Why a grid battery ships oversized

A battery's rating is not a description of the hardware. It is a promise about output, and like most promises it has a maintenance plan attached that nobody reads out loud.

I have written here before about what actually ages a grid battery and about where you draw the boundary when you quote its efficiency. This is the third member of that family, and it is the one with money in it. The question is not how fast a battery degrades. It is who agreed, in advance and in writing, to pay for the degradation, and what they assumed you would do with the thing.

The reference battery

When PJM needed to know what new capacity costs, it hired the Brattle Group and Sargent & Lundy to build reference designs and price them. The 2026/2027 report includes a battery energy storage system as a candidate resource. Its specification table is unusually honest, because it was written for accountants rather than for a press release.

The reference BESS is 200 MW-ac at the point of interconnection, four hours, lithium-ion, containerized. Round-trip efficiency 85%. Economic life 15 years. Salvage value zero. Use case: daily cycling.

Installed energy capacity: 1,030 MWh-dc.

Two hundred megawatts for four hours is 800 megawatt-hours out the door. The table specifies a thousand and thirty going in. Some of that gap is ordinary conversion and reserve, which the table does not itemize. But the report itemizes one piece of it explicitly, and that piece is the subject of this report.

Thirteen percent of the cells exist to be used up

Here is the sentence, from the body of the report:

"To account for degradation of the energy capacity, our cost estimate assumes that the facility will include an initial 13% overbuild, or 135 MWh-dc, with augmentations planned for Year 5 and Year 10."

One hundred and thirty-five megawatt-hours of cells are installed on day one with no expectation of being available on the last day. They are consumed on purpose. They are the margin that lets a system whose capacity is falling keep meeting a fixed number in a contract.

And when that margin runs out, the answer is not to accept a smaller battery. It is to open the containers and add more cells. That is what the industry means by augmentation, and the definition is not folklore. A law firm's 2026 briefing on storage procurement states it plainly: "In the context of energy storage, 'augmentation' refers to the process of adding storage capacity to a project over time and is typically seen in the context of battery energy storage projects."

The Brattle report also lays out the choice a developer is making, and it is a genuine trade rather than a best practice:

"Developers are currently using a range of approaches to maintain sufficient capacity to provide the rated AC output at the POI over a four-hour period, including overbuilding the initial capacity and augmenting the capacity in future years. Overbuilding the initial capacity provides the developer greater cost certainty and reduces the frequency and costs of frequent augmentation events. On the other hand, a smaller overbuild defers capital expenditures to future augmentations that reduces the initial capital costs of the facility and may allow the owner to take advantage of declining module costs, depending on future cost trends."

Read that as an engineer and it is a hedge on the price of lithium cells five and ten years out. Overbuild heavily and you have bought your future capacity at today's price. Overbuild lightly and you are betting cells get cheaper, and paying a mobilization crew more often for the privilege.

The degradation number has conditions attached

The rate itself, from the same page:

"S&L estimates that BESS energy capacity (in MWh or duration at full power) degrades by 4% in the first year and 2% in subsequent years, assuming daily cycling and a 5% minimum state of charge."

Four then two. That is the number you will see quoted in a hundred places with the second half of the sentence removed. The second half is the whole sentence. "Assuming daily cycling and a 5% minimum state of charge" is not a footnote about test conditions, it is a description of a duty cycle, and the report's own footnote says why that matters:

"Degradation occurs due to many factors, including time, ambient conditions, state-of-charge, operational profiles, depth of discharge and manufacturing defects."

Time is on that list first, which is worth sitting with. A battery that never cycles still ages. But operational profile and depth of discharge are on it too, and those are choices someone makes every day for fifteen years.

How the national planning model handles all of this

Now go up a level, to the models that decide how much storage the United States builds.

NREL publishes cost and performance projections for utility-scale lithium-ion storage specifically so they can be dropped into capacity expansion models, including NREL's own ReEDS. The 2025 update, by Cole, Ramasamy and Turan, has to compress everything above into a small number of parameters. Watch where the degradation goes.

"We have allocated all operating costs (at the one-cycle-per-day level) to the FOM. By putting the operations and maintenance costs in the FOM rather than the VOM we in essence assume that battery performance has been guaranteed over the lifetime, such that operating the battery does not incur any costs to the battery operator."

