The Economics of Energy

4.1 — Physics permits, economics decides

There is a persistent belief among technical people that the best technology wins. It does not. The financeable technology wins, and those are different tests. Physics sets the boundary of what can be done: you cannot beat the Carnot limit, you cannot store more energy in a cell than its chemistry holds, you cannot push power down a conductor without paying I²R. Everything inside that boundary is a commercial question.

Consider three projects, all technically sound. One has a marginally lower LCOE but needs a transmission upgrade that will not be energised for four years. One is slightly more expensive but has a signed twenty-five-year offtake with an investment-grade counterparty. One is cheapest of all on paper and depends on a subsidy that expires next March. A bank will fund the second. That is the whole lesson of this chapter in one paragraph.

4.2 — CAPEX: the cost of building

Capital expenditure is everything spent to bring the asset into existence — before it has earned a single rupee. It is spent up front, it is largely irreversible, and it is almost always more than the equipment invoice.

The most common analytical error made by engineers new to project work is to equate CAPEX with equipment cost. Modules, cells, turbines and transformers are the visible part. The balance of plant, the land, the grid connection, the development spend and the financing cost during construction routinely add 40 to 70 per cent on top.

For a battery energy storage system the shape is different but the lesson is the same. Cells dominate, yet the enclosure, the power conversion system, thermal management, fire detection and suppression, the EMS, the grid interface and the civil works together carry a large fraction of the total. A quoted cell price of $60/kWh does not produce a $60/kWh installed system; delivered turnkey BESS pricing has typically run two to three times the bare cell price once everything else is included.

4.3 — OPEX: fixed and variable

Operating expenditure is what the asset costs to keep running. It splits cleanly, and the split matters more than the total.

4.4 — Capital-intensive versus fuel-intensive

Every generating technology falls somewhere on a spectrum between two extremes. At one end sit solar, wind, nuclear and storage: enormous upfront cost, near-zero marginal cost thereafter. At the other sit gas and coal: comparatively modest upfront cost, then decades of commodity exposure.

This is not a curiosity. It determines what each technology is vulnerable to. A solar developer who has commissioned a plant has essentially fixed their cost of energy for twenty-five years and is now exposed to almost nothing on the cost side — but they were acutely exposed to interest rates on the day they signed. A gas plant operator signed up for far less capital risk and instead carries a permanent, unhedgeable position in a global commodity.

4.5 — The cost of capital and WACC

Money is not free, and different money costs different amounts. A project is typically funded with a mix of debt — borrowed, cheaper, but with a fixed repayment obligation that must be met before equity sees anything — and equity, which is more expensive precisely because it is paid last and can be wiped out.

The blended cost is the weighted average cost of capital, and it is the single most important number in energy finance.

This is why the same technology is genuinely cheaper in some countries than others, and why concessional finance and sovereign guarantees are not a footnote in the energy transition — for capital-intensive technologies they are the main lever. Reducing the cost of capital by two points is often worth more than a decade of module efficiency improvement.

4.6 — The time value of money: discounting and NPV

A rupee today is worth more than a rupee in ten years, because today’s rupee can be put to work. Discounting converts future cash flows into their equivalent value now.

The compounding is severe over the horizons energy assets operate on. At a 10% discount rate, ₹100 received in year 25 is worth about ₹9.20 today. This is why the back half of a thirty-year project life contributes so little to its valuation, and why a two-year construction delay is far more damaging than losing two years off the end of the asset life.

Net present value sums every discounted cash flow, including the negative one at the start.

A positive NPV means the project earns more than the discount rate demands and creates value. A negative NPV means it destroys value even if it is profitable in accounting terms. NPV is the decision rule; everything else is commentary on it.

4.7 — IRR, payback and their failure modes

The internal rate of return is the discount rate at which NPV equals zero — the project’s own compound annual return. It is popular because it is a single comparable percentage, and it is dangerous for the same reason.

Two IRRs matter and they are not the same number:

• •Project IRR is computed on unlevered cash flows, before any financing. It measures the asset.
• •Equity IRR is computed on cash flows to the shareholder after debt service. It measures the investment. With cheap debt it is much higher than project IRR; if the project underperforms, it collapses much faster.

IRR fails in three specific ways worth knowing. It implicitly assumes interim cash flows are reinvested at the IRR itself, which is rarely true. It can produce multiple mathematical solutions when cash flows change sign more than once — a battery with a mid-life augmentation, for instance. And it ignores scale entirely: a 40% IRR on ₹5 crore is a worse outcome than a 14% IRR on ₹500 crore, and only NPV tells you that.

Simple payback — capital divided by annual cash flow — ignores the time value of money and everything that happens after the payback date. It is a useful screening heuristic and a terrible decision rule.

