There is a way of reading a cancellation announcement that makes it useful, and a way that makes it entertainment. The entertaining way is to collect them. The useful way is to ask, each time, whether the thing that killed the project was in anyone's model — and to notice how often the answer is no.
The largest cause is a decision, not a number
Rystad's breakdown of cancellations puts strategy pivot at the top, and the gap between the two percentages is the finding. Forty-eight per cent of the capacity and twenty-four per cent of the projects means the projects abandoned for strategic reasons were, on average, twice the size of the others.[1]
That inverts the intuitive picture. A small project dies because the numbers do not work. A very large project dies because the company that owns it decided to be a different company.
bp is the clearest case. Its exit from the Australian Renewable Energy Hub gave up a 63.7% stake in a 26 GW wind and solar development with a plan for 10 million tonnes a year of green ammonia — and the stated reason was the "recent strategy reset, which will see BP grow its upstream oil and gas business".[7] Nothing about that project's economics changed on the day of that announcement. The shareholder base's patience did.
There is no cell in a techno-economic model for that.
What the stated reasons actually were
The public record is unusually good here, because listed companies have to explain themselves. Here is what four of them said.
| Sponsor | What was given up | What kind of number this is | Stated reason |
|---|---|---|---|
| Air Products | Three US projects, incl. Massena NY liquid green hydrogen | A pre-tax charge of up to USD 3.1bn | For Massena: "regulatory developments rendering existing hydroelectric power supply ineligible for the Clean Hydrogen Production Tax Credit (45V)" and "slower than expected development of a hydrogen mobility market". For the others: "challenging commercial aspects" and "unfavorable project economics"[4] |
| ArcelorMittal | DRI-EAF at Bremen and Eisenhüttenstadt | An abandoned investment plan of €1.3bn | "cannot proceed with these investments, given the realities of the market and the economics of low-carbon emissions steelmaking" — and, decisively, "even with the financial support, the business case … is not strong enough"[5] |
| Fortescue | Arizona Hydrogen and PEM50 Gladstone — both already past FID | A write-down of USD 150m pre-tax | US energy policy shift and a strategic redirection in Australia[6] |
| Cleveland-Cliffs | Hydrogen-based DRI at Middletown; returned a USD 500m DOE grant and will spend USD 1bn relining a coal blast furnace instead | A returned grant, plus a redirected capital plan | "it's clear by now that we will not have availability of hydrogen"[9] |
Those four numbers are deliberately not added together, and the middle column is why. A pre-tax accounting charge, an abandoned investment plan, a write-down and a returned grant are four different objects. Summing them would produce a headline with no defensible meaning — which is, incidentally, how most "billions lost in hydrogen" figures are built.
And ArcelorMittal's sentence is the one to keep
Of everything in that table, one clause should worry a developer more than the write-downs: "even with the financial support, the business case … is not strong enough."[5]
That is a company saying that a subsidised project still did not clear. Most hydrogen models are built to answer the question "how much support does this need". ArcelorMittal is answering a different one: whether the supported project is worth doing at all, against everything else the capital could do. Those are not the same test, and only the second one is the one that gets taken.
FID is not the finish line
The most durable assumption in hydrogen project tracking is that final investment decision is the point of no return. It is not.
Fortescue's USD 150m write-down covered two projects that had already taken FID.[6] Stegra reached financial close in January 2024 on the largest hydrogen-based project financing in Europe, and by 2026 needed a €1.4 billion rescue round, with second-lien lenders converting debt to equity and the project timeline placed under review.[11] Liquid Wind was declared bankrupt in May 2026, six days after its founding chief executive made optimistic public statements.[10]
And HyScale100 — 2.1 GW at the Heide refinery, with €900 million of public funds allocated and €194 million earmarked by Schleswig-Holstein that was never disbursed — was cancelled in August 2026 for "lack of planning certainty".[8] Money allocated is not money spent, and a project can die between the two.
The shape of the pipeline is not what the headlines imply
Figure 1
And the cancelled projects are not the interesting part
Every hydrogen project registered in Europe since 2020, by where it ended up.
- Operational
- Neither built nor cancelled
- Cancelled
Cancellation is the visible outcome and it is the smaller one. The 79% in the middle is the real story: projects that are neither dead nor built, most of them sitting in feasibility, some formally paused. A pipeline is not a queue — it is a distribution, and most of the mass is in a state that no announcement will ever mark.
