Test report DSG-1580 · Rev E · tested October 12, 2026

AI Datacenter InfrastructureDevice under test

Data Center Fires and Power Failures Have Doubled Since 2024

Data center fires and power failures rose from eight incidents in 2024 to 20 in 2025, with UPS and battery rooms the most common failure point, register data shows.

Read
4 min
Words
757
Node
3nm
Operator
Elena Vasquez

Spec summary

  1. Data center fires and power failures rose from 8 incidents in 2024 to 20 in 2025 and 19 in 2026.
  2. 47 of 93 recorded incidents across 2010–2026 occurred in 2024 or later.
  3. UPS and battery rooms were the most common named failure point, with 22 of 93 incidents.
  4. The US logged 33 incidents over the period; the UK logged 14.
  5. 52 of 93 incidents were fires or thermal events, 18 classified as major.

Data center fires and power failures more than doubled between 2024 and 2025, with the annual count rising from eight incidents to 20, according to newly published incident data.

The figures come from the Energy Incident Register, an independently run public register of safety incidents in energy and grid infrastructure. Its records show the sector recorded 19 incidents in 2026, keeping the rate near the elevated 2025 level rather than reverting to the historical baseline.

The register now tracks data center incidents from 2010 to 2026 across 26 countries, drawing on public reporting. Of the 93 data center incidents logged over the full period, 47 occurred in 2024 or later — meaning roughly half of all recorded incidents in 16 years took place in the most recent three.

Where are the incidents concentrated?

The United States accounts for the largest share, with 33 incidents across the 16-year period. The register attributes this to the country's market share rather than to a disproportionate failure rate; the US hosts the world's largest data center fleet.

The UK recorded the next highest incidence rate, with 14 incidents across the period. Together, the two markets account for exactly half of all logged incidents — 47 of 93.

The register does not break down the remaining incidents by country in the published data, but the 26-country coverage indicates incidents are spread across major hosting markets rather than clustered in one region.

What is failing?

Fires and thermal events dominate the incident profile. Some 52 of the 93 incidents were fires or thermal events, and 18 of those were classified as major.

The failure data points to specific weak points in facility power chains:

  • Uninterruptible Power Systems (UPS) and battery rooms: 22 incidents, the most common named failure location
  • Backup generators: 15 incidents, the second most common
  • No part named: 53 incidents, where public reporting did not identify the failing component

That last number is significant for anyone reading the data for engineering conclusions. In the majority of incidents — 57% — no component was named in the reporting. The register's component-level figures therefore cover only the 40 incidents where a part was identified.

Within that identified subset, UPS and battery rooms account for more than half. That distribution is consistent with lithium-ion battery installations and the thermal runaway risks they carry becoming a larger share of data center power infrastructure as operators densify facilities.

What happened this year?

Two recent incidents illustrate the scale of the events now entering the register.

In September, a fire broke out at a data center in Gasan-dong, Seoul, South Korea. In June, a fire hit a Tata Communications center in India.

Both incidents drew coverage from trade press, and both are the type of publicly reported events the Energy Incident Register captures. The register's methodology is based on public reporting, which means the true incident count depends on disclosure — operators are not required to report failures to the register, and incidents that never become public do not appear in the data.

Why does the doubling matter?

The timing of the increase coincides with the fastest capacity expansion in the sector's history. Data center construction has accelerated since 2023, driven by AI training and inference workloads, and the incident data suggests the growth phase is adding risk.

The 2024-or-later cluster — 47 of 93 incidents — maps onto the period when operators were commissioning new capacity at record rates, often under compressed construction schedules.

The component data reinforces that reading. New large-scale facilities rely heavily on battery-based UPS systems to bridge grid power and generator startup. With 22 of the 40 identified failures occurring in UPS or battery rooms, the part of the facility that has grown most during the buildout is also the part failing most often.

The register does not attribute causes, and the data alone cannot establish whether the increase stems from more facilities, faster commissioning, battery chemistry changes, or reporting effects. What it does establish is the trend line: eight incidents in 2024, 20 in 2025, 19 in 2026.

For operators, the practical takeaway from the register is narrower: the two most commonly identified failure points — battery rooms and backup generators — are both maintainable, inspectable systems. The 53 incidents with no named component also point to a disclosure gap that limits how much the industry can learn from its own failures.

The Energy Incident Register continues to accept and track publicly reported incidents across its 26-country coverage, and the dataset spans 2010 through 2026.

via energyincidentregister.com (Original)

Filed under

  • data-center-incidents
  • power-infrastructure
  • ups-systems
  • facility-safety
  • data-center-reliability
Share this article:

More from Elena Vasquez

Elena Vasquez

Show full bio

Senior reporter covering people, launches and day-to-day practice in hardware layer of artificial intelligence at Die Signal.

284 articles

Same lot · LOT-C1C8

« Previous articleNext article »