FOM is fixed operations and maintenance, a flat annual charge per kilowatt. VOM is variable operations and maintenance, a charge per megawatt-hour actually moved. The report puts everything in the fixed bucket and sets the variable bucket to zero.

It then explains that the fixed bucket is where the cell replacement lives:

"One of the primary differences in the level of FOM was whether augmentation or other performance maintenance were included in the cost. Lower FOM numbers typically include only simple maintenance while higher FOM numbers include some capacity additions or replacements to address degradation. We have adopted a FOM value from the high end and assume that the FOM cost will counteract degradation such that the system will be able to perform at rated capacity throughout its lifetime. The FOM value selected is 4% of the $/kW capacity cost for a 4-hour battery."

So the modeled battery does not degrade. Not because anyone believes that, but because 4% of capital cost per year has been set aside to make the statement true, on the authority of the Brattle analysis quoted earlier. The published spread of FOM values across the surveyed literature runs from near zero to roughly $120 per kilowatt-year, which tells you how much of this is convention rather than measurement.

The sentence that carries the risk

Immediately after, in the same paragraph:

"We assume that this FOM is consistent with providing approximately one cycle per day. If the battery is operating at a much higher rate of cycling, then this FOM value might not be sufficient to counteract degradation."

There it is, stated by the authors, in the document, unhidden.

Put the two modeling choices side by side and the shape of the problem appears. Variable O&M is zero, which means that inside the model, a cycle is free. Degradation is prepaid through a fixed annual charge, which does not move when you cycle harder. A capacity expansion model handed those parameters can therefore dispatch a battery as aggressively as the market rewards, and it will never see a bill for the wear, because the wear was invoiced once, up front, at a rate that assumed one cycle a day.

This is not a criticism of NREL. It is a national planning model, the assumption is disclosed in plain English on the page, and the alternative (endogenous degradation that responds to dispatch) is a much harder object to put in a linear program. It is a warning about what happens downstream, where the parameter is copied out of the report and the caveat is not.

Two more numbers from the same section deserve to travel with it. The chosen lifetime is 15 years, and the reason given is the honest one:

"This is a conservative choice, which we make because of the uncertainty in battery lifetimes (existing 4-hour batteries are all new, so no system has yet reached 15 or more years)."

Nobody has watched a four-hour grid battery reach the end of the life we are all planning against. Every 15-year figure in circulation, including the one in the Brattle table, is an extrapolation. And the round-trip efficiency selected is 85%, which is the same number Brattle used, which is a reminder of how few independent estimates there really are underneath a field that quotes these figures with great confidence.

It ends up in the contract

None of this stays theoretical, because a lender will not finance a promise that nobody signed. The 2026 procurement briefing describes the instrument: "LTSAs are utilized to memorialize the long-term warranty and performance guarantees provided by a battery supplier." A long-term service agreement is where the capacity guarantee, the augmentation obligation and the measurement method live.

The same briefing is careful to point out that capacity is only one of the things that decays: "a storage resource can degrade with respect to its charging speed," its stored capacity, and its ability to hold energy over the life of the project. A guarantee written against only the second one is a guarantee with two holes in it. If your contract promises 70% of nameplate energy at year ten and says nothing about how quickly the system can absorb a charge, you can satisfy the contract and still miss the morning window you actually bought the asset for.

This is the seam where my own work sits. The gap between a battery's rated capacity and its delivered capacity is not fixed by hardware, it is managed daily by whatever decides depth of discharge, resting state of charge, thermal setpoints and cycle count. That is a control problem, and it determines whether the 4% assumption holds or quietly does not. I help design the AI battery-cycling systems for a veteran-owned (HUBZone) energy-storage integrator. I do not own that company and earn nothing from this link; I flag it because it is a field I build in and not only write about. Full policy here.

The signal

A grid battery's nameplate is not a measurement. It is a service level, held up by three things that are all decisions rather than facts: cells installed beyond what the rating requires, scheduled replacements at year five and year ten, and an annual maintenance charge sized on the assumption of one cycle a day.

Every one of those is visible in public documents, stated by the people who wrote them, with the conditions attached. What gets lost between the document and the summary is always the same clause: the part after "assuming."

So when you next read that a battery degrades 2% a year, or lasts fifteen years, or costs 4% of capital to maintain, the useful follow-up is not whether the number is right. It is: at what duty cycle, and who is contractually obliged to make it true?