Report 053 · Energy Storage
Every storage headline is written in megawatts. Grid planners do not pay in megawatts. They pay in accredited capacity, and in PJM's next capacity auction a 4-hour battery is credited at 59 percent of its nameplate while a 10-hour battery gets 78. Same chemistry, same cells, different number, and the thing that moved it was duration.
There is a number that governs whether a battery project pencils out, and it almost never appears in the coverage of that project. It isn't the cost per kilowatt-hour, and it isn't the round-trip efficiency, though I've written about how loosely that one gets quoted. It's the capacity credit: the fraction of a resource's nameplate rating that the grid operator is willing to count as firm capacity when it plans for the worst hours of the year.
For a nuclear plant that fraction is close to one. For a battery it is not, and how far below one it sits has almost nothing to do with the cells and almost everything to do with how many hours the thing can run.
PJM, the grid operator for thirteen states and DC, publishes its accreditation values as a plain table before each capacity auction. Here is what storage gets in the 2028/2029 Base Residual Auction, alongside a few reference points from the same page:
4-hour storage, 59 percent. 6-hour storage, 68 percent. 8-hour storage, 71 percent. 10-hour storage, 78 percent. For comparison, on that same list: nuclear 96 percent, diesel utility 93 percent, demand response 91 percent, coal 85 percent, gas combined cycle 78 percent, gas combustion turbine 67 percent, onshore wind 34 percent, tracking solar 10 percent, fixed-tilt solar 7 percent.
Read the storage rows again as a business problem. Build a 100-megawatt battery with four hours of energy behind it and you may sell 59 megawatts of capacity. Build the same 100 megawatts of power with ten hours behind it, roughly two and a half times the cells, and you may sell 78. You spent a great deal more to gain nineteen points, which tells you the curve is steep at the bottom and flattens fast. That shape is the entire economics of storage duration, and it's sitting in a one-page PDF that nobody reads.
It also tells you the 10-hour battery is being treated as a better capacity resource than a gas combined-cycle plant, which is not the sentence most people expect to write about batteries.
The method behind those numbers is called Effective Load Carrying Capability, ELCC. PJM's own description is compact: it "is a method to calculate the capacity contribution of all resources," and it "captures the expected performance of resources during tight RTO-wide system operation hours that can be caused by high loads and/or poor resource performance."
The key phrase is "during tight" hours. Accreditation does not ask what a resource can do on an average Tuesday. It asks what it does in the specific hours when the system is close to running out, because those are the only hours that determine whether the lights stay on. A resource that delivers reliably in those hours earns a high number. A resource that delivers the same megawatt-hours at low-risk times does not.
PJM adopted this for all resources, not just renewables. The Federal Energy Regulatory Commission accepted the method in January 2024, and it took effect with the 2025/2026 delivery year. Note the consequence: coal at 85 percent and gas combustion turbines at 67 percent are also being marked down, because forced outages during a winter storm are just as real a failure to show up as an empty battery. Accreditation is not an anti-storage instrument. It is an anti-nameplate instrument.
PJM also chose marginal rather than average accreditation, which matters more than it sounds. Marginal means the question is what the next resource of that type contributes to a system that already has the ones you built last year.
That marginal framing is where the interesting physics lives, and PJM states the mechanism in one sentence: "increasing one intermittent resource alone, such as solar, leads to saturation, reducing the resource's capacity contribution. Solar paired with an energy storage resource, however, could have a higher combined contribution."
Storage saturates too, and for a reason that is intuitive once you see it. A short-duration battery is very good at shaving the sharpest point of a peak. Add enough of them and the peak stops being sharp. What's left is a flatter, longer stretch of high net load, and a 2-hour battery cannot cover a 5-hour stretch no matter how many of them you install. The resource fixes the problem it is good at, and by fixing it, changes the problem into one it is bad at.
The ERCOT study commissioned from PowerGEM puts this precisely. As penetration grows, "the contribution of the next MW declines since the net load is being shifted to periods where solar or wind output is lower, or the duration need for the next MW of battery has grown since the initial batteries have already been deployed for the highest peak periods."
The same report makes the broader point that most coverage misses entirely: "The ELCC of a resource class is not a static value; it is contingent on the penetration of the resource class, underlying load characteristics, and interactions with other resource classes among other variables." There is no such thing as the capacity value of a battery. There is only the capacity value of the next battery, on this grid, this year.
