Transformer failure and control fault cause Azure data centre power loss
Feb 2026Published 12 Oct 2026Responsible editor: Chief Editor
Get the next one by email
Free alerts on data centre incidents in the United States, and an email when EIR-2254 is updated. No account needed; unsubscribe in one click.
Get a free sample brief Three complete verified records and the incidents-per-GW table, by email within a minute.
By signing up you agree to the privacy policy.
Companies named in this entry: Microsoft
What happened
On 7 February 2026, one data centre within Microsoft's West US cloud region in California, United States, lost power. According to a post incident report that Microsoft published, an electrical failure in an on-site transformer caused the loss of utility power to the data centre, although the utility itself remained functional. Generators started as designed, but a cascading failure within a control system prevented the automated transfer of load from utility power to generator power.
The uninterruptible power supply batteries carried the load for several minutes until they were fully depleted, and customer impact began at 07:58 UTC. Power returned to around 90 percent of IT racks by 09:31 UTC, and the site was running fully on generators by 11:29 UTC. Some equipment needed further electrical control system troubleshooting before power could be restored. Six storage scale units lost power. Four recovered quickly, but two had prolonged recovery, which delayed restoration of dependent compute and platform services. Service impact ended at 04:24 UTC on 8 February, and the site returned to utility power at 03:42 UTC on 9 February.
Contributing factors
A cascading failure in a control system prevented automatic load transfer to generators even though the generators started. Dependencies of compute and platform services on the storage scale units slowed overall service restoration. The source does not explain why the on-site transformer failed.
Lessons learned
- Test automatic transfer schemes end to end under load, including control logic and failover to generators, not only whether generators start.
- Size UPS battery runtime with a clear margin over the worst-case time to transfer to generators, and alarm when that margin is at risk.
- Map the dependencies of storage and compute platforms on power sequencing so that recovery after a power loss is staged and prioritised.
More lessons on detection and monitoring, from every incident on the register.
Source
Other entries that name Azure
Published 2026-10-12. Written from public reporting. Descriptive, not investigative. See the data accuracy disclaimer.