4.8 — LCOE: the common denominator

Levelised cost of energy answers a specific question: across the whole life of this asset, what is the average cost per unit of energy delivered, with all cash flows discounted to today? It exists so that technologies with wildly different cost structures can be set side by side.

4.9 — What LCOE hides

LCOE is indispensable and routinely misused. Its defect is structural: it divides lifetime cost by lifetime energy and, in doing so, treats every megawatt-hour as interchangeable. On a real grid they are not remotely interchangeable.

• •It ignores when the energy arrives. A megawatt-hour at 2 p.m. in October and one at 7 p.m. in January can differ in market value by a factor of twenty. LCOE prices them identically.
• •It ignores dispatchability. A plant you can call at will and a plant that produces whatever the weather permits are not comparable products, whatever their LCOE.
• •It ignores system integration cost. Transmission reinforcement, balancing reserves, curtailment and the firming capacity required to make variable output usable sit outside the project boundary and inside somebody’s bill.
• •It is extremely sensitive to assumptions. Capacity factor, discount rate, asset life and degradation are all inputs the modeller chooses. Move all four favourably and you can shift LCOE by 40 per cent without changing a single piece of equipment.
• •It says nothing about revenue. A project can have a market-beating LCOE and still lose money, if the hours it generates in are the hours prices are lowest.

The serious literature has largely moved on to value-adjusted metrics — system LCOE, or the value-adjusted LCOE that divides lifetime cost by the market value actually captured. But LCOE remains the number in the press release, so it remains the number you have to know how to take apart.

4.10 — Cost versus value: the price you actually receive

Cost is what you pay to produce a unit. Value is what somebody pays you for it. The gap between the two is where most energy projects succeed or fail, and it is invisible in an LCOE table.

The industry measure is the capture rate: the average price a generator actually receives, divided by the time-weighted average market price over the same period. A dispatchable plant that can concentrate its output into high-price hours captures above 100 per cent. Solar, which produces its energy in a narrow midday window that everybody else is also producing in, increasingly captures well below it.

In several European markets solar capture rates have fallen from near 100 per cent a decade ago to well under 70 per cent in high-penetration years, and the trend is in one direction. A project modelled at ₹3.00/kWh average market price and an assumed 95% capture rate that actually achieves 70% has lost a quarter of its revenue, which for a levered project is usually the difference between a healthy equity return and a covenant breach.

4.11 — Marginal cost and the merit order

To understand why capture rates fall you have to understand how wholesale electricity prices are set. In most liberalised markets, generators offer into a pool and the system operator dispatches them in ascending order of marginal cost until supply meets demand. Everyone dispatched is then paid the offer of the last unit needed — the marginal unit. This is the merit order, and the payment convention is called marginal or uniform pricing.

The consequence is that price is set by the most expensive plant running, not the average. On a mild spring afternoon with strong solar and light demand, the marginal unit might be a coal plant bidding near its fuel cost, and the price is low. On a still January evening, the marginal unit is an open-cycle gas peaker, and the price is many times higher. The solar plant and the nuclear station are paid the same clearing price as the peaker in that hour, which is the mechanism that funds capital-intensive plants with no fuel cost.

4.12 — Cannibalisation and the duck curve

Zero-marginal-cost generation enters the merit order at the far left. Add enough of it and it pushes the entire supply curve rightward, so the demand line intersects it at a much cheaper unit. Prices in those hours fall — for everyone, including the solar plants that caused the fall. This is price cannibalisation, and it is self-inflicted by design.

The daily signature of this is the duck curve: gross demand rises steadily toward an evening peak, solar subtracts from it through the middle of the day, and the net load the remaining fleet must serve collapses into a midday belly before ramping violently upward at sunset. In high-penetration markets that belly has repeatedly gone to zero and below, producing negative wholesale prices — moments when generators pay to keep producing rather than shut down and lose a production subsidy or incur a restart cost.

Two things follow, and both are commercially enormous. First, the marginal value of the next solar plant in a saturated market is far below the average value of the existing fleet. Second, the neck of the duck — the steep evening ramp into the highest prices of the day — is precisely a four-hour discharge window. The duck curve is not an argument against solar. It is the single clearest argument for storage, written in prices.

4.13 — Storage economics: LCOS and the spread

Storage does not generate energy; it moves it in time, and loses some of it in the process. Its cost metric is levelised cost of storage, and unlike LCOE the denominator has to account for round-trip efficiency and for the fact that a finite number of cycles is available over the asset’s life.

