
Microsoft experienced a widespread outage affecting Microsoft 365 services from November 12 to November 14, 2026, disrupting users and businesses worldwide. The outage primarily affected Exchange Online, leaving many users unable to access Outlook, send or receive emails, or connect to their mailboxes through different devices and applications. Other services impacted included Microsoft Teams, SharePoint, OneDrive, Copilot, Defender, and additional Microsoft 365 applications that rely on Exchange infrastructure. The connectors OpenAI’s and Anthropic’s AI services use to reach Microsoft accounts were also disrupted. The disruption generated tens of thousands of user complaints, with Downdetector reporting a surge of nearly 130,000 Outlook-related reports in the United States. Some organizations, businesses, and government agencies experienced interruptions to communications and daily operations, while delayed email delivery created additional challenges.
Microsoft traced the outage to a combination of a latent software defect, increased network traffic following a Teams failover, and approximately 14% of gateway server capacity being unavailable due to a separate issue. Under the resulting load, a flaw in an internal Windows process caused servers to exhaust CPU and memory resources and become unresponsive, triggering a cascading failure across the network. Microsoft reported that more than 80% of gateway servers became unresponsive at the worst point, while automated recovery mechanisms struggled because restored servers were overwhelmed by queued traffic. The company ruled out malicious cyber activity and gradually restored services by applying a corrective update, restarting servers in stages, rerouting traffic, and adding capacity. Microsoft extended email delivery timeouts to preserve delayed messages and confirmed that services had recovered, with email backlogs subsequently cleared by November 14. The incident highlighted the risks of interconnected cloud infrastructure, cascading failures, and weaknesses in automated recovery and load-testing procedures.
Read more about it here.