365Fanatic / M365 status / Post-incident report
MO1550821 · EX1550543

Microsoft 365 outage of 12 and 13 November 2026: the final report

On 12 November 2026 Exchange Online and a long list of other Microsoft 365 services were unreachable for many customers for almost a full day. Microsoft released the final post-incident report on 20 November 2026. This page summarises it in plain language; you can download the report itself.

Final report of 20 November 2026. This is the latest version and replaces the preliminary report of 15 November; this page now follows the final report. The next steps run into 2027 and some have no date yet.

Download the report (PDF) Our status page for this incident

Microsoft's report as shown in the admin center (final Post Incident Report, 20 November 2026, PDF, 15 pages). © Microsoft Corporation. The latest version is always in the admin center.

What happened

  • Start: Thursday 12 November 2026, 16:20 UTC. End of the impact: Friday 13 November, 16:10 UTC; extended monitoring and clearing of the email backlog followed. Microsoft declared the incident resolved on Saturday 14 November at 04:15 UTC. The severity varied by region, connection method and time; access may have been partial or intermittent.
  • Exchange Online was hit hardest: most users could not reach their mailbox in Outlook, Outlook on the web, Outlook for iOS and Android, Exchange ActiveSync or through the Messaging and REST APIs. Search was degraded, integrations with Exchange failed and email was delayed for most users.
  • No mail was lost and all queued messages were delivered. Microsoft extended the window during which Exchange Online keeps retrying from 24 to 48 hours before a non-delivery report is generated. Delayed messages for consumer mailboxes that had been inactive for more than 30 days did expire, and those senders received non-delivery reports. Customer data at rest was not affected.
  • Services named: Exchange Online, Microsoft 365 Copilot, Teams, SharePoint Online, OneDrive, Defender, Purview, Fabric, Intune, Planner, Power Apps, To Do, Power BI and Universal Print. Teams, SharePoint and OneDrive were mostly hit early in the incident and recovered earlier, because their traffic moved off the affected gateway. Features that depend on Exchange Online data (search, calendar, presence, people) stayed affected for longer.
  • Examples from the report: in Copilot most prompts failed and the apps did not start; in Teams calendar, presence, Teams Rooms and desk phones (offline in the admin center) and creating or editing channels had problems; in Purview the portal was unreachable and policy changes and alerts were delayed; in Defender Safe Links and releasing or deleting quarantined mail sometimes failed and post-delivery protection (ZAP) acted more slowly; and in Universal Print print jobs could not be created or retrieved.
  • Microsoft confirms it was not a cyberattack and found no evidence of malicious activity. The change that caused the defect was not made by AI tooling. You do not need to change anything in your tenant.

Cause

According to the final report, four factors came together and caused a global outage of the networking gateway in front of many Microsoft 365 services, but mainly affecting Exchange Online. Each factor is normal on its own; together they exposed a latent defect:

  1. A latent defect in connection processing. Under certain load, Windows servers became overwhelmed and stopped responding: inside an internal Windows process, tasks waited on each other (lock contention), which exhausted CPU and memory. The defect came from an update enabled on Microsoft 365 networking gateway servers; neither the defect nor the update applies to broader Windows deployments. Only the newer server types were hit, not the older generation: they have more cores, so more concurrent requests and more contention. The rollout began in July, went out gradually worldwide under the safe deployment procedure and ran for more than three months without impact. Normal load testing did not catch it: the traffic conditions in the test labs were different and the labs used older machines with fewer cores.
  2. Reduced server capacity. Nearly 14% of the gateway's server capacity was out of service as part of routine operations when the latest version of the service is deployed on these servers. There is enough spare capacity, so this reduction alone would not have caused impact.
  3. A Teams failover. Microsoft keeps two independent gateway systems for isolation and availability, so many services can fail over to the other one during a disruption. On 12 November at 01:07 UTC a routing issue on the gateway in the US East region caused Teams connection timeouts, after which Teams automatically failed over to the gateway that also serves Exchange Online. Shifts like this are routine and had never affected Exchange Online before. For about sixteen hours everything stayed healthy, because the gateway servers were not yet serving the daily peak of those regions.
  4. Peak load. The first failures in North America occurred around the start of the business day on the US East Coast; broad failures began at about 16:00 UTC as US traffic approached its daily peak.

