Cut-paper illustration of a cloud above a laptop and a row of envelopes, with the cable between them cleanly cut in two

On Thursday morning, November 12, Exchange Online mailboxes started refusing to open. Outlook on the web, desktop Outlook, phones, it did not matter. By Friday the outage had spread to 74 countries and territories, and it took Microsoft until the early hours of Saturday, November 14, to call it restored. About 30 hours, by Microsoft's own record.

The cause was not an attack, and it was not AI. Microsoft's preliminary explanation is one sentence long: an issue during an earlier network capacity upgrade surfaced after a more recent network change. The failure itself was ordinary. What should hold your attention is how long recovery took, and what that says about the plan you will reach for the next time your email goes away.

What happened

Microsoft's Service Health record puts the start of the Exchange Online incident, EX1550543, at 10:05 UTC on November 12. Its first public post came at 17:54 UTC that day, saying it was investigating mailbox access problems in the Americas. The wider Microsoft 365 incident, MO1550821, followed.

What stopped working was broad. Users could not reach Exchange Online mailboxes through Outlook on the web, MAPI, Exchange ActiveSync, or the REST API. Everything that leans on a mailbox or a calendar went with it: scheduling and creating chats in Teams, Copilot, SharePoint, Planner, To Do, Power Apps, and parts of Defender and Intune. Microsoft's record lists 20 affected services. Exchange Server running on premises was not affected.

Downdetector logged roughly 130,000 to 140,000 Outlook problem reports at the peak. Hospitals, government offices, and airlines were among the organizations pushed back onto phone lines and paper. Microsoft posted its final status update at 04:17 UTC on November 14, marking both incidents service restored, and its status account confirmed restoration for the last customers and regions around 06:09 UTC.

The cause, in Microsoft's own words

Here is the preliminary root cause as it appears in the MO1550821 record:

"The incident was introduced by an issue during an earlier network capacity upgrade that surfaced following a recent network change."

The same entry adds: "This was a technical issue not related to cyber or AI / agents." Microsoft has committed to a preliminary post-incident report within two business days and a final report within five. As of this writing, neither has appeared, so treat everything in this post as the preliminary account, not the finished one.

The mechanism, as described in Microsoft's status posts, was that a number of servers became unexpectedly unavailable and started failing automated health checks, which put elevated load on the servers that were left. Microsoft rolled back a recent network configuration change. Restored servers then went unresponsive again once they took traffic, which is why the outage dragged on long after the rollback.

One report to treat with care: The Verge, citing sources, linked the outage to configuration changes in Azure virtual networks. Microsoft has not confirmed that, and its incident record does not list Azure as an affected service. Until the post-incident report says otherwise, that is an unconfirmed theory, not a fact.

The durable point: recovery is the outage

Two failures stacked. An earlier upgrade introduced a defect that sat unnoticed, and a later change woke it up. The rollback was the easy part. Getting servers healthy again under real load was not, and that is what turned a bad morning into a 30 hour outage.

This was not Microsoft's first network-change incident this year. On July 23, 2026, Microsoft said a bug in its automated network maintenance system removed IP routes from more devices than intended, affecting Azure and Microsoft 365 in the West US region until the change was rolled back. Same shape: a change, a latent defect, a slow recovery.

For anyone running a tenant, the lesson is plain. Your vendor's recovery speed is a number you do not control, and your plan has to assume a day, not an hour. What you do control is what your users do meanwhile, how they hear from you, and whether the mail that was in flight comes back or quietly disappears. Those are the things worth fixing this week, while the details are still easy to find.

What to do this week

  1. Warn your users about the phishing. Microsoft Threat Intelligence has reported phishing messages impersonating Microsoft 365 service health notifications, citing the real incident number EX1550543. If a service health email asks anyone to click through to anything, it is not from Microsoft. Send that warning now, while people still remember the outage.

  2. Check your own tenant. Microsoft says delayed mail was processed by November 14, but trust that the way you trust any vendor statement: verify it. Run a message trace for November 12 to 13 and confirm nothing bounced or vanished. Delayed mail is the kind of problem nobody notices until a customer asks where their invoice went.

  3. File the SLA claim if you qualify. Exchange Online promises 99.9 percent monthly uptime. Below 99.9 percent earns a 25 percent credit, below 99 percent earns 50 percent, and below 95 percent earns 100 percent, and each applies only to the affected service's share of the fee, not the whole bill. A tenant whose users were down the full 30 hours lands around 95.8 percent uptime, in the 50 percent tier. Microsoft does not apply credits automatically, and claims for this outage are due by the end of December 2026. Two caveats: the SLA counts downtime by whether users could send or receive in Outlook on the web, so check your tenant's actual impact before filing, and consumer plans have no SLA credit at all.

  4. Decide your out of band channel now. When email and Teams are both down, how do you tell 200 people what is happening? A status page, a phone tree, a group chat on a different platform. Write it down while the memory is fresh, and make sure it does not depend on the service that just failed.

Sources

The outage is over. The full report is still coming, and it may change the details. What will not change is the list above. Do the four things while the outage is still fresh, and the next one costs you less.