Source: Energy Industries Council, via Innovation News Network · Data as of 2026-08
Show the data ▸Hide the data ▾
| Status | Projects | Share |
|---|---|---|
| Operational | 59 | 9.5% |
| Neither built nor cancelled | 491 | 78.7% |
| Cancelled | 74 | 11.9% |
| Total registered since 2020 | 624 | 100% |
Six hundred and twenty-four European projects registered since 2020. Fifty-nine operational. Seventy-four cancelled.[3]
The cancellations get written about because they are events. The 491 in the middle are not events, and they are 79% of the pipeline. Most of them sit in feasibility; some are formally paused. A project in that state produces no announcement, generates no headline, and quietly consumes the years in which it might have been built.
This is why cancellation counts understate attrition so badly. The Hydrogen Council counted around 50 public cancellations over eighteen months and put it at roughly 3% of the pipeline.[12] That is an accurate count of a category that is not where most projects go to die. They go to the middle bar, and stay there.
The IEA's measurement of the same phenomenon is cleaner because it looks at what moves rather than what is announced: new final investment decisions fell in 2025 for the first time in the sector's history.[13]
When the numbers did do the killing
Figure 2
What killed European capacity, by the developers' own stated reasons
Share of cancelled or stalled capacity, across 23 projects in 11 European countries.
32% of projects.
18% of projects — so the financing failures were disproportionately large projects.
11% of projects.
A December 2024 snapshot and a single source: treat it as historical reference, not as a picture of 2026. Note also that the capacity-weighted and count-weighted shares diverge — financing failure took 29% of the capacity but only 18% of the projects, which means the projects that died for lack of money were the big ones. The remaining share is spread across causes the source does not group.
Source: Westwood Global Energy Group · Data as of 2024-12
Show the data ▸Hide the data ▾
| Stated reason | % of capacity | % of projects |
|---|---|---|
| High costs | 40 | 32 |
| Failure to obtain financing | 29 | 18 |
| Lack of demand | 9 | 11 |
| Total pipeline affected | 29.2 GW = 20.3% of the European pipeline | 23 projects |
For the projects where an economic cause is stated, cost dominates and financing is close behind. The detail worth stopping on is that the two shares diverge: financing failure accounted for 29% of the lost capacity but only 18% of the projects.[2] The projects that died for lack of money were disproportionately the large ones — which is the same signal as the strategy-pivot finding, arriving from a different direction. Scale does not protect a project. It changes what kills it.
Demand is the smallest stated cause in that breakdown, at 9% of capacity, and that number should be read carefully rather than taken as good news. A project that never signs an offtake does not usually announce that demand killed it. It stalls, and joins the 491.
What to put in the model instead of a cancellation probability
None of this supports a probability of cancellation. Attaching one would be pretending that a sponsor's strategic reset is a random variable with a knowable distribution, and it is not.
What can be carried is the exposure:
- Name the sponsor's commitment concentration. A project that is one of thirty in a portfolio is a different risk from one that is the company. bp went from roughly thirty hydrogen projects to a target of five to seven, and the survivors were not chosen on their individual merits.[7]
- Record the date of the sponsor's last strategy reset, and treat a recent one as information. A company that reset last quarter has already made its cut. One that has not is carrying the risk that it will.
- Separate "needs support" from "is worth doing". ArcelorMittal's sentence is the test: model the case with the subsidy and ask whether the sponsor would still choose it over its alternative uses of capital.
- Do not treat FID as terminal. Track post-FID projects for cost escalation and schedule, because that is where Fortescue's, Stegra's and Liquid Wind's problems appeared.
- Count the stalls, not the cancellations. A project that has not moved in eighteen months is a worse signal than one that was formally cancelled, because the cancelled one at least produced a decision.
- Distinguish allocated from disbursed. HyScale100 had €900 million allocated and died anyway. Public money in a press release is not public money in a bank account.
The uncomfortable conclusion is that the best-modelled project in the sector can be cancelled by a board meeting it was never party to. That is not an argument against modelling. It is an argument for being precise about which risks the model carries and which it does not — and for saying so out loud, in the document, where the committee can see it.