Summary
- The current 12 October register page contains 92 entries across 27 countries.
- Its incident descriptions include 52 fires or thermal events and 17 incidents classed as major.
- The database is a reported-incident sample, not a statistically representative failure rate.
Energy Incident Register has published a 12 October snapshot of its data centre power-system incidents, recording 92 publicly reported cases across 27 countries between 2010 and 2026. The register identifies 52 as fires or thermal events, with 17 classified as major. It also reports that UPS systems and battery rooms are the most frequently named equipment category, appearing in 22 entries, followed by backup generators in 15 cases.
These numbers require careful interpretation. The register gathers incidents that have appeared in public reporting and checks them against cited sources. It does not measure the population of all data centres or every failure that occurred at each site. Without a denominator of installed facilities, equipment hours, power capacity and reporting completeness, the figures cannot establish that a particular operator or technology has become more dangerous over time. More entries may reflect increased infrastructure deployment or more comprehensive reporting as well as actual changes in risk.
The register’s current page states that 46 of the 92 cases are dated 2024 or later and reports 14 incidents in the United Kingdom. A contemporaneous secondary report gave slightly different overall totals, apparently reflecting a changing dataset. The 92-entry figure in this account is taken from the register’s specifically dated incident page rather than blended with the different total in secondary reporting. Both sources underscore that the dataset is being expanded, so comparisons should state the snapshot used and avoid presenting the aggregate as a permanent industry-wide total.
UPS systems and batteries are essential to continuity. They can provide immediate power when a primary electrical feed falls outside operating limits, allowing automatic transfer or supporting an orderly response. The same assemblies contain stored electrical energy and require protection against overcurrent, thermal excursions and faults in the charging and switching equipment. An incident entry naming a UPS room does not, by itself, identify which component initiated a fire or whether battery chemistry was causal.
Standby generators are the second most frequently named equipment group in the register. Their role is different from that of a UPS: they support longer interruptions but normally require a start and transfer sequence. Fuel handling, exhaust, cooling, lubrication and control systems add maintenance requirements. Incident lessons therefore extend beyond installed generator capacity to the adequacy of testing, preventive maintenance, segregation and the ability to recover from a failed start or transfer event.
The absence of a named failed component in many reports is itself a constraint on incident analysis. A news article may record a service interruption or emergency response without publishing a forensic finding. It would be inappropriate to assign a root cause, infer a manufacturer or use an incident to make a broader claim about lithium-ion batteries when the reporting does not establish one. The register explicitly distinguishes public evidence from details that remain unknown.
The geographic breakdown deserves similar caution. The United Kingdom and United States have highly developed data centre markets and a substantial body of accessible reporting. Countries with fewer reported incidents might have smaller installed fleets, weaker media coverage or different disclosure practices. Ranking countries by a simple incident count would confuse reporting volume with an engineering probability of failure. Insurers or operators making comparisons need exposure data, standard definitions and a consistent observation period.
The database groups together events with different operational and safety consequences. A controlled power-path interruption is not equivalent to a destructive battery-room fire, and an electrical injury has different prevention measures from a generator fuel spill. The register's value is that readers can inspect individual cited records rather than assuming all 92 events describe the same severity. Engineers should use the underlying narratives to identify applicable controls and separately judge whether the recorded conditions resemble their own sites.
The register also demonstrates the limits of publicly available root-cause information. If a component has not been identified, it may be because an investigation is ongoing or because detailed findings were never released. This should encourage cautious attribution in incident coverage. For example, battery fires are sometimes discussed as a single category despite differences in chemistry, enclosure design, charging controls and the initiating fault. The dataset can support questions for site audits, but it cannot replace those site-specific assessments.
For engineering teams, the practical value is a catalogue of failure modes to examine during hazard analysis, maintenance planning and integrated systems testing. It can prompt questions about cable segregation, thermal monitoring, generator testing, battery isolation and site emergency procedures. Its numbers are a documented collection of reported events, not an actuarial forecast. Any comparative risk analysis would require more complete exposure and operating data.
