Accounting is the money ledger of a TEA: the one-time cost to build the plant (capex), the recurring cost to run it (opex), and the money that comes back in (revenue and credits). Capital is a stock spent once; opex and revenue are annual flows. The three are put on a common footing — capital annualized, output divided in — to produce the levelized cost and the margin against it.
Every figure carries its basis — currency, cost year (prices drift; escalate old figures), location, and scope (where the system boundary and the battery limits are drawn). Two figures add or compare only on the same basis; a borrowed reference cost is converted first.
Capex — the one-time build
Capex is the capital spent once to build the plant — a stock, not an annual flow. It is built equipment-first: cost the dominant items per unit operation, sum to a purchased-equipment base, then gross that up through installation, off-sites, indirects, contingency, and working capital to total capex. That total — not the bare ISBL or purchased-equipment cost — is what the capital recovery factor annualizes into the capital share of levelized cost; when a TEA quotes “the capex,” this is the defensible figure. The full build-up method — with an interactive layer-by-layer walkthrough — is its own concept: capex estimation.
Opex — the recurring flow
Operating cost recurs every year, split by how it responds to output:
Variable opex scales with production — feedstock and energy (below), plus consumables (catalyst/sorbent make-up, chemicals, process water) and per-unit waste handling.
Fixed opex is largely output-independent — operating and maintenance labor, maintenance materials (a few percent of ISBL/yr), and overheads (insurance, property tax, admin).
Variable opex = consumption per unit (from the mass and energy balance) × input price; fixed opex is factored, like capital. Annual opex plus the annualized capital charge, spread over output (set by the capacity factor), gives levelized cost.
Feedstock & energy — the dominant variable line
For a commodity made at scale, feedstock and energy is usually the majority of variable opex and a large fraction of total cost — converting cheap inputs into product is the business.
feedstock & energy cost per unit = Σ (consumption per unit × input price)
Consumption is physics, price is markets. Consumption per unit comes from the mass and energy balance — better conversion and tighter heat integration lower it. Prices (electricity per MWh, gas per GJ) are set by markets, vary by location and time, and are often volatile — which is why the input price is typically the single largest uncertainty and the dominant driver in a sensitivity. Where the system boundary is drawn decides which price governs: buy hydrogen at the fence and the H₂ price dominates; make it inside from electricity and the power price does.
Revenue & credits — money in
Three kinds of money come in:
Product revenue — market price × quantity sold. For a commodity the producer is a price-taker: levelized cost is the floor it must beat, the market price is what it receives, and the gap is margin. Comparing the two — not assuming price equals cost — is how a TEA reads competitiveness.
Byproduct credits — revenue or a cost offset from a salable co-product (oxygen from electrolysis, exported steam or power) at its market price × the quantity the mass balance produces — bounded by the quantity a market can actually absorb.
Policy / carbon credits — non-physical inflows created by policy (low-carbon production subsidies, capture payments), often volatile or time-limited.
Costs can be reported gross, or net of credits (a net cost of production) — a modeling choice that must be stated and never mixed across compared figures.
Limits & typical error
It’s an accuracy class, not a point — a factored estimate is ±30–50% (FEL-1/2; see uncertainty & accuracy class). More significant figures imply precision the method can’t supply.
The factored stack multiplies the base error — each layer (install factor, OSBL, indirects, contingency) rides on the one beneath, so a 25% error in the equipment base is still ~25% at the total. Cost-year drift adds silently.
Quoting the wrong line understates the plant — purchased-equipment cost is only ~a quarter to a third of installed (off by the ~3–4× installation factor); ISBL is ~half of total fixed capital (off by ~2×). Dropping contingency reports an unlikely best case.
Per-unit fixed opex is hyperbolic in utilization — spread over output it scales as 1/output, so its per-unit contribution rises sharply at low capacity factor, while variable opex per unit is roughly flat. Maintenance-as-%-of-capital is a blunt proxy; omitted consumables (catalyst, water, waste) understate silently.
Feedstock cost compounds two uncertain factors — cost = consumption × price, so a 15% error in either moves it ~15%; price volatility usually dwarfs the engineering uncertainty.
A big byproduct credited at full price overstates revenue — a world-scale plant can swamp the regional market for its co-product, driving marginal value toward zero; the credit is bounded by the buyable quantity, not the produced quantity.
Scope errors double-count or drop cost — a Lang factor (already bundling OSBL and indirects) plus OSBL again, an interface item booked in both ISBL and OSBL, an initial catalyst fill counted as both capital and opex, or a stream credited as a byproduct on one side and bought as a feed on the other. Each layer and crossing is counted exactly once.