A hidden bug dormant since July combined with traffic shifts and reduced server capacity to trigger one of Microsoft's largest outages last week. The incident lasted just under 24 hours, from November 12 at 16:20 UTC to November 13 at 16:10 UTC, and affected Microsoft 365 and Exchange Online infrastructure.

Microsoft's preliminary Post-Incident Report identifies three contributing factors. A defect deployed to the cloud gateway in July caused CPU and memory resources to become exhausted under certain conditions. The bug went undetected in pre-release testing because test hardware was less powerful than production systems with higher CPU core counts on newer machines.

The second factor involved a traffic shift. About 15 hours before the outage, Teams traffic moved from an alternate gateway to the primary gateway where Exchange Online operates. This was a standard fail-safe procedure, but it increased load on the primary gateway with a different mix of requests. The third factor was an existing fault that left the gateway server fleet operating at 14 percent below optimal capacity, eliminating the headroom normally available to absorb demand spikes.

As these conditions combined, gateway servers froze. Backup systems that attempted to route requests away from unresponsive servers encountered the same lockouts on supposedly healthy backup gateways, creating a cascading failure. More than 80 percent of affected gateway servers froze. Mechanisms designed to isolate affected systems by region appear to have failed, though Microsoft's current report does not address why.

Recovery proved difficult because restarting servers was ineffective. When servers came back online, large request queues immediately triggered the same resource exhaustion. Services like Teams and OneDrive could be rerouted around affected infrastructure, but scaling limitations prevented Exchange Online from making a similar move. Microsoft patched and gradually returned servers to service, adding capacity and traffic-distribution guardrails. The company also extended delivery timeouts on delayed emails to prevent them from bouncing as undeliverable. Microsoft plans to publish its full final incident report on November 20.