The capacity factor is not a report card

A nuclear plant runs at 90%. A wind farm at 35%. The conclusion sounds obvious — until you ask what the numbers actually mean.

Capacity factor works like an odometer. It tells you how many kilometres a car has driven — in this case, how many megawatt-hours a plant produced compared to what it could have produced running flat out all year. One hundred percent equals 8,760 hours of full-power generation. At 90%, a nuclear unit produces 7,884 equivalent full-load hours of electricity per MW of installed capacity. At 35%, a wind farm delivers 3,066. That difference is real. One megawatt at 90% does produce more annual energy than one megawatt at 35%.

But an odometer does not tell you whether the car arrived on time, drove safely, consumed too much fuel, or ended up in the right place. Capacity factor has the same limitation. It measures a single dimension — annual energy output per unit of capacity — and says nothing about when that output arrives, how reliably it shows up during moments of scarcity, how quickly it can adjust, what it is worth on the market, or what the rest of the grid needs to do to accommodate it.

The error is not the number. It is using the number as a universal ranking.

The peaker paradox

The best way to see the boundary of capacity factor is to look at a natural gas peaking plant.

A peaker is designed to run during the hours when electricity demand spikes — cold mornings, hot afternoons, moments when other plants are unavailable. It may run only a few hundred hours per year, producing a capacity factor of five to ten percent. By the logic of simple comparison, it would rank near the bottom of any energy-source table.

In practice, a peaker with a low factor but high reliability during scarcity hours is one of the most valuable assets a grid can hold. It earns its place not by producing energy continuously but by standing ready when the system would otherwise fail. The low factor is not a weakness. It is a description of its job.

The IAEA makes this distinction explicit for nuclear plants. Their reporting system separates the load factor — equivalent to the standard capacity factor — from the unit capability factor, which measures how much of the time a unit was technically available, regardless of whether it was dispatched. The same mathematical ratio can reflect deliberate restraint or genuine failure. Without knowing which, the diagnosis is wrong.

Gas peakers take this to its logical extreme: a plant that is theoretically available but chosen not to run earns an identical factor to one that suffered repeated breakdowns. The number absorbs both realities into a single value. The first question after reading any capacity factor should always be: what caused it to be this high or low?

Four causes, one number

Capacity factor collapses four fundamentally different situations into a single figure.

For a solar farm, low nighttime output is a physical constraint of the fuel source. For a wind turbine, the factor reflects the local wind regime, turbine design, and siting — none of which is under the operator's control once the turbine is built. For a nuclear or gas plant, the factor may reflect planned maintenance, unplanned breakdowns, curtailment from grid congestion, economic decisions to hold capacity in reserve, or regulatory constraints on output.

These causes call for different responses. A chronically underperforming nuclear unit may signal ageing infrastructure or management problems. A natural gas plant with a low factor may be operating exactly as designed. A wind farm curtailed by grid congestion may have a higher capacity factor if the transmission bottleneck were resolved. None of this is visible in the raw percentage.

A second source of confusion is the slip between related concepts. Full-load hours are just capacity factor expressed in time rather than percentage — they contain no additional information. Availability or capability factor measures what was technically ready, including hours when the plant was not dispatched. Capacity credit and effective load carrying capability (ELCC) ask what a plant contributes to the grid's ability to meet peak demand — a question that is profoundly time-dependent and probabilistic, not a year-long average at all.

Using capacity factor as a proxy for any of these is a category error.

Can nuclear plants regulate?

A common rhetorical move in energy debates is to treat a high capacity factor as evidence of inflexibility. If nuclear plants run at 90%, surely they cannot respond to demand changes?

The IEA documents load-following operation in nuclear fleets in France, Belgium, Slovakia, and Sweden. Plants in those systems routinely ramp output up and down to track demand. Technically, large thermal units can regulate. The question is whether they choose to.

The answer depends on economics, not physics. A large reactor with high fixed costs and low marginal fuel costs earns more by running continuously than by modulating. Regulatory frameworks in some countries further push toward baseload operation. The result is a plant that can regulate but has limited incentive to do so — and that choice is reflected in a high capacity factor.

