Over the decades I’ve observed the Levelised Cost of Electricity (LCoE) has evolved into a marketing metric, stemming from a prototype analytical one - albeit when I first saw it I thought it was a good one. This misstep is what happens when a single number gets asked myopically to compare projects that aren’t comparable – unlike the Project IRR. Energy auction designers, designers and advisors seem to just keep on sticking with it. Now with “green fuels” I see the LCoH and LCoNH3 popping up I can’t help but think here we go again…

The LCoE seems to be the most-cited number in the energy transition and government technology rankings still rely on it and every sponsor deck I see still has the comparison chart (where they are the in the lowest 25%..).

It is also, by now, indefensible for most of what it gets used for. Lazard’s 2024 LCOE+ report puts unsubsidised utility-scale solar at $29 to $92 per MWh. A 3x range presented as a comparable point estimate, wider than most US auction bid spreads. The metric survives because its critics keep using it so the systemic problem isn’t the arithmetic; it’s the lifting one single metric is doing.

The marketing metric problem.

Back in the day the LCoE was a legit screening tool. Somewhere in the last fifteen years it’s role has slipped. Policy capacity targets are calibrated against LCoE forecasts. Government technology pathways are sized off numbers that were meant to set up a conversation, not kill it.

Sophisticated developers and infra funds don’t make investment decisions on LCoE. Internal models run on equity IRR, DSCR profiles, and contracted-cashflow quality. LCoE “enters the chat” as an external-comms output. The audience that treats LCoE as decisive is the audience that should treat it least decisively: government modellers, auction designers, regulators setting reserve prices.

The usual rebuttal I hear in conference side groups / panels and foyer is “LCoE was never meant to do those things - but we have to live them.” Fair, but it’s doing them now – social media wasn’t meant to trigger youth depression but here we are! So we either pull it back to its screening role or hold it to a higher standard the vastly accelerating industry requires. Doing neither while defending the metric on its old terms is the worst option, and it’s the option in front of us - will anybody challenge it - I have some thoughts for a later piece.

A solar MWh and a wind MWh are not the same product

To my mind the deepest LCoE error isn’t the maths, it’s the denominator. Cost over energy gives a number that gets compared across technologies as though every MWh has the same value. It doesn’t. The [IEA’s VALCOE work makes the point using India: “solar PV’s LCoE falls below coal by the mid-2020s, but on a value-adjusted basis the picture inverts. As solar penetration crosses roughly 10 per cent, midday production value collapses and flexibility value climbs. By 2030, solar PV becomes less competitive on VALCOE even with continued cost reductions.”

The empirical evidence is unambiguous. CAISO logged around 1,180 hours of negative wholesale prices in 2024, roughly 13 per cent of all hours and more than double 2023. Total renewable curtailment in CAISO hit 3.4 million MWh, up 29% year on year, with solar accounting for 93%. The midday solar MWh is increasingly the one nobody wants, and the LCoE denominator does not know that… :-)

The fix exists. The IEA publishes VALCOE. CAISO runs TOD-adjusted PPAs. AEMO publishes time-of-day price duration curves. I read the “VALCOE is too complex for auctions” objection as simply conservative inertia. If a procurement team can write a TOD-adjusted PPA, they can rank bids on a TOD-adjusted comparator surely?!

While we’re on the denominator: AEMO has flagged] MLF volatility as material for investment decisions and is consulting on reform including a glide path. The standard junior-analyst response is that MLF and DLF are merchant risks that don’t belong inside LCoE. That’s wrong. They scale the MWh that reach the connection point. If your LCoE divides by gross generation rather than network-delivered energy, you’re calculating cost per MWh that someone imagined…

The market-conditional WACC

LCoE assumes a single WACC for the project life. Cost of capital follows a path conditional on secondary markets that may not co-operate including lower Ke and Kd as the project changes in maturity and ownership.

The aggressive auction bids of 2020 to 2022 priced a year-three sell-down to a super or pension fund at a Ke materially below the development sponsor’s hurdle. Development equity carried construction risk, exited, operating equity arrived at a discount. The LCoE looked competitive because it had embedded the sell-down. Through 2022 and 2023 central banks tightened at the fastest pace in decades and unlisted infrastructure repriced. Sponsors who had bid on a static WACC anchored to 2021 conditions were left with contracted tariffs that did not cover their actual cost of capital. For capital-intensive renewables, a 100 basis-point WACC shift moves LCoE by 10 to 15 %. Two hundred basis points flips the auction ranking massively - it did in the 2G, 3G and 4G auctions too.

The smart pushback, from someone with more finance theory than development cycles behind them might be that time-varying WACC just shifts the arbitrariness from one input to another. True in some respects, but the argument is about disclosure, not accuracy. A static-WACC LCoE presents itself as naive about financing and bidders are definitely not! They’re optimising against an auction that rewards the most aggressive defensible WACC trajectory. Either publish the financing-path assumption or admit the metric is hiding the assumption that decides the ranking.

In my view this is not gaming, it’s more of a selection bias. Competitive auctions reward the most aggressive defensible assumption, and across hundreds of bids that selects for the most generous WACC path. I saw this around the world from the early European off-shore wind licenses, South Africa and South America.

Deferred and disguised costs

Across the analysis I have seen, tax and decommissioning are where the model hides things - they are so far out ’“that their discounted value makes them immaterial”…I disagree and they aren’t subtle issues.

Pre-tax WACC discounting post-tax cashflows: balances inside the model, doesn’t balance in reality, almost nobody catches it. The US ITC under the IRA is 30 % base and reaches 50% with bonus adders. One bidder treats the credit as a capex offset (cutting effective LCoE by around 22 per cent), another runs it as a revenue stream with different timing, and the auction ranks them as comparable. Australian LGC revenue, historically $30 to $50 per MWh, is occasionally folded into the energy denominator, contaminating cost-per-MWh with a regulatory-revenue line. MACRS 5-year against a 25-year economic life creates a tax-shield NPV worth several per cent that nobody traces. Carry-forward losses sit in models that assume full utilisation. Franking benefits get double-counted in some sponsor models, missed in others.

Decommissioning is the inverse error. The convention is a single terminal-year line, often undiscounted, sometimes waved off as covered by residual value. The bond or surety actually costs 1 to 3 per cent of decommissioning capex annually through operations, the carry cost of the terminal obligation, and most LCoE models capture neither. Residual scrap value is real on the other side: copper, aluminium, steel, recycled silicon at 5 to 10 per cent of original capex. Many wind projects repower rather than decommission. For batteries, augmentation costs run at 15 to 25 per cent of original capex over a 15-year life and many LCoE-style storage comparisons leave that line out entirely.

From benchmark to verification

We all know that LCoE isn’t going away as it’s too embedded in policy and too convenient for charts. The question is what runs alongside it.

Auction designers could require a TOD-adjusted comparator. Published LCoE numbers should carry an explicit financing-path footnote. Tax conventions should be standardised across competing bids. Decommissioning should split into operating-phase carry cost and terminal cost. None of this is technically hard. What’s missing isn’t capability, it’s the verification layer that catches inconsistency before it shapes policy.

The lowest-LCoE project on paper is rarely the lowest-cost project in reality. More often it’s the project with the most generous defensible assumptions. Closing that gap is what multi-dimensional optimisation and verification is for.

Sources and further reading for the real nerds!

First published on Nick Crawley’s Substack.