A developer, a lender and a journalist can each hold a defensible figure for the capital cost of the same 100 MW alkaline electrolyser, and be a factor of eight apart. Not because one of them is wrong, and not because the market is volatile. Because the three of them are measuring to different fence lines, and none of the three published numbers says where the fence is.
This is the single most common way a hydrogen cost figure misleads, and it is almost never a case of bad faith. It is a missing sentence.
Three boundaries, one machine
The European Commission's Joint Research Centre publishes electrolyser cost at three explicit boundaries in the same report. For alkaline, in the same reference year, for the same technology:[1]
Figure 1
The same alkaline electrolyser, priced at three boundaries
Each figure is correct. They answer different questions.
The electrochemical core. What a stack tender prices.
Stack plus balance of plant: rectifiers, water treatment, gas separation, cooling.
What it costs to have a plant. Civils, grid connection, EPC, owner's cost, contingency.
Reported in euros per kilowatt of electrolyser capacity. Bars are anchored at zero; the lighter extension on the third bar is its published range. The installed-project figure is for 100–200 MWel in a European context.
Source: European Commission, JRC — Clean Energy Technology Observatory, Water Electrolysis 2025 Status Report · Data as of 2025
Show the data ▸Hide the data ▾
| Boundary | Alkaline (€/kW) | PEM (€/kW) |
|---|---|---|
| Stack only | 323 | 563 |
| Full system | 1,016 | 1,209 |
| Installed project, 100–200 MWel | 2,630–3,050 | not reported |
Between the first bar and the third there is a factor of 8.1 to 9.4. Nothing about the electrolyser changed. What changed is how much of the project came inside the fence.
The three answer genuinely different questions, and each is the right number for someone:
- Stack only is what a stack supplier quotes and what a manufacturing cost curve tracks. It is the number that falls when a factory learns.
- Full system is what arrives on site as equipment: stack plus rectifiers, water treatment, gas separation, cooling, controls. It is the right number for comparing vendors.
- Installed project is what a lender funds and what an LCOH model needs. Civils, buildings, grid connection, land, permitting, EPC management, owner's cost and contingency all sit here.
The European Hydrogen Observatory states its own scope in the same spirit: its published CAPEX covers "stack, balance of plant (BoP), other utilities, and other CAPEX", where other CAPEX runs to "land and grid fees, insurance, permitting, feasibility study, contingency, EPC management", on advanced-stage 100 MWel projects.[6] That sentence is not boilerplate. It is the difference between a figure you can put in a model and a figure you cannot.
Where the money actually goes
If the stack were most of the cost, tracking stack prices would be tracking project cost. It is not, and three independent sources put the share in the same territory.
The Institute for Sustainable Process Technology ran a detailed engineering design of a 1 GW plant. Stacks came out at 13% of total installed cost for alkaline and 23% for PEM; indirect costs, owner's cost and contingency together took 44% and 38%.[5] The World Bank puts the stack at "only 20 to 50 percent of direct CAPEX in modern large-scale PEM and ALK electrolyzer projects", with installation and EPC "often the largest cost contributors, accounting for 40-50 percent of total CAPEX".[4] The IEA, looking at Chinese equipment, reports that the stack "only represents 15-20% of the total investment" and that around 60% of the investment required is engineering, procurement and construction.[2]
Figure 2
What a built plant is made of
Detailed engineering design of a 1 GW plant. The stack is the smallest block in both technologies.
- Stacks
- Balance of plant and other direct
- Indirect, owner's cost, contingency
A front-end engineering study, not a built plant. Used for cost STRUCTURE — the share each element takes — which is what a detailed design establishes. The 2020 absolute values are deliberately not used, and the shares are corroborated by the World Bank (stack 20–50% of direct CAPEX) and the IEA (stack 15–20% of total investment; EPC around 60%).
Source: Institute for Sustainable Process Technology — Gigawatt Green Hydrogen Plant · Data as of 2020
Show the data ▸Hide the data ▾
| Element | Alkaline (% of TIC) | PEM (% of TIC) |
|---|---|---|
| Stacks | 13 | 23 |
| Balance of plant and other direct cost | 43 | 39 |
| Indirect, owner's cost and contingency | 44 | 38 |
The three sources disagree on the exact share, which is what independent sources do. They agree on the shape: the stack is a minority of the cost, and the largest single block is the work of building the thing.
What this does to the China story
The most repeated cost claim in the sector is that Chinese electrolysers are several times cheaper. At stack level, that is true and well evidenced: the IEA reports Chinese alkaline stacks at roughly a quarter the price of Western equivalents, and attributes it to fierce domestic competition driving down tender prices.[3]
Then the boundary moves, and the claim does not survive it.