The distinction matters because critics sometimes conclude from high nuclear capacity factors that grids cannot integrate nuclear with variable renewables. The actual constraint is economic dispatch logic, not a physical inability to ramp. Whether market design can change that incentive is a legitimate policy question. Whether a specific reactor can ramp is largely a technical one, and the answer is often yes.

The symmetric case applies to wind and solar. A low capacity factor for renewables does not indicate inefficiency or fuel waste — there is no fuel. It reflects that the resource is sometimes unavailable. The practical implication is that more installed megawatts are needed to deliver the same annual energy, and that the temporal match between production and demand needs to be evaluated separately. But "low capacity factor" does not mean "useless." It means: know what you are counting.

What the grid actually needs to know

Adequacy, flexibility, and market value are three distinct questions. Capacity factor speaks to none of them directly.

Adequacy asks whether a power system can meet demand during the hours when it is hardest to do so. ENTSO-E models this probabilistically: the loss of load expectation (LOLE) expresses risk in expected hours of shortfall, not a prediction that exactly that many hours will fail. Capacity credit and ELCC — the increase in peak demand a resource can support without degrading that adequacy metric — are the proper tools for this analysis. Both are sensitive to the exact profile of output during scarcity hours, geographic correlation with other plants, and the penetration of similar resources already in the mix. A grid with a great deal of solar PV may find that the next megawatt of PV contributes diminishing adequacy value, as peak net demand shifts toward the evening.

Year-average capacity factor cannot substitute for these hourly, probabilistic calculations.

Flexibility asks how quickly and reliably a plant can change output. The relevant measurements are ramp rates, start-up time, minimum stable load, minimum run time, reserve service availability, and the cost of providing each. A large battery is fast but limited in duration. A gas turbine can run for days but takes minutes to start and carries fuel supply risk. Demand response can suppress consumption during peaks — if the underlying contracts and metering actually allow it. All of these resources can sometimes provide multiple services, but not simultaneously without constraint: holding spinning reserve means the plant cannot dispatch its full output for energy at the same moment.

The IEA's flexibility framework emphasises that generation, demand, storage, and networks must be evaluated together. The same wind park with the same capacity factor in two different grid contexts contributes differently to flexibility because the surrounding system changes what it needs. One dimension that capacity factor metrics routinely miss is the grid's inertia-based stability services — a problem that becomes structural as conventional generation is retired.

Market value is where capacity factor most directly enters the economics — but through a more precise channel than simple ranking. LCOE, the levelised cost of energy, uses annual output in its denominator. A plant with higher capacity factor spreads its fixed costs over more megawatt-hours, all else equal. But LCOE assigns the same value to every megawatt-hour regardless of when it is produced. A unit that reliably generates during evening demand peaks or cold-snap days produces electricity worth more than one that floods the market during midday solar surpluses.

The IEA's VALCOE framework attempts to correct this by adding energy, capacity, and flexibility value to cost calculations. It is a better basis for comparing technology portfolios, but it remains sensitive to price assumptions, system scenarios, and the specific market design in which a technology operates.

For any honest cost comparison, the relevant question is incremental: what changes in total system cost — including generation, network, storage, balancing, and reserve — when one clearly specified alternative replaces another, against the same reliability standard?

The same standard applies to claims about integration costs. Asserting that wind and solar carry no integration costs understates the real demands they place on balancing, networks, curtailment management, storage, and reserve capacity. Asserting that all such costs are caused by variable renewables overstates it: network expansion is also driven by demand growth, electrification, site selection, ageing infrastructure, and the retirement of existing capacity. Neither direction of error is neutral in policy terms. The only defensible attribution compares two well-specified scenarios with the same reliability standard, and is explicit about which costs follow from which choice.

