Green Hydrogen,
Black Numbers

How to use it

Run the tests in order. Each returns one of three results.

Pass

Evidenced, and the evidence would survive a hostile reading.

Conditional

Plausible, but resting on something not yet secured. Name the thing and the date it is expected.

Fail

Contradicted by the evidence, or unevidenced with no route to evidence.

A single fail stops the screen. Do not proceed to the next test on the assumption that a later strength compensates: it does not, because these are sequential dependencies rather than weighted criteria. A project carrying three conditionals is not two thirds financeable. It is a project with three unowned risks, and the correct action is to price the delay rather than to average the score.

Most projects I review stop at Test 2.

Test 1. Is there demand with no cheaper route?

Hydrogen competes with electrification, with efficiency, with material substitution, and with waiting until a regulation forces the issue. Identify the specific demand and establish whether a cheaper decarbonisation route is available to it. If one is, the project rests on subsidy rather than on the merits of the molecule, and it will last as long as the subsidy does.

The demand that survives is narrower than most published forecasts assume, and it is industrial. Existing hydrogen consumption in refining, ammonia and methanol. Primary steel where scrap supply cannot meet demand. High-temperature process heat with no electric route and no siting option for one. Shipping and aviation derivatives, where energy density and existing fleet constrain the alternatives. Long-duration storage where salt geology exists.

Evidence that satisfies this test

A named counterparty, their current fuel or feedstock cost, the delivered cost of the electric alternative at their site, and the reason that alternative is unavailable to them.

Test 2. What is the delivered cost at the customer gate?

Production cost is the figure usually quoted. Delivered cost adds compression, storage, transport, losses, purity conditioning and the cost of firming intermittent supply. On distributed projects those additions routinely exceed production cost, so the delivered figure is what the project competes on and the only one worth modelling.

Ask for the delivered figure with each line shown separately, then ask what it becomes at fifty per cent of design utilisation. A plant sized for a demand curve that arrives three years late runs well above its modelled cost per kilogram for the whole of that period, and the equity absorbs the difference.

Evidence that satisfies this test

A delivered cost stack with each component separately sourced, and the same stack rerun at half utilisation and at the sponsor's own low case for electricity or feedstock price.

Test 3. Who signs, for how long, and are they good for it?

Offtake quality determines financeability, and it is the most commonly overstated element in a hydrogen investment case. Memoranda of understanding and letters of intent are routinely presented as offtake. Neither commits a buyer to purchase, and a letter of intent from a counterparty who has not board-approved the capital expenditure for conversion at their own end commits them less still.

Examine contract length against asset life, the creditworthiness of the buyer, whether the price is indexed to something the buyer can hedge, and what the buyer's alternative is on the day the contract expires. A ten-year contract into a fifteen-year asset leaves a five-year merchant tail, and the equity return usually sits in that tail.

Evidence that satisfies this test

Executed contracts, buyer credit ratings, the indexation clause, and the buyer's own board paper approving their side of the conversion capital.

Test 4. What does the project need that it does not control?

Hydrogen projects depend unusually heavily on infrastructure decided by other parties: grid connections, pipeline conversion, storage caverns, port handling, and permitting regimes still being drafted. Every one of those is a schedule risk the project carries and cannot manage.

List each external dependency, name who owns the decision and when it is expected, then rebuild the model assuming each one slips by two years. Few projects survive that exercise, and a confident published timeline is a poor guide to which will. On the 800 MW Southern Green Hydrogen programme, the work that mattered most was establishing that deferring capital expenditure by five to seven years reduced cost and improved market timing. Sequencing is a decision, and it is usually a better one than acceleration.

Evidence that satisfies this test

A dependency register with named owners and expected decision dates, and a model rerun on a two-year slip for each.

Test 5. Does the safety case hold, and will the community accept it?

I chaired the Energy Institute working group that produced the first cross-industry, regulator-supported quantitative methodology for hydrogen safety distances, and the group that authored UK guidance on hydrogen refuelling systems co-located with petrol stations. That work makes one thing plain. Safety cases written by people who have specified hydrogen but never operated it tend to fail at the planning stage rather than at the technical one.

Consent is the test most diligence omits and the one that most often sets the schedule. Green infrastructure projects stall for years, and are sometimes killed outright, by community and regulatory conflict. A project with no community engagement plan and no separation-distance evidence pack is carrying an unpriced delay of unknown length.

Evidence that satisfies this test

A safety case authored or reviewed by operators, current versions of the applicable standards, a quantified separation-distance justification, a funded competence and training regime, and a community engagement record that predates the planning application.

Use and attribution

Green Hydrogen, Black Numbers is published under a Creative Commons Attribution 4.0 licence. Reproduce it, translate it, put it in your investment committee pack, teach from it. Credit it in this form and link back to this page.

Version 1.0, July 2026. Corrections and disagreements are welcome and will be credited in the next version.