Economics

What LCOH actually tells you

8 min read · Updated 2026-09-28

Levelised cost of hydrogen is the number nearly every molecular-energy decision turns on, and it is the number most often quoted with no indication of how it was produced. Two competent analysts can build LCOH figures for the same project a factor of two apart without either making a mistake, purely through assumption choice.

That does not make LCOH useless. It makes a bare LCOH figure useless.

What the calculation is

LCOH is the total lifetime cost of a production system, discounted to present value, divided by the hydrogen it delivers over that life. It converts an awkward mix of upfront capital and ongoing operating cost into a single figure per kilogram, so options can be ranked consistently.

The inputs are unremarkable: capital cost, cost of capital, plant life, electricity price, electricity consumption per kilogram, load factor, stack replacement, fixed and variable operating cost, water. The sensitivity of the output to a handful of them is where all the disagreement lives.

The four inputs that decide the answer

Electricity price

The dominant term for electrolytic hydrogen, typically the majority of total cost. It also admits the widest range of defensible assumptions: a power purchase agreement price, a merchant market average, a curtailment-only price, or a levelised cost of dedicated generation. These are all legitimate and they are not close to each other.

Worse, electricity price and load factor are not independent. Cheap power is often cheap because it is intermittent, and taking only the cheapest hours lowers the price and the load factor at once — which pull the answer in opposite directions.

Load factor

The second dominant term, and the one most often assumed rather than derived. A plant running at 90% utilisation spreads its capital over three times the output of one running at 30%. Since electrolysers are capital-intensive, that moves LCOH substantially.

The honest way to set it is from an hourly model of the actual renewable resource and the actual grid arrangement. A great many published figures use a round number instead, and round numbers tend to be optimistic.

Cost of capital

An input that is frequently invisible in the presentation and enormous in the result. A capital-intensive project’s levelised cost is highly sensitive to the discount rate, and the appropriate rate for a first-of-a-kind project with no contracted offtake is not the rate used for a contracted utility asset. Models built with an infrastructure discount rate and a merchant risk profile flatter themselves considerably.

Capital cost and the boundary it covers

Stack cost, system cost and installed cost including balance of plant, civil works, grid connection and contingency are different numbers, sometimes by a factor of two or more. Vendor figures usually quote the narrowest boundary that is defensible. A model that imports a quoted capex without establishing what it includes has an error built in before it starts.

The boundary problem, generally

The single most consequential decision in a hydrogen cost model is where you draw the system boundary, and it is made before any arithmetic happens.

Production cost at the plant gate is the figure almost always quoted. It is also almost never the commercially relevant one. What a buyer pays is a delivered, certified cost at their boundary — which adds compression, storage, transport, any conversion and reconversion losses, and the cost of meeting a certification scheme.

Those additions are not marginal. For hydrogen in particular, the storage and transport penalty is severe enough that a project with a lower plant-gate cost can lose decisively on delivered cost. Comparing two projects on plant-gate LCOH can therefore rank them backwards.

The same applies to derivatives. LCOA for ammonia and LCOM for methanol inherit every hydrogen assumption and add synthesis capital, energy and — for methanol — a CO2 cost whose price and accounting status may matter more than its quantity.

How to read somebody else’s number

Four questions resolve most of the ambiguity:

1. What is the boundary? Plant gate, delivered, or delivered and certified.
2. What electricity price and load factor, and are they consistent with each other?
3. What discount rate, and does it match the project’s actual risk?
4. What does the capex include? Stack, system, or installed with grid and contingency.

If a figure cannot be traced to answers for those four, it is a marketing number. That includes figures in otherwise serious publications, and it includes optimistic and pessimistic ones equally.

How to build one worth having

Build it up from components rather than calibrating backwards to a number someone else published — benchmarking to a target reproduces that target’s assumptions invisibly. State every input with its source and mark the ones that are judgements. Model the resource hourly rather than assuming a load factor. Then find the sensitivities that dominate, because they tell you where reducing uncertainty is actually worth effort.

And produce a range with named drivers rather than a point estimate. A single number invites a false precision that the underlying data does not support, and an investment committee will find that out.

The output that stays useful is the model itself, not the figure it produced this quarter. Inputs move — electricity prices, capital costs, carbon prices, certification rules — and a model you can rerun is worth considerably more than a PDF that was accurate once.

That is why techno-economic analysis here delivers the model rather than only the result, and why technology assessment starts by fixing the boundary before anything is compared. It is also, frequently, why a review of an existing model is the faster engagement: the problems are usually in the boundary and the untested sensitivities rather than in the arithmetic.