A dashboard, not a score

The implication is not that capacity factor should be discarded. It belongs in its own column — the one labelled "annual energy per unit of installed capacity." Within a single technology type and operating context, an unexpected drop signals something worth investigating. In project finance, it is indispensable: a capital-intensive installation that produces fewer megawatt-hours has higher unit costs. These are genuine uses.

What it cannot do is serve as the single axis for ranking electricity sources. A more honest evaluation runs across at least four dimensions:

Adequacy contribution — not annual average output but capacity credit, ELCC, and the probabilistic risk profile during the hours and scenarios the grid is most stressed. Methods: LOLE/EUE analysis, ERAA scenario modelling.

Operational flexibility — ramp rate, start time, minimum stable generation, reserve capability, and the reliability of those services when the system needs them. Methods: technical specification, market product qualification, observed dispatch data.

System cost impact — not the project's LCOE but the incremental change in total system costs across a comparable scenario pair with the same reliability standard. Methods: consistent scenario comparison with transparent baseline assumptions, allocation rules, and sensitivity analysis.

Physical and societal constraints — emissions, land use, water consumption, material inputs, waste, safety, and the distribution of costs and benefits across communities. Methods: lifecycle analysis, local impact assessment, explicit decision-making on acceptable tradeoffs. This column is not a measurement problem. It is a political question.

Publishing a table with these four dimensions, complete with assumptions, ranges, and sensitivities, gives readers something to interrogate. Collapsing them into a single percentage gives them something to misread.

The Dutch addendum

In the Netherlands, the grid congestion problem adds a layer that capacity factor discussion often ignores. A megawatt-hour may have high national value but be physically non-transportable because the relevant transmission corridor is saturated. Conversely, demand response — contractually guaranteed reductions in industrial consumption during peaks — can defer expensive network investments without generating a single additional megawatt-hour. Its contribution is invisible in any generation-side capacity factor.

This means the comparison framework must start from the missing system function: what does the Dutch grid lack in which location, during which hours, with which certainty? The answer may be evening capacity in a congested region, seconds-level frequency response, multi-day seasonal storage, or something else entirely. Whichever source best fills that specific gap — at acceptable cost, risk, and societal impact — is the relevant comparison, regardless of what its annual capacity factor happens to be.

The 90% and the 35% are not wrong. They describe something real. The odometer is a useful instrument — it just does not tell you where you are going, whether you arrived, or whether the engine is healthy. A system that needs reliable electricity across all hours and seasons requires a richer set of instruments to read.

Sources

Definitions and measurement

• EIA Glossary: Capacity factor — Standard definition and formula for capacity factor
• EIA: How much electricity does a power plant generate? — Practical context for reading capacity factor data
• IAEA PRIS: Glossary and performance indicators — Nuclear-specific definitions including load factor and capability factor

Adequacy and capacity value

• ENTSO-E: European Resource Adequacy Assessment — Probabilistic modelling of European adequacy using LOLE and ELCC
• NREL: Resource Adequacy and Capacity Accreditation (2024) — Current methodology for capacity accreditation of variable resources
• NREL: Comparison of Capacity Value (2012) — Foundational comparison of capacity value methods across technologies
• NREL: Renewable Electricity Futures, vol. 1 — Planning reserves and capacity value in high-renewable scenarios

Nuclear flexibility and integration

• IEA: Nuclear — Global Energy Review 2020 — Evidence for load-following in French, Belgian, Slovak, and Swedish nuclear fleets
• IEA: The Power of Transformation (2014) — Flexibility requirements in systems with high shares of variable renewables

Economics and system cost

• IEA: Global Energy and Climate Model — LCOE/VALCOE — Methodology and data for levelised and value-adjusted cost comparisons
• IEA: Electricity 2026 — Flexibility — System-level flexibility requirements and the role of market design
• IEA: Sustainable Recovery — Electricity — Integration of energy and capacity costs in scenario analysis

Physical and societal dimensions

• IPCC AR6 WGIII, chapter 6 — Lifecycle emissions, land use, and material inputs across electricity technologies
• ACER: Electricity infrastructure monitoring report 2024 — European grid adequacy and congestion data including the Netherlands