Microsoft has traced the worldwide Microsoft 365 outage of November 12–13, 2026, to a Windows bug that overwhelmed servers handling cloud connections. Its preliminary incident report, dated November 15, describes how a software change, a shift in traffic and reduced server capacity combined to disrupt services, with Exchange Online suffering the greatest impact.

The disruption began at 4:20 PM UTC on November 12 and lasted until 4:10 PM UTC on November 13, according to cybersecuritynews.com. Most users could not reach Exchange Online mailboxes through Outlook or other clients, while many experienced delayed email.
Windows bug surfaced after months without visible problems
The defect originated in a fix prepared specifically for Microsoft 365 networking gateway servers. Deployment began in July, and the change operated globally for more than three months without visible problems before the outage.
Microsoft said neither the defect nor the fix applies to general Windows deployments. The affected software was part of its cloud infrastructure, rather than a problem with customers’ Windows systems.
Inside an internal Windows process, competing tasks were waiting to access shared resources, a condition known as lock contention. Under certain traffic conditions, that contention exhausted processor and memory capacity. Connection requests then accumulated until servers stopped responding.
Microsoft’s testing did not expose the issue because it focused on weaker machines. The fault became worse on newer hardware with more processor cores, according to the preliminary account.
Teams traffic shifted as gateway capacity fell
Two conditions brought the defect into view. About fifteen hours before the outage, Teams traffic moved from an alternative gateway to the primary gateway shared with Exchange Online.
At the same time, nearly 14% of gateway server capacity was unavailable because of an unrelated issue. Microsoft said the remaining capacity was sufficient on its own, but the combination of conditions triggered the defect.
The report therefore distinguishes the capacity loss from the software fault: the unavailable servers were affected by a separate issue, while the traffic shift and reduced capacity together exposed the problem in connection handling. The software change had already been running for months before those conditions arose.
Automatic rerouting spread failures across regions
As servers failed, automated systems moved connections to healthy endpoints. The additional work triggered the same fault on those servers, spreading failures across regions in less than two hours.
At the worst point, more than 80% of servers became unresponsive. Microsoft-hosted incident discussions also recorded that unavailable servers were placing additional load on the infrastructure still operating.
A separate failure in the boundaries intended to contain faults allowed the disruption to spread globally. Restarting servers did not provide a lasting recovery: when waiting traffic returned, the restarted machines quickly failed again.
The affected services also differed in their ability to redirect connections. Teams, SharePoint and OneDrive could move eligible traffic away from the affected path. Exchange Online could not fully switch paths at its operating scale, and it experienced the greatest impact during the disruption.
Engineers restored earlier connection handling
To recover the service, engineers developed an update that returned Windows connection handling to its earlier working state. The recovery process involved removing large groups of servers from service before applying the change.
Engineers then restarted those servers and restored traffic gradually. This approach followed the unsuccessful restarts in which machines became unresponsive again as waiting connections returned.
Microsoft also added capacity during the recovery and extended the email delivery window. That extension was intended to keep queued messages from being returned as undeliverable while the service recovered.
The incident account identifies both the connection-handling update and the staged return of servers to service as parts of the recovery, alongside the additional capacity and email delivery measures.
Repair details due November 20
Microsoft’s report remains preliminary, with repair details due November 20. Administrators can identify the disruption in service health records under incidents EX1550543 and MO1550821.
Microsoft’s published guidance said changes to tenant or device configuration were unnecessary. The Windows bug affected Microsoft’s cloud infrastructure, not customer Windows systems, and the company said the defect and the gateway-specific fix did not apply to general Windows deployments.
The preliminary findings describe both the software defect that overwhelmed connection-handling servers and the separate containment failure that allowed the disruption to spread worldwide.