FULL STORY

Microsoft 365 Global Outage: From Crash to Apology

Microsoft 365 suffered a rare global outage lasting over 15 hours, with Outlook and Exchange Online inaccessible worldwide. Microsoft later apologized, attributing the failure to a network capacity upgrade rather than attacks or AI.

2026-11-13 ~ 2026-11-14 · 2 episodes · 70 posts

Episode 1 · Microsoft 365 global outage drags past 15 hours, hitting Outlook and Teams (2026-11-13, 66 posts)

On November 12 (starting around 16:17 UTC), Microsoft suffered one of its largest service disruptions in years: Outlook.com and Exchange Online became inaccessible worldwide, with collateral impact on Teams, SharePoint Online, Microsoft 365 Copilot, Defender XDR, and OneDrive — even Microsoft's own employees couldn't send or receive email. By around 1:30 PM ET, Downdetector had logged more than 120,000 user reports of Outlook being down, peaking at roughly 130,000; Microsoft 365 saw over 16,000 affected users and Copilot nearly 6,000; within an hour, outage reports spanned 74 countries and regions, with the five affected services peaking at 19,650 reports per minute (versus a typical average of fewer than 15).

Confirmed

  • Microsoft officially confirmed the root cause: a core process on affected servers was overwhelmed, leaving large numbers of servers unexpectedly unavailable and failing health checks; the load shifted to the remaining servers, triggering a cascading failure. Microsoft isolated parts of its infrastructure, restarted affected components, and gradually restored traffic to validate stability, while pausing all new deployments.
  • An incident advisory (MO1550821) was posted to the M365 admin center, with EX1550543 as the Americas-region Exchange Online incident ID; during remediation, the mail timeout limit was temporarily extended to 48 hours.
  • The outage lasted more than six hours: Kentik traffic data shows it began around 17:35 UTC, and 6.5 hours later Exchange Online traffic was still about 90% below normal.
  • The BBC reported that Microsoft admitted recovery was progressing slower than expected and that services were unstable early in the restoration: officials said recovering servers could become unresponsive again once they resumed taking traffic, prompting temporary rate limiting on top of existing safeguards for troubleshooting.

Unconfirmed

  • The Verge's Tom Warren, citing sources, reported the outage was linked to an Azure VNet (virtual network) configuration change; other leaks claimed a release/deployment drove CPU usage to spike and put racks into maintenance mode. Microsoft has not officially confirmed these specific triggers.
  • Engineer Luke Elin, with 35 years of systems experience, offered a root-cause hypothesis from an engineering perspective that was widely reshared and endorsed by industry insiders: traffic didn't spike — capacity dropped. After servers failed health checks and were removed from the cluster, the process for establishing secure connections overloaded, creating a "restart storm," leaving the cloud service in a metastable state unable to absorb such shocks.

Why it matters

  • justinhart did the math: under a "four nines" (99.99%) availability commitment, only 52 minutes of downtime are allowed per year — burned through in a single morning; the "three nines" annual allowance of 8.76 hours was exhausted that same night, directly calling into question the credibility of Microsoft's enterprise SLAs.
  • The recovery process itself exposed systemic fragility: per CantaloupeNews, Microsoft at one point said it struggled to push the fix forward because engineers were flooded with Authenticator verification pushes — the overloaded verification system in turn stalled the repair effort.
  • One developer (esrtweet) seized on Microsoft's statement that recovered servers might become unresponsive again upon receiving traffic to mock its capacity planning; gary marcus called it a very "2025" kind of failure. Given the event's impact on enterprise email — a core communications infrastructure worldwide — the post-mortem is well worth studying.

46 more related posts →

Episode 2 · Microsoft Apologizes for Global Outage, Blames Network Upgrade Not AI (2026-11-14, 4 posts)

Microsoft apologized for a widespread global outage, attributing it to a network capacity upgrade issue exposed by recent changes, not a cyberattack or AI. Many users remain skeptical, questioning whether AI-generated code was the real cause.