Why grid battery fires keep dropping

If you follow energy news at all, you've seen the footage: a battery storage site burning, crews standing back because you can't really put a lithium fire out, you wait for it to spend itself. Those images are real, and they anchor a reasonable instinct: we are bolting enormous batteries onto the grid and next to buildings, so surely the fire risk is climbing with them.

The data says the opposite. A December 2025 fact-check by Science Feedback put it plainly: grid battery fires are "a real concern, but... probably quite uncommon." It cited EPRI's Stephanie Shaw, who reported that "only 0.3% of projects experienced a failure that led to a fire with potential safety concerns in 2024," and that "over the past 6 years failure rates have dropped by 98+%." EPRI's own BESS Failure Incident Database, the most complete public tally of these events, states the failure rate "dropped by 99% from 2018 to 2025 as lessons learned from early failures have been incorporated into the latest designs and best practices." An earlier read of the same dataset, reported by Energy-Storage.news, showed the rate falling "around 9.2 failures per GW... deployed in 2018 to around 0.2 in 2023." A 97 to 99 percent drop, depending on the window, and it happened during the years deployment went vertical. More batteries, fewer fires per battery. That is the story.

I work on the systems side of energy storage, so let me explain the two things doing the work here, because the coverage usually stops at "batteries got safer" without saying how.

Part one: the chemistry changed under everyone's feet

Grid storage quietly standardized on a different battery than the one in your phone or your EV. Most stationary systems now use lithium iron phosphate (LFP) rather than the nickel-rich (NMC) chemistries built for maximum energy density. Science Feedback names the reason directly: "LFP batteries are less prone to experiencing thermal runaway than batteries made with NMC cathodes," which is "one reason many grid battery manufacturers have adopted" it.

Here's the materials read. LFP gives up some energy density, so the pack is bigger and heavier for the same stored energy. In a car that penalty matters. On a concrete pad next to a substation, it mostly doesn't, and you get a far more thermally stable cathode in return. Stationary storage is exactly the application where you can afford to trade density for stability, and the industry made that trade. It isn't that batteries became magic. It's that grid storage picked the battery whose worst day is less bad.

Part two, the one I'd lead with: the cells were rarely the problem

This is the finding that reframes the whole fire conversation, and almost no headline carries it. When EPRI's analysts sorted what actually caused these incidents, Energy-Storage.news reported that "65% of incidents could be linked to operation and integration of batteries, and just 11% were caused by failures of battery cells or modules." Read that again. The cell, the thing everyone pictures spontaneously igniting, was the direct culprit about one time in nine. The dominant failure mode was everything around the cell: how the system was integrated, commissioned, and run.

That is a profoundly hopeful result, because integration and operation are engineering problems, and engineering problems get solved. It is also exactly where my own work lives. I help design the AI battery-cycling systems for a veteran-owned energy-storage integrator, which in plain terms means the logic that decides how a pack charges and discharges, watches each string for the early drift that precedes trouble, and keeps cells inside the envelope where they stay boring. The 99% drop EPRI describes is not luck. It is thousands of those integration and control lessons getting written back into the next design. Safer batteries are partly better chemistry and largely better systems.

Safer is not safe

The honest caveat: 0.3% is not zero, and EPRI notes it is aware of "33 non-public incidents" beyond the public database, so the tally undercounts. A more stable chemistry lowers the ceiling on how bad a failure gets; it does not remove the need for spacing, gas detection, fire codes, and active monitoring. "We use LFP" is the start of a safety case, not the end of one. The systems that fail are usually the ones where someone treated the chemistry as permission to stop paying attention.

Disclosure: as noted above, I help design the AI battery-cycling systems for a veteran-owned (HUBZone) energy-storage integrator. I don't own the company and earn nothing from this link; it's disclosed because the topic is one I build in, not just write about. Nothing here is sponsored; here's the full policy.

The signal

The takeaway isn't "grid batteries are safe now, relax." It's that the risk moved. It walked out of the chemistry, where it was hard to control, and into integration and operation, where it's tractable. So when you see a storage project proposed near you and the reflex question is "what if it catches fire," ask the better one: who is operating and monitoring this, and how do they see trouble coming before it arrives? The fire rate fell 99% because more and more operators had a good answer to exactly that. The chemistry set the floor. The engineering did the rest.