Report 036 · Energy Storage
Battery datasheets quote a cycle life, one number, as if a cycle were a fixed unit of wear. It isn't. New simulations from Oak Ridge National Laboratory show that two identical packs doing two different grid jobs age along different physical pathways, at different rates. Which means the operating strategy isn't a thing you decide after buying the battery. It's part of the battery's design life.
I've written a fair amount here about battery chemistry: why LFP won the grid, why a zinc-bromide cell can't burn, why grid battery fires keep getting rarer. Chemistry is the part of the story everyone wants to talk about, because it's the part that feels like it settles things. Pick the right cell and you've bought the right outcome.
Except the field keeps producing the same inconvenient observation: identical hardware, deployed in different places, doesn't age the same way. Two sites, same cells, same integrator, same climate control, and years later one pack has meaningfully more capacity left than the other. The usual explanation is a manufacturing lottery. A team at Oak Ridge National Laboratory has now put a physics-based number on a better explanation, which is that the packs were never doing the same job.
Be precise about the claim, because this is a modeling result and not a field measurement. Surya Mitra Ayalasomayajula, Michael Starke, and Srikanth Allu built a hierarchical, physics-based simulation framework that models lithium-ion behavior at the cell, module, and pack level, with mechanistic aging built in, and ran it on a high-performance computing cluster. Their paper, "From Cell to System: Accelerated HPC Simulations of BESS Aging under Frequency Regulation and Arbitrage use cases," won a Best Paper Award at the 2026 IEEE Electrical Energy Storage Applications and Technologies Conference in Tucson this January. ORNL published a release on it July 7, 2026. The work was funded by the Department of Energy's Office of Electricity.
The engineering achievement is speed. Their abstract describes coupling that framework with HPC "to accelerate systems level evaluation by upto two orders of magnitude," letting them simulate 150 to 200 kWh systems over 500 to 1,000 cycles in days. The framework can handle more than 10,000 cells at once and multiple lithium-ion chemistries. As Allu put it, "Simulations at this scale can significantly reduce the need for costly, time-intensive, full-system testing to evaluate battery aging."
That matters more than it sounds. The honest reason nobody has good answers about 15-year pack behavior is that finding out takes 15 years, or an enormous test lab, and the product cycle is faster than either. If you can get a credible answer in days, you can ask questions you previously just had to guess at.
They compared two of the standard things a grid battery gets paid to do.
Frequency regulation is the twitchy job. The grid runs slightly fast or slightly slow, and the battery pushes or pulls small amounts of power constantly to hold the frequency steady. Frequent, shallow cycling, all day.
Energy arbitrage, or cost reduction, is the patient job. Charge when power is cheap or abundant, hold, then discharge hard into the expensive peak. Fewer cycles, but deep ones that move a lot of energy.
The result is that these are not the same amount of wear scaled up or down. They are different mechanisms. In Allu's words: "We found that these two applications drive different degradation pathways within the battery, highlighting how operating conditions influence aging at the material level." The deeper, energy-intensive arbitrage profile wore the battery down faster than the shallow regulation duty.
Note the phrase "at the material level." That's the part that makes this more than an accounting exercise. The claim isn't that one job uses up more of a fixed budget of cycles. It's that the physical process by which the cell degrades is different depending on how it's driven. Deep, high-rate cycling stresses the electrode structure and drives different reactions than a lifetime of small, shallow adjustments. Both end in lost capacity. They don't get there by the same road, and a single cycle-life number on a datasheet can't represent both.
The simulations turned up a second thing worth flagging for anyone specifying a system: aging varied not just between packs but among cells within a pack, and low-voltage architectures showed more variation in degradation than high-voltage ones. A pack is only as healthy as its worst cells, so how you wire it is a lifetime question, not only an efficiency one.
Here's the practical translation. When a datasheet says 6,000 cycles to 80% capacity, that number was measured under some specific test protocol: a defined depth of discharge, a defined rate, a defined temperature. Change any of those and you are no longer in the conditions the number describes. The spec is a measurement, not a promise, and it silently assumes a duty cycle that may look nothing like yours.
This is the same discipline I keep coming back to from my own research life in spectroscopy, where I've argued that a spectrum is a measurement, not a photograph. Every number an instrument or a test protocol gives you is the answer to a specific question asked under specific conditions. The error is never in the number. It's in carrying it somewhere it wasn't measured.
Worth keeping the other half of the picture too, because operation is not the only clock running. Calendar aging is real and independent: cells lose capacity sitting still, faster when held at high state of charge and high temperature. How much depends heavily on conditions. In one open peer-reviewed dataset of second-life cells cycled on synthetic residential and commercial grid duty profiles, Moy and colleagues reported that "despite being in storage for 3-5 months, the six cells undergo little to no calendar aging," with capacity falling 3.85% on average across 24 months of second-life testing. Under those conditions, cycling stress dominated. Under different storage conditions, the balance shifts. That is exactly the point: the answer depends on the regime, which is why you want a model that can be asked about your regime rather than a single number borrowed from someone else's.
If degradation depends on how you cycle, then how you cycle becomes a control problem with money on both sides of it. Every dispatch decision is a trade: revenue today against pack life tomorrow. Bidding aggressively into a high-price hour earns real money and spends real asset life, and until recently the second half of that trade was mostly a shrug and a warranty clause.
This 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.
And I want to be careful not to oversell what that means, because this publication is where I get to be honest about my own field. A dispatch optimizer that accounts for degradation is not magic and it does not make cells last longer than physics allows. What it can do is stop treating pack life as a free input. A controller that knows a deep discharge costs more life than a shallow one can decline a marginal arbitrage hour that doesn't pay for the wear it causes. That is a real gain, and it is a bounded one. Anyone promising an AI that extends battery life without a tradeoff is selling you something, and the ORNL result is a reason to be more skeptical of that pitch, not less, because it shows the tradeoff is physical.
The industry buys batteries on chemistry and cycle count, then assigns them a job afterward as if the job were free. ORNL's simulations say the job is part of the specification. Two identical packs given two different duty cycles are, in a real physical sense, two different assets with two different lifetimes, aging by two different mechanisms. If you're sizing storage, the question to bring to the table isn't only which cells you're buying. It's what you're going to ask them to do every day for the next decade, and whether anyone has modeled what that specific pattern does to them.
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.