Figure 3
Installed cost: the Chinese advantage narrows to an overlap
Ranges published by the IEA for installed systems, by equipment origin and destination.
Installed cost basis, reported in US dollars per kilowatt. Not converted to euros: an FX assumption is exactly the kind of undeclared input this piece is about.
Source: IEA — Global Hydrogen Review 2025, Executive Summary · Data as of 2025
Show the data ▸Hide the data ▾
| Case | Installed cost (USD/kW) |
|---|---|
| Chinese electrolyser, made and installed in China | 600–1,200 |
| Chinese electrolyser, installed outside China | 1,500–2,400 |
| Non-Chinese electrolyser, installed outside China | 2,000–2,600 |
The lower two ranges overlap across a 900 USD/kW band. The IEA's own conclusion is that the cost advantage of cheap Chinese electrolysers "reduces significantly when systems are installed outside of the country".[2]
Both facts are true at once, and they are not in tension — they are the same fact seen at two boundaries. A cheap stack lands inside a project whose civils, grid connection, permitting, EPC management and contingency are priced in the country where the plant is built, and those are the majority of the cost. The IEA adds a second effect that does not show up in a price at all: imported equipment "can drive up operational costs" where it has to be adapted to local standards or underperforms.[2]
What a 50% cut in stack price actually buys
This is the arithmetic that follows from Figure 2, and it is the reason stack-price headlines mislead about project economics. Halve the stack price, hold everything else, and see what moves at project level.
Figure 4
Halve the stack price. Total installed cost falls by 6.5% to 12%.
Fall in total installed CAPEX from a 50% reduction in stack cost, computed on each source's published stack share.
Arithmetic, not a model: the fall in total installed cost equals half the stack's share of it. The World Bank share is of DIRECT CAPEX rather than total, so its effect on a full installed project is smaller than the range implies — it is included as the upper bound of the published shares, not as a like-for-like case.
Source: GaugeH₂ calculation on stack shares published by ISPT, the IEA and the World Bank · Data as of 2026-08-30
Show the data ▸Hide the data ▾
| Published stack share | Share of cost | Fall in total CAPEX |
|---|---|---|
| ISPT, alkaline 1 GW | 13% | 6.5% |
| IEA, Chinese equipment | 15–20% | 7.5–10.0% |
| ISPT, PEM 1 GW | 23% | 11.5% |
| World Bank, low end of range (direct CAPEX) | 20% | 10.0% |
The grey bar is the implicit assumption in a headline that treats stack price as project cost. The distance between it and the coloured bars is the size of the error.
This is not an argument that stack cost does not matter. It is an argument about which lever moves which number. Stack price is the right thing to track if the question is whether manufacturing is learning. It is close to the wrong thing to track if the question is whether a project clears a hurdle rate.
Why the 2020 projections missed, in one sentence
IRENA's 2020 study — still the most-cited cost anchor in the sector, six years on — put alkaline systems above 10 MW at 500–1,000 USD/kW and projected below 200 USD/kW in the long run.[7] By 2024 the IEA was reporting non-Chinese installed systems at 2,000–2,600 USD/kW.[2]
Part of that gap is boundary: IRENA's figure is a system, the IEA's is an installed project, and by now you can see how much of the distance that alone explains. Part of it is real: the IEA attributes the rise in electrolyser cost to inflation and slower-than-expected deployment.[2]
What is worth noticing is that IRENA said so at the time. Its projection rested on an assumed learning rate of 16–21%, and the report flagged in its own text that this was "significantly lower than the 36% learning rates experienced over the last 10 years for PV".[7] A learning rate is a function of cumulative volume. The volume did not arrive.
Reading a €/kW figure in one minute
Five questions. If a published figure cannot answer the first one, the other four do not arise.
- Where is the fence? Stack, system, or installed project. If the source does not say, treat the figure as a stack or system number — that is what is usually being quoted — and expect the installed cost to be two to three times higher.
- Is the renewable generation inside or outside? A dedicated wind or solar asset can exceed the electrolyser. Most published electrolyser CAPEX excludes it; most project CAPEX includes it.
- What plant size? Per-kW cost falls with scale. A 10 MW figure and a 200 MW figure are not comparable, and the JRC range above is explicitly 100–200 MWel.
- Are owner's cost and contingency in? In the ISPT design these plus indirect costs were 38–44% of the total. A figure that omits them is not describing a fundable project.
- What year, and in what currency, unconverted? Costs moved materially between 2022 and 2026, and a figure converted at an unstated exchange rate has an assumption inside it that nobody declared.
The reason to insist on this is not pedantry. It is that a hydrogen project is defended in front of people whose job is to find the assumption you did not state. A number whose boundary is written down survives that meeting. A number that is merely low does not.