Data Center Fires and Power Failures Double Since 2024
Rising thermal events and power system faults highlight infrastructure stress as AI power demands scale up.
New data from the Energy Incident Register reveals that data center fires and power failures have more than doubled since 2024. As AI workloads push electrical infrastructure and high-density rack power requirements to unprecedented levels, critical facility outages and hardware faults are escalating globally.
Key details
Data tracked by the Energy Incident Register, an independent public registry monitoring safety incidents across energy and grid infrastructure, indicates a sharp increase in facility failures. The register recorded 8 fires or power failures at data centers in 2024, rising to 20 incidents in 2025 and 19 incidents in 2026. Out of 93 data center safety incidents cataloged across 26 countries between 2010 and 2026, 47 occurred in 2024 or later, showing a heavy concentration during the recent AI buildout.
Thermal safety events represent the largest category of incidents. Of the 93 total cataloged events, 52 were fires or thermal incidents, with 18 categorized as major emergencies. Component-level reporting indicates that Uninterruptible Power Supply (UPS) units and battery rooms were the most common failure points, accounting for 22 incidents. Backup diesel and gas generators represented the second most frequent source of failure, with 15 reported incidents. Geographically, the United States accounted for 33 incidents over the 16-year tracking window, followed by the United Kingdom with 14 incidents.
Why this matters
The rapid adoption of megawatt-scale AI clusters and high-density liquid-cooled server architectures has significantly increased stress on power distribution and backup hardware. Operating facilities at near-capacity with continuous heavy compute loads leaves minimal margin for electrical or thermal errors. Unplanned power outages and electrical fires not only interrupt training runs and inference availability, but also result in severe capital damage to expensive GPU servers and specialized power management components.
Context
The surge in power-related incidents mirrors broader challenges across the data center industry as operators rush to deploy high-density compute infrastructure. Fast-tracked power deployments, reliance on on-site fuel cells or peaker turbines, and continuous heavy electrical draw have put immense pressure on secondary power equipment like transformers, UPS battery rooms, and backup generators. Previous industry reports from PJM Interconnection and local utilities have highlighted similar grid instability risks and rapid load swings triggered when massive compute clusters unexpectedly disconnect.
Risks and open questions
A major open question is whether current power infrastructure and safety standards can keep pace with accelerating AI rack power densities, which now often exceed 100 kW per rack. Additionally, in 53 of the 93 cataloged incidents, official public reporting omitted the specific root cause or failing component. This lack of detailed disclosure makes it difficult for industry engineers and utility regulators to systematically address safety vulnerabilities across different hardware topologies.
What happens next
Operators and equipment manufacturers are expected to face stricter safety compliance standards and insurance scrutiny for high-density AI data centers. Facilities are increasingly evaluating new 800V DC power architectures, solid-state transformers, and advanced battery energy storage systems (BESS) to replace traditional UPS battery rooms and mitigate thermal runaway risks. Regulators and grid operators will continue monitoring large load interconnections to ensure backup power systems do not compromise local distribution grids.
Source: Data Center Dynamics Published on AI Usage Global, author: AUG Bot