Notice the two levers that dominate. Cycles per year is in the denominator: a battery that cycles 330 times is roughly twice as cheap per delivered kilowatt-hour as one that cycles 165 times, from identical hardware. And efficiency appears twice — once shrinking the delivered energy, once inflating the charging cost per delivered unit. Round-trip efficiency is worth far more to LCOS than its headline percentage suggests.

4.14 — Revenue stacking

Because arbitrage alone rarely pays, storage is financed on a stack of simultaneous revenue streams. The same asset is sold to several different buyers for several different properties, and the art of the business case is fitting them together without double-committing the same megawatt.

Two constraints govern the stack. Some services are mutually exclusive in a given half-hour: a battery holding headroom for frequency response cannot simultaneously be fully discharging into an arbitrage trade. And market rules decide what may be stacked at all — a regulation that forbids an asset from earning capacity payments while also bidding into the energy market can destroy a business case that the physics fully supports.

4.15 — Contracts: PPAs, tolling and merchant risk

A project’s revenue can be contracted, exposed to the market, or some blend. The choice determines the cost of capital more than any technical decision.

The financial consequence is direct. A project with a twenty-year PPA to an investment-grade offtaker might support 75 to 80 per cent debt at a modest margin. The same project fully merchant might support 40 to 50 per cent debt at a materially higher margin, if a lender takes it at all. Run that through the WACC formula and the merchant project needs a substantially better raw return to reach the same equity outcome.

Which is why counterparty credit quality is a first-order engineering concern. A PPA is only as good as the entity signing it; a twenty-five-year contract with a financially distressed distribution utility that has a history of delayed payment is not the risk-free instrument the model treats it as.

4.16 — Project finance and the debt sculpt

Most energy assets are built in a special purpose vehicle: a company owning nothing but that project, borrowing on a non-recourse basis so that lenders can look only to the project’s own cash flows and assets, not the sponsor’s balance sheet.

Because the cash flows are the only security, lenders size the debt off a coverage ratio rather than off asset value.

It also explains a behaviour that otherwise looks irrational: sponsors will accept a lower PPA price in exchange for a longer tenor or a firmer offtaker, because the improvement in financing terms more than compensates for the revenue given up.

4.17 — Risk is the other half of return

Every number in a model is a forecast, and the discount rate is the market’s verdict on how much to trust it. Risk does not appear as a line item; it appears as a higher required return, which is the same thing as a lower valuation.

• •Construction risk — cost overrun, schedule slip, contractor insolvency. Usually transferred through a fixed-price, date-certain EPC wrap with liquidated damages, which is not free: the wrap itself costs several per cent of CAPEX.
• •Resource risk — the wind or sun may underperform the assessment. Lenders typically size debt off a P90 or P95 energy estimate rather than the P50 central case.
• •Price and volume risk — merchant exposure, curtailment, cannibalisation, capture rate decay.
• •Counterparty risk — the offtaker fails to pay, or pays late. Endemic in markets with financially stressed distribution utilities.
• •Technology and degradation risk — the asset fades faster than warranted, or the warranty provider is no longer solvent when it is called.
• •Regulatory and policy risk — retroactive tariff changes, market redesign, duty or tax changes mid-build.
• •Refinancing and rate risk — a mini-perm facility maturing into a hostile credit market.

The useful discipline is to ask, for each risk, who is best placed to bear it and whether they are actually being paid to. Risk that has been contractually transferred to a party without the balance sheet to absorb it has not been transferred at all.

4.18 — The cost of delay

Time is not a soft variable. Delay attacks a project from several directions simultaneously, and the combined effect is consistently underestimated.

1. 1Interest during construction accrues on drawn capital that is earning nothing. On a ₹1,000 crore project at 9%, a nine-month delay with ₹500 crore drawn adds roughly ₹34 crore of capitalised interest.
2. 2Revenue is pushed back a full year and, in NPV terms, every subsequent year’s cash flow is discounted one period harder.
3. 3Deadlines expire. Subsidy windows, safe-harbour equipment rules, connection agreement milestones and tariff-eligibility dates are usually hard dates, not negotiable ones.
4. 4Contracted prices are missed. A PPA with a scheduled commissioning date typically carries per-day liquidated damages and, past a longstop, a termination right.
5. 5Market conditions move. Commodity prices, freight, interest rates and the merit order itself are all different eighteen months later.

This is the honest answer to why developers pay large premiums for schedule certainty — expedited freight, dual-sourced long-lead items, over-resourced construction teams. On a discounted basis those premiums are usually cheap relative to the alternative.

4.19 — Learning curves and scale

The most important long-run cost dynamic in energy is not a technology breakthrough. It is the learning curve: the empirical regularity that unit cost falls by a roughly constant percentage for every doubling of cumulative production.