ERCOT is an energy-only market with no capacity payments, so its ELCC work is used to inform planning reserve margins rather than to write checks. That makes the results useful in a different way: they are unusually granular about time of day, and they show how violently the number swings.
From the 2025 summer risk-period table, storage measured against the afternoon window and then against the evening window:
1-hour storage: 70.72 percent in the afternoon, 13.71 percent in the evening. 2-hour: 93.29 and 27.43. 3-hour: 93.29 and 41.14. 4-hour: 93.29 and 54.85. 5-hour: 93.29 and 68.15.
Look at the afternoon column. From two hours upward the number is pinned at 93.29 percent, essentially flat. Duration buys you nothing extra there, because the afternoon solar-driven net-load peak is narrow and a 2-hour battery already covers it. Now look at the evening column, after the sun has gone and the peak is long. The 1-hour battery collapses to 13.71 percent. It is, for the hours that matter most, worth about an eighth of what its nameplate says.
The winter table tells the mirror story: 1-hour storage at 25.46 percent in the morning window and 23.14 in the evening, rising to 93.37 percent for 4-hour and 5-hour units in both. The study's own summary is that 3 hours or longer is needed to hold above 91 percent in the summer afternoon window, while the summer evening and both winter windows need 5 hours or longer to stay above the mid-seventies.
Then the part that should worry anyone reading Texas storage headlines. In that same 2025 snapshot, the installed fleet was roughly 6,898 megawatts of 1-hour storage and 7,651 megawatts of 2-hour, against 202 megawatts of 3-hour, 247 of 4-hour, and 20 megawatts of 5-hour. The overwhelming majority of what got built is exactly the duration that performs worst in the evening and winter windows. Every one of those megawatts is counted at full value in a press release.
PJM says 59 percent for a 4-hour battery. ERCOT's study says 93.29 percent for the summer afternoon and 54.85 for the summer evening. These do not contradict each other and should never be averaged together.
They are different systems with different load shapes and different resource mixes, computed with different methods for different purposes: PJM's is a single annual class rating applied across all of its risk hours for a capacity market, ERCOT's is a set of seasonal risk-window values for a market that has no capacity product at all. The transferable finding is not any specific percentage. It is the structure: capacity credit falls as penetration rises, rises with duration, and is defined entirely by the hours the system is at risk rather than by the equipment.
It's also worth knowing that these numbers arrive in a tightening market. PJM's 2028/2029 auction, announced 14 July 2026, procured 138,318 megawatts of unforced capacity and cleared at $325 per megawatt-day for the entire footprint, the price cap, a 2.5 percent decrease from the prior year's cap. When capacity is scarce and priced at the ceiling, the difference between 59 percent and 78 percent accreditation stops being a modeling detail and becomes the project.
The practical version of all this, sizing a system against the specific hours a site or an operator is actually exposed rather than against a nameplate, is the work I do on the energy side, and I'll be exact about my role: I help design the AI battery-cycling systems for a veteran-owned (HUBZone) energy-storage integrator. I don't own that company and earn nothing from this link; I flag it because it's a field I build in, not just write about. Full policy here.
A megawatt is a rate, not a promise, in the same way that a cycle life figure is a claim about a duty cycle rather than about a cell. Capacity accreditation is the grid's way of converting a rate into a promise, and the exchange rate is set by duration and by how many others already made the same bet.
So when a project is announced in megawatts, three questions recover most of the truth. How many hours of energy sit behind that power rating, since that single number moves PJM's credit from 59 to 78 percent? Which hours is this grid actually at risk in, morning, afternoon or evening, because a battery that covers the wrong window is credited near an eighth of nameplate in Texas? And is the accreditation marginal, meaning the value falls as more of the same thing gets built, which it does?
The megawatts in the headline are real. They're just not the units the grid keeps score in.
Note on comparison: the PJM and ERCOT figures are computed by different methods, for different systems, and for different purposes. They are presented side by side to show the structure of capacity accreditation, not as competing estimates of a single quantity. No accreditation value from one market should be applied to the other.
U.S. Army combat veteran (Counter-IED / Electronic Warfare), peer-reviewed researcher in microwave spectroscopy, and founder & CEO of Shroombiosis. Consults on laboratory operations, AI, and supplement formulation.