Together, the peak, the shifted Teams traffic and the reduced capacity made the gateway hit the July defect. Requests built up until servers stopped responding. The gateway consists of thousands of "capacity units"; when servers fail, traffic management automatically routes traffic to other healthy units in the same region. That happens often and is normally what makes the service resilient, but now the rest took extra load and the same defect struck there: a cascade. In less than two hours a large part of the North American servers was in trouble; at the lowest point more than 80% were unresponsive. When a region ran out of healthy capacity, traffic management sent the traffic to other regions, which had the same defect, and within about two hours of the start in North America servers in every region were affected.

First step in recovery: moving the traffic of Teams, OneDrive, SharePoint and the Office apps to the redundant gateway, which relieved those services and let them recover earlier. After that, Exchange Online took longer than usual to recover. Servers coming back were immediately overwhelmed by the waiting traffic and failed again; the automated recovery was not designed for such widespread failure with sustained reconnection demand. Because the defect came from a change deployed months earlier, finding the cause took longer and the usual safeguards (stopping a rollout, rolling back or switching off a new feature) could not revert it quickly. And because the defect was already across the fleet, recovery had to be run for each capacity unit around the globe.

How it was fixed

  • Fixing one gateway server took five steps: (1) take the server out of service so it takes no load when restarted, (2) restart it from its failed, unresponsive state, (3) apply the patch, (4) restart it a second time for the patch to be active and (5) put it back into service.
  • Microsoft built an update that reverted the defective change in connection handling and applied it to part of the capacity at 07:20 UTC on 13 November. It was then expanded in groups, first to regions in or entering peak hours.
  • Large groups of capacity units were moved out of traffic rotation, all unresponsive servers in them were recovered at the same time and the group was returned to service. That was then expanded across all regions. In addition, traffic was moved to alternative connection paths, capacity was added and the email delivery retry window was extended from 24 to 48 hours.

Timeline

  • July 2026: The update for the gateway servers starts to roll out in July and spreads gradually worldwide, for more than three months without problems.
  • Thu 12 Nov, 00:25 - 01:07: Teams users in US East get timeouts through the alternative gateway; traffic is automatically failed over to the gateway shared with Exchange Online. Load rises gradually until 16:00; isolated server failures begin in some regions.
  • Thu 12 Nov, 16:03 - 16:20: Monitoring sees more connectivity problems (16:03) and automation starts to redirect traffic (16:17). At 16:20 the alert for Exchange Online and the notice EX1550543 follow. Start of the incident.
  • Thu 12 Nov, 16:25 - 16:33: Confirmed: problems across multiple Exchange connection methods (Outlook, Outlook on the web, ActiveSync, APIs). The first support case arrives and complaints appear on sites such as DownDetector.
  • Thu 12 Nov, 17:03 - 18:00: The outage grows and spreads across multiple regions despite the automatic redirection. Microsoft finds errors on the part that receives access requests (17:45) and starts restarting parts step by step (18:00).
  • Thu 12 Nov, 18:27 - 19:20: Notice MO1550821 for the other affected services, including Teams, Copilot and Defender (18:27). At 19:00 Microsoft widens the response and pauses all deployments and configuration changes. At 19:20 it is clear the high load is not caused by more traffic, but by many servers being unexpectedly unavailable.
  • Thu 12 Nov, 20:00 - 21:00: A targeted change bypasses components that cause failed health checks (20:00). While validating it, many servers turn out to be unresponsive and restarts begin (20:30). An alternative health check works on a subset of servers and is applied more broadly (21:00).
  • Thu 12 Nov, 22:00 - 23:55: The first recovery actions turn out not to work. Microsoft widens the investigation, looks at whether isolating traffic per region helps and introduces protective measures in one location (23:00). Servers that come back fail again once they receive traffic (23:55).
  • Fri 13 Nov, 00:30 - 01:15: Connection negotiation is identified as the bottleneck: a process on the servers is overwhelmed and stops processing requests (00:30 - 01:00). Measures to reduce the pressure help partly (01:15).
  • Fri 13 Nov, 01:53 - 05:00: Microsoft turns off part of the automatic redirection, adds capacity, adjusts traffic protocols and blocks traffic in a targeted way. That gives partial recovery, but not a lasting one; machines get stuck in bad states and extra virtual capacity is scaled up.
  • Fri 13 Nov, 07:20: Mitigation found and applied to part of the capacity: an update that reverts the defective change.
  • Fri 13 Nov, 08:00 - 11:35: The update is rolled out step by step, first to regions in or entering peak hours. The service begins to stabilise (10:00), Asia Pacific is recovered around 11:00 and Europe at 11:35.
  • Fri 13 Nov, 13:00 - 15:54: Recovery is sustained across all regions and the queues of delayed email shrink (13:00). Webmail is back at expected levels globally (14:17). Customer traffic has been moved to alternative connection paths (15:05) and most major access methods are back at normal levels (15:54).
  • Fri 13 Nov, 16:10: No more active impact. Extended monitoring and clearing of the email backlog follow. End of the impact.
  • Sat 14 Nov, 04:15: After extended monitoring: final updates to EX1550543 and MO1550821; the incident is resolved.
  • 15 Nov: Microsoft releases the preliminary post-incident report.
  • 20 Nov: Date of the final post-incident report, with the full list of repair items and next steps.