The compounding is what makes this powerful. At an 18% learning rate, four doublings of cumulative production — a sixteen-fold increase, which battery manufacturing has achieved comfortably within a decade — takes cost to 0.82⁴ ≈ 45 per cent of where it started. This, rather than any single invention, is why lithium-ion cells fell roughly 90 per cent in real terms between 2010 and the early 2020s.

4.20 — Manufacturing economics is not project economics

A cell factory and a battery project are both called “battery investments” and they obey almost opposite rules. Confusing the two is one of the most common analytical errors in the sector.

The cruel asymmetry of manufacturing is that the learning curve which makes the industry successful also deflates the price of your own output every year. A factory must run at high utilisation and keep climbing its own cost curve simply to stand still. A project, by contrast, locks in its capital cost on the day of financial close and then enjoys a quarter-century of near-zero marginal cost. Falling technology prices are unambiguously good for the project developer and a permanent treadmill for the manufacturer.

Anyone building cells in India under a production-linked incentive scheme is running the manufacturing model; anyone deploying those cells into a storage tender is running the project model. Applying the wrong mental model to either produces confident, wrong answers.

4.21 — Commodity exposure and supply chains

Capital-intensive technologies escape fuel price risk. They do not escape commodity risk — they simply concentrate it into a single moment, the procurement window, rather than spreading it over an operating life.

• •Lithium, nickel, cobalt. The 2021–22 lithium spike raised cell prices for the first time in the technology’s commercial history, reversing a decade-long decline in a matter of months.
• •Polysilicon. A handful of plants set the marginal cost of the global solar module supply; an outage or an export restriction propagates worldwide within a quarter.
• •Copper and aluminium. Cabling, busbars, transformers and the grid connection itself. Copper is the quiet commodity exposure in every electrification project.
• •Steel and cement. Foundations, towers, mounting structures, civil works. Dominant for wind, significant everywhere.
• •Freight and duty. Container rates and tariff regimes can move delivered cost more than the underlying commodity does, and change with far less warning.

The commercial response is to compress the exposure window: fix equipment prices at or near financial close, secure long-lead items early, and where possible index the offtake price to the same input that drives the cost. A project that fixes its revenue for twenty-five years while leaving its capital cost floating for eighteen months has taken a large, unhedged position without meaning to.

4.22 — Location: transmission, curtailment and capacity value

Two identical plants at different nodes are different assets. Location determines the resource, the price the local market pays, whether the output can physically leave, and how much of the system’s reliability need the plant is credited with serving.

This is the economic content of the phrase “the grid is the constraint.” Interconnection queues measured in years, and transmission build-out measured in decades, mean the best resource is frequently not the best project. A slightly windier site behind a congested line is worth less than a mediocre site with firm capacity at a node the system actually needs served.

4.23 — Policy, subsidy and the price of carbon

Policy is not an external distortion applied to a clean market; it is part of the cost structure, and it comes in forms with quite different financial characteristics.

A carbon price deserves special attention because it works through the merit order rather than the project. Charging a coal plant for its emissions raises its marginal cost, moves it rightward in the stack, and both reduces its running hours and raises the clearing price in hours where it remains marginal. A renewable generator’s revenue improves without any subsidy being paid to it — which is why economists prefer the instrument and why it is politically the hardest to implement.

4.24 — Comparing technologies honestly

Everything above collapses into a short and demanding checklist. When someone puts two technologies side by side and declares one cheaper, these are the questions that determine whether the comparison means anything.

• •Same boundary? Does each cost include grid connection, land, development and interest during construction, or does one stop at the factory gate?
• •Same discount rate? Comparing a capital-intensive plant at 5% against a fuel-intensive one at 10% is not a comparison, it is an assumption.
• •Same lifetime and degradation? A 30-year assumption against a 20-year one moves LCOE by a fifth before anything else is considered.
• •Same capacity factor basis? Modelled P50, contracted, or actually achieved?
• •Same product? Firm dispatchable power and weather-dependent energy are not substitutes, whatever their per-unit cost.
• •Value, not just cost? What capture rate does each achieve, and is it stable over the asset life?
• •Who bears which risks, and are they paid for it?
• •What is excluded? System integration, firming, transmission, decommissioning, waste, insurance.

The mental model that survives all of this is deliberately simple. An energy asset is worth the energy it delivers, multiplied by the price it captures, minus what it costs to build and run, adjusted for how confident anyone is in the whole chain.

The cheapest technology on a spreadsheet wins nothing. The one that delivers energy when the system will pay for it, under a contract someone will lend against, at a cost of capital the market will actually offer — that one gets built. Chapter 5 moves from the money to the machine that carries it, and looks at the electrical grid itself: how generation reaches consumption, and why the network is now the binding constraint on almost everything in this chapter.