption, and changed it by a factor of four between editions. |
| Subsidies and tax | Lazard and the DOE state exclusion of production incentives explicitly. Agora recommends excluding public funding because it is "very much dependent on the country and the region". The EHO models three support mechanisms, defaulted to zero. Corporate tax and depreciation do not enter the IEA, IRENA, Agora, EHO or Hydrogen Council methods at all. |
| Plant life, currency and base year | Plant life: 25 years (IRENA 2022, EHO), 30 years (BEIS, DOE), 20 rising to 25 across Lazard editions — and for electrolysis the IEA annexes state no plant life at all, only a stack life in hours. Base year: BEIS is unambiguous ("all estimates are in 2020 real values"); the DOE states a 2022 dollar basis; Lazard states none in any edition examined; Agora states none. NREL's H2A is the only model that exposes the basis-year to reference-year conversion as an explicit mechanism. |
The checklist
If you are reading a published LCOH, or writing one, these are the twelve questions. A figure that cannot answer them is not wrong — it is unlocated.
- Lower or higher heating value?
- What is the discount rate, is it a WACC or a hurdle rate, and is it real or nominal?
- Pre-tax annuity or after-tax discounted cash flow? If after-tax, what tax rate and what depreciation schedule?
- Where is the CAPEX boundary — stack, system, or installed project?
- How is stack replacement treated: annuitised, capitalised and discounted, folded into CAPEX, or not modelled?
- Is degradation applied, and to output or to efficiency?
- What capacity factor or full load hours, and derived how?
- What is the electricity price, and does it include network charges, levies and taxes?
- Are by-product oxygen and heat credits in or out?
- Is water in the boundary, and is desalination costed?
- Is the figure pre- or post-subsidy?
- What plant life, what currency, what base year, and are the terms real?
And one that only one publisher answers: at what pressure and purity is the kilogram delivered? Agora recommends 30 bar and a purity of 5.0 — 99.999% — and states that if a system operates below the reference pressure, the compression cost belongs in the LCOH.[2] The DOE states 30 bar without a purity.[10] The IEA, IRENA, the European Hydrogen Observatory, Lazard and the Hydrogen Council state neither. A kilogram at 30 bar and 99.999% and a kilogram at atmospheric pressure and 99.9% are different products, and the cost of getting from one to the other is not small.
Why we are publishing this rather than a number
There is no maintained, boundary-explicit cost reference for green hydrogen. IRENA's benchmark is from 2020. Lazard published a levelised cost of hydrogen up to its 2024 edition and it does not appear in the 2026 one.[13][14] Agora's calculator is version 1.1 from July 2023 and tells you not to use it in a project.[3]
We are not going to close that gap by adding a tenth number to the nine that already disagree. What is missing is not another figure. It is the discipline of saying, next to any figure, which twelve choices produced it.
This page is versioned and dated, and it will be corrected in place with the change recorded below. If you find a methodology we have characterised wrongly, that is a correction we want.