All times are UTC and come from Microsoft's report (shortened). Dutch time is UTC+1 in November.

Repair items

  • Developed an update that reverted the connection handling change that introduced the performance regression. (completed)
  • Permanently updated the Windows image to include the fix, which will be applied on all new and existing server deployments. (completed)
  • Extended the email delivery window so delayed messages were not returned as undeliverable before the incident was mitigated. (completed)
  • Added additional capacity as a proactive measure to help with recovery. (completed)
  • Additional monitoring for traffic-redirection patterns and connection-handling health. (completed)
  • Implemented temporary guardrails to avoid automatic traffic distribution and avoid cascading impact. (completed)
  • Returned servers from maintenance. (completed)

Next steps

  • Traffic distribution design can cascade failures to other regions. Fix traffic distribution to eliminate such cascaded failures to other regions. (CY26 Q4)
  • Regional failure zones are too large. Design traffic distribution for smaller fault zones. (TBD)
  • Teams failover contributed to exposing the defect impacting mail flow and Outlook connectivity. Partition the network gateway capacity into pools that serve different workloads to isolate failures. (CY27 Q1)
  • A routing issue in one region moved all Teams traffic worldwide to the primary network gateway. Teams traffic fails over only for the affected region and not worldwide. (CY26 Q4)
  • Failover testing did not identify the defect in isolation. Scoped failover testing to cover peak conditions and other changes during rollout as part of safe deployment. (TBD)
  • Change auditing and diagnostics processes took longer to identify the Windows change as the root cause. Update diagnostics collection and change auditing for Windows changes. (CY27 Q1)
  • Distributing and validating the corrective configuration across the global fleet took too long for timely recovery. Improve recovery procedures for fleet-wide failures. (CY27 Q1)
  • Servers were re-overwhelmed on return to service. Add gradual traffic ramp-up and admission control when servers rejoin; build controls to reduce traffic by protocol, workload or region during recovery; improve retry and back-off behavior within the service. (TBD)
  • Service retries may spread the issues to other capacity units. Improved implementation of retry and back-off behavior with other services. (TBD)
  • Testing did not adequately represent the combination of hardware and traffic mix encountered during this incident. Expand load testing to the newest, highest-core hardware with production-representative traffic, including regular tests at peak-load limits. (CY27 Q1)
  • The network gateways did not successfully manage a defect in combination with other factors creating impact at peak traffic conditions. Add additional capacity for more headroom. (CY27 Q1)
  • Some capacity units had reduced server capacity (part of routine maintenance), which can be risky in combination with other factors. Detection for extended capacity loss that risks running out of headroom. (CY26 Q4)
  • First customer notification came about 20 minutes after broad impact. Improve detection and automated notification to reduce time to communicate. (CY26 Q4)

Microsoft's pages

Source

This summary is by 365Fanatic and follows Microsoft's post-incident report of 20 November 2026. The wording is ours; Microsoft's report is authoritative for the exact text.

Last updated on 21 november 2026.