The Fire Record: Notable BESS Fires and Where the List Lives

Every battery hearing ends up at the same place: somebody holds up a photo of Moss Landing and somebody else holds up a chart showing failure rates falling off a cliff. Both are real. This lesson gives you the record behind each, starting with the incidents and then the database most people are quoting without having opened.

Notable incidents

Unless noted, size, chemistry, enclosure and outcome come from EPRI's BESS Failure Incident Database as downloaded in October 2026. Sizes are MW / MWh; a dash means EPRI lists no figure.

A few patterns stand out. The worst human toll came early: Arizona in 2019 and Beijing in 2021, both explosions that hit firefighters. The largest environmental event, Moss Landing in 2025, was an older NMC system packed into a building. Recent fires at modern outdoor sites have mostly stayed in one or a few units, yet still brought evacuations, shelter orders, school closures and days of smoldering. And Idaho has its own entry.

Where the list lives

The list most people mean is EPRI's BESS Failure Incident Database. It used to live on EPRI's Storage Wiki, and the old storagewiki.epri.com database page now redirects to the Energy Storage Hub. EPRI's methodology page says:

• It began in 2021, after the South Korean fires and the Arizona explosion, and includes incidents back to 2011.
• The main table tracks utility-scale and commercial and industrial systems. A second table tracks other incidents: manufacturing, transport, storage and recycling. Residential systems are not tracked.
• Its focus is incidents with wider public health and safety impact, not routine operational failures.
• Entries come from media reports and public documents such as root cause reports and press releases, found through searches of English-language media, keyword alerts and verified crowdsourcing, with translated local-language reports for some Taiwan and South Korea incidents. EPRI says it cannot guarantee every incident is captured, especially before 2018 and 2019.
• Source documents are archived, and the whole table can be downloaded.
• It was renamed from "Failure Event" to "Failure Incident" to match emergency response language.

As downloaded in October 2026, the stationary table had 165 rows. Thirty-two are placeholders that name only a country and year, for incidents EPRI knows about without public detail. Of the rows with locations, South Korea had 56 and the U.S. 37. Nineteen rows are flagged as explosions.

This is a research database, not a reporting system. Inclusion depends on someone writing about the incident, and two-thirds of the incidents EPRI tried to classify in 2024 lacked enough public information to name a cause. States are filling gaps on their own. After three fires in summer 2023, New York formed an Inter-Agency Fire Safety Working Group, which released draft recommendations on February 6, 2024 for lithium-ion systems over 600 kWh, covering peer review of permit applications, emergency response planning, local fire department training and updates to the state fire code. New York adopted energy storage code updates based on that work on July 25, 2025, and the 2025 New York State Fire Code that carries them took effect December 31, 2025, according to NYSERDA.

The failure-rate claim

EPRI compares incidents to the cumulative capacity deployed. Its database page says the rate per cumulative deployed capacity dropped 99 percent from 2018 to 2025, and that its graph includes 33 non-public incidents. EPRI's October 2026 report, Risk Factors and Failure Statistics of Lithium-Ion BESS, put the drop at 98 percent; Energy-Storage.news updated its coverage after EPRI clarified that figure. The May 2024 version of the analysis found a 97 percent drop from 2018 to 2023. The 2026 report also found failure rates now comparable to large oil-filled transformers, that 93 percent of recorded utility-scale incidents caused no injuries or deaths, and that no incident spread beyond the facility perimeter.

Meanwhile the yearly count of incidents in the public table has not fallen much: 17 rows dated 2018, 30 in 2023, 26 in 2024 and 23 in 2025.

Rate = incidents / cumulative GWh deployed

Rate 2025 / Rate 2018 = 0.02 (a 98% drop)
(23 / C2025) / (17 / C2018) = 0.02
(23 / 17) x (C2018 / C2025) = 0.02
1.35 x (C2018 / C2025)      = 0.02
C2025 / C2018 = 1.35 / 0.02 = about 68

Reading the debate fairly

The industry case rests on that rate, on EPRI's finding that newer vintages fail less, on the move away from big indoor installations, and on the absence of U.S. injuries or deaths since 2019. It is a strong case about the fleet.

Residents and opponents make several points. The database depends on public reporting, so it likely undercounts. Most causes are unknown. A per-GWh rate describes the fleet, not the consequences for the people living next to the one site that fails, and evacuations, school closures and multi-day smoldering keep happening at modern sites. Some argue the rate is flattered because the denominator is full of new systems that have not had time to age into failure. EPRI's data cuts against that last point: it found failure rates fall with age, and 72 percent of failures with known age happened in the first two years. So young fleets tend to fail more, not less.

Where the evidence points: the improvement is real and large, the database is incomplete, and a falling rate does not answer the local question of what happens, and who pays, when this particular site is the one in the table next year.