Ask a non-specialist what size a battery is, and they usually say how much energy it holds (kilowatt-hours). Ask an engineer, and they give you two numbers: how much energy it holds, and how fast it can deliver it. A 100 MW / 200 MWh battery is a “2-hour” system - running flat out, it takes two hours to empty. A 100 MW / 400 MWh system is a “4-hour”. That single ratio - duration - is the most consequential sizing decision on any grid-scale battery. And across Europe, it is being made against a moving target. Projects permitted in 2023 were mostly 1–2 hour. Projects entering detailed design in 2026 are mostly 2–4 hour. The reason is not fashion; it is the interaction of arbitrage spread shape, capacity-market de-rating, and a real shift in the ancillary-services stack. Duration is the one parameter that reads every piece of market design at once.
What duration means physically
Duration is the ratio between a battery’s usable energy (MWh) and its power rating (MW). A 100 MW / 100 MWh system is 1-hour; 100 MW / 200 MWh is 2-hour; 100 MW / 400 MWh is 4-hour. The power rating is set by the inverter (PCS) and the grid connection; the energy rating is set by the number of cells. Longer duration means more cells for the same PCS - more capex on batteries, the same capex on power electronics. That cost geometry shapes every duration decision.
Arbitrage favours shorter - until it doesn’t
The simplest arbitrage window is: charge at midday when solar is saturated and discharge into the evening peak. In a market with a single deep spread per day, a 1-hour battery captures a large share of that day’s value - cycling once, deeply, at the maximum spread. Adding a second hour of duration captures shoulder prices that are typically 20–30% lower than the peak; adding a third captures lower still. NREL’s Storage Futures study and the JRC’s duration analyses model the same pattern: marginal revenue per MWh of duration declines as duration lengthens, and the optimum depends on the daily spread distribution. In continental European markets with heavy midday negative pricing and a single sharp evening ramp - the Iberian and Italian pattern today - the pure-arbitrage optimum sits around 2–2.5 hours for most sites. Once multi-day spreads from weekends and weather-driven renewable lulls are included, the optimum widens toward 3 hours.
Ancillary services favour shorter; capacity markets favour longer
Frequency-regulation services like secondary reserve (aFRR) pay mostly for power, not energy. Capacity remuneration is typically quoted in €/MW/h regardless of whether the provider is a 1-hour or a 4-hour asset. A shorter-duration battery therefore has a lower capex-per-MW of aFRR capacity and competes more aggressively in that market. Activation (energy) remuneration adds variable revenue that favours duration - a longer battery can sustain a call for longer - but capacity is the larger line.
Capacity markets push in the opposite direction. Great Britain, Ireland, Italy’s MACSE, and now Spain all de-rate shorter-duration batteries for capacity-market participation. Indicatively, 2-hour systems de-rate into the 40–60% range, 4-hour into 70–85%, and 6-hour close to 95%. If a developer assumes a capacity-market annuity in the pro-forma, the 4-hour asset clears more of that annuity per MW of interconnection than the 2-hour.
Where the two curves cross
For each site, a developer can build two revenue curves. The “merchant + aFRR” curve typically peaks at 2–2.5 hours and declines after that. The “merchant + aFRR + capacity market” curve keeps rising and typically peaks at 3.5–4 hours, depending on the cleared capacity price. The right duration is where total stacked revenue minus duration-scaled capex is maximum. With capacity payments at an indicative €10–12k/MW/year 15-year annuity, and cell prices moving from ~$115/kWh today toward ~$90/kWh by 2030 (BloombergNEF), the crossover for most continental European sites sits in the 3–4 hour range. Below €8k/MW/year, the crossover retreats to 2.5–3 hours. Above €15k/MW/year, it pushes to 4.5–5. The lesson is that duration is a function of capacity-market clearing price, not a standalone engineering parameter.
What Great Britain tells us
The GB market is about a decade ahead of continental Europe on this trajectory. Early GB projects (2015–2020) were overwhelmingly 1-hour, optimised for frequency response. As the EFR and dynamic-containment stacks matured and capacity-market de-rating methodology settled, the mainstream shifted to 2-hour through 2022–2024. By 2025 the share of new-build 4-hour projects in GB had risen materially, with the first 6- and 8-hour projects entering planning. McKinsey’s 2025 BESS revenue analysis flagged the same directional shift across every mature European market: as ancillary services saturate and capacity markets take a larger share of the stack, durations lengthen.
Cost geometry still matters
The cost difference between 2-hour and 4-hour is not 2×. BoS (balance of system), PCS, transformer, interconnection, land, site works and permitting are largely fixed between the two sizes; only the battery cells and the HVAC / fire-suppression scale with energy. Unit-capex trajectories from Lazard’s 2024 LCOS analysis and NREL’s 2024 cost update put a 4-hour system at roughly 150–170% of the capex of the equivalent 2-hour system - not 200%. That is why 4-hour can out-economics 2-hour even when the 2-hour has higher revenue per MWh of energy.
What to lock in on day one
Three design decisions can be deferred cheaply; three cannot. The deferrable set is battery chemistry generation, cell supplier selection and augmentation schedule - all of these adjust as cells improve. The set that cannot be deferred is grid-connection capacity (which fixes the MW), site footprint (which caps total containers), and MV infrastructure sizing (which caps maximum final energy). A site designed for 2-hour today with no physical or electrical headroom cannot be re-sized to 4-hour later. Developers across Europe who expect capacity markets to pay well from 2026 are overwhelmingly specifying the civils for 4-hour even where the day-one installation is 2-hour.