Everyone is watching batteries. The record says grid.

The public conversation about energy safety is almost entirely about battery fires. The register does not show that. Grid infrastructure accounts for nearly half of every incident recorded, and electrical faults outnumber fires.

This report sets out what the register holds, counted on 2026-09-25. It is deliberately explicit about what the numbers cannot tell you, because a safety record that overstates itself is worse than no record at all.

This quarter73 incidents occurred between July and September

These are the incidents the register records as having happened in the quarter, by the date the reporting gives. 49 at grid infrastructure, 8 at solar pv, 5 at battery storage, 5 at wind (onshore), 4 at hybrid co-located, 2 at wind (offshore). 5 were major. 35 of the 73 are still provisional, published from first reports and rechecked automatically as more sources appear.

The 5 major incidents

What it cost, where the reporting says

Consequences are recorded only where a source states them. Most reports state none, and the absence of a figure is recorded as an absence, not a zero.

Quarter by quarter

Read a rise with care. The register widens its search as it goes, so part of any increase is the register seeing more, not the world producing more. 25 in July, 11 in August, 37 in September.

Significant incidents this quarter, a selection

18 of the 47 significant incidents. The full quarter, every entry with its sources, is in the register.

What failedThe component and the mode, where the reporting says

The most useful thing a record can say is what broke. News reporting says it less often than anyone would like, which is why not determined is the largest mode in every technology. Where it is stated, this is the pattern.

Finding 01Grid infrastructure is the largest category, by a distance

Substations, transformers, switchgear and transmission account for 336 of 713 entries. Battery storage, which dominates the coverage, accounts for 105.

Share of all published entries. Grid infrastructure is 47%, battery storage 15%.

Finding 02Electrical faults outnumber fires

302 entries are classified as electrical and 277 as thermal. Thermal runaway is the subject of most of the industry’s safety writing, and it is not the largest category in the record.

Finding 03Most incidents happen on an ordinary day

573 of 713 entries, 80%, occurred during normal operation. Commissioning, which absorbs a large share of attention in contract negotiation and in insurance wordings, accounts for 50 entries, or 7%.

Severity across all entries: 458 significant, 160 minor, 87 major, 8 near-miss.

Finding 04The map shows where reporting happens, not where risk is

United States accounts for 255 entries and United Kingdom 100. United States is not 3 times more hazardous. It reports more incidents, in English, to outlets that are indexed and searchable.

73 countries in total. 19 entries have no country stated in their sources, and are recorded as such rather than guessed.

For anyone pricing or financing assets outside the English-speaking world, this is the practical point. A clean loss history for a region may describe the reporting, not the region.

Finding 05Battery storage incidents are flat in the record

Recorded battery storage incidents by year of occurrence: 6 in 2021, 11 in 2022, 16 in 2023, 11 in 2024, 18 in 2025, 15 in 2026 so far. Deployed capacity has grown steeply over the same period.

This is the finding we hold most loosely. It is consistent with the industry’s argument that battery safety has improved, and it is equally consistent with our coverage of battery incidents being better in some years than others. It is offered as an observation, not a conclusion, and the entries behind it are all public.

Named this quarterWho the reporting named, by role

Every name below appears because a source named it in that role for an incident this quarter. Being named is not a finding of fault. The largest suppliers appear most because they have the most equipment in the field and the most reporters who know their names.

Operators

Manufacturers

The verified recordEach incident this quarter, with what sits behind it

Empty cells mean no source states it. Nothing here is inferred.

This table is the quarter’s slice of the verified dataset. The same rows export as CSV from your dashboard.

The paid recordWhat sits behind each entry

Everything in this report is drawn from the free register. Behind each entry, for subscribers, sits the structured record where a source supports it: the site, operator, owner, manufacturer, integrator and EPC; the root cause and the failed element; capacity, age at incident, downtime and loss estimate; and every source with its verification status. Team plans add CSV export of the verified dataset and this report in full each quarter. Enterprise adds the API and weekly bulk files.

That is where the questions this report raises get answered for a specific site, a specific supplier or a specific book. Three complete records are free, sent by email within a minute, so the depth can be judged before anything is paid for. Plans are on the plans page.

How much to trust itWhat each entry is built on

A record that does not say how solid it is invites the reader to guess, and readers guess generously. These are the register’s own figures for the 628 entries with a verified record behind them.

So: a count by technology or country rests on almost every entry and can be leaned on. A rate per GW rests on about a quarter of them. A cut by manufacturer rests on fewer than one in five, and by integrator on one in twenty. The register publishes those cuts, and it publishes these percentages next to them, so nobody has to find out the hard way.

MethodHow to check any figure in this report

The register is independently run and self-funded. It takes no payment from any operator, vendor or service provider in the industries it covers.