Duration is all you need

The argument belongs to Ben Haley and was first published by his firm, Evolved Energy Research; Lubershane’s contribution is an introduction recommending it, plus a frame of his own about duration-limited resources. Haley starts from a fact about the system being replaced. Coal, gas and nuclear plants had no effective limit on how long they could run, so if a plant was needed in one hour there was little concern about whether it could also run in the next, and capacity adequacy could often be assessed without regard to chronology. Hydro-dominated regions were the exception, constrained by water, and have long planned around duration explicitly. What is being built now is overwhelmingly four-hour lithium-ion, which suits the problem it was bought for: nearly every system has needle peaks, one or two hours where load rises well above the surrounding hours, and a short battery covers those while earning from ancillary services. The essay is organized around what happens once those peaks are covered.

The answer is that the system stops being capacity-limited and becomes energy-limited, and it does so to itself. Haley’s team ran net load, meaning demand after wind and solar generation, across 100 scenarios for the United States, calculating at every discharge duration how much dispatchable capacity must persist in order to satisfy reliability criteria. Two structural features fall out of the resulting curve: the system needs only a limited quantity of capacity at short durations, and as duration lengthens the required capacity declines but stays meaningful. He draws the parallel to renewables, where early solar happened to generate during peak demand until enough of it was built that the moment of system risk moved to the hours when wind and solar output is lowest. Short-duration storage follows the same saturation logic, so an additional four-hour battery must either accept a lower capacity rating or add hours to reach the remaining need. The mechanism is that each addition flattens the peak it was built to serve, which he calls being a victim of its own success.

Which technologies then win depends on where along the duration axis the need sits, and the modelled answer is not a simple march toward ever longer storage. Lithium-ion occupies the short durations in virtually every run. Storage of some type almost always serves needs out to roughly sixteen hours, and beyond that it typically cannot compete with generation that has no duration limit at all, such as gas. Storage becomes economic again only at very long durations, above roughly ninety hours, where iron-air chemistry, with its high power cost, low energy cost and low efficiency, is frequently competitive in the modelling. He keeps that last finding qualified: the capacity needed at those durations is smaller than at medium durations, and what multi-day storage can potentially beat there is other low-utilization resources such as thermal plants running peaker-like profiles. Cutting across all of this, a lot of four-hour lithium-ion may still be built to serve longer peaks despite reduced capacity accreditation, because arbitraging daily energy prices can pay enough to carry it, most often in solar-heavy systems. What Haley concludes is narrower than his title: the binding constraint migrates from instantaneous discharge to sustained delivery, planning framed in gigawatts obscures that, and overreliance on short-duration storage and load flexibility, including data center flexibility, to keep pace with rapid load growth may itself introduce reliability risk.