
Short version: on Thursday, November 12 and Friday, November 13, 2026, Outlook and Exchange Online were down for most users for almost 24 hours, from 11:20 a.m. Thursday to 11:10 a.m. Friday (Eastern Time). It was not an attack. According to Microsoft's preliminary report, a defect that had been sitting unnoticed since July in the servers that receive Microsoft 365 connections was triggered by a combination of more load and fewer available servers, and the automatic recovery systems themselves spread it worldwide in under two hours. Mail sent to you was not lost: it waited in a queue, and Microsoft extended the window before returning it to 48 hours. Below is the full explanation and what to prepare for next time.
What went down, and for how long
The first official notice went out at 16:20 UTC on Thursday the 12th (11:20 a.m. Eastern), under ID EX1550543: users could not reach their Exchange Online mailboxes "through multiple connection methods". That covered Outlook desktop, Outlook on the web, mail on phones (Exchange ActiveSync), and apps that read mail through the API.
Two hours later Microsoft opened a second incident, MO1550821, for the rest of Microsoft 365. The final list of affected services in the report is long: Exchange, Copilot, Teams, SharePoint, OneDrive, Defender, Purview, Fabric, Intune, Planner, Power Apps, To Do, Power BI and Universal Print.
But not everything failed the same way. The report separates two levels:
- Primary impact: most users could not open their mailbox, and mail delivery was delayed for many.
- Secondary impact: in the other services, only what depends on Exchange Online failed. For example, the Teams calendar, or Copilot when it had to read your mail. Teams chats and meetings worked without issue for most of the outage, and SharePoint and OneDrive files could still be opened.
By around 1:30 p.m. Eastern, Downdetector had more than 120,000 Outlook reports, according to NBC News and the AP. Microsoft declared the impact over at 16:10 UTC on Friday the 13th (11:10 a.m. Eastern) and closed the incidents that night, after watching the backlog of delayed mail drain.
The cause: three things at once
Microsoft released a preliminary Post Incident Report (PIR) on November 15. Admins of any affected tenant can download it. According to that document, the outage did not have one cause but three, and none of them would have been enough on its own.
1. A latent defect from July
In front of Exchange Online and many other services sits a layer of servers Microsoft calls the networking gateway: they receive your connection and pass it to the server that holds your mailbox. Those servers run Windows.
In July, Microsoft deployed a change built specifically for that gateway. The report stresses that neither the defect nor the fix applies to other Windows deployments. The change went through its staged rollout, reached the whole world over a few weeks, and ran without problems for more than three months.
But it had a defect: under certain load conditions, threads inside an internal Windows process ended up waiting on the same resource (lock contention, in programming terms), and the machine ran out of CPU and memory until it stopped responding.
Load testing missed it for a telling reason: the problem gets worse on servers with more cores and newer hardware, and Microsoft's labs tested mostly on weaker hardware with fewer cores and a different traffic mix.
2. More load than usual: Teams moved onto the same gateway
Microsoft runs two independent gateways, precisely so a service can move from one to the other if one fails. About fifteen hours before the outage, at 01:07 UTC on Thursday, the automatic system moved Teams traffic from the alternative gateway to the primary one, the same one Exchange Online uses.
That move is routine. But it raised the load on the primary gateway and changed the kind of traffic it received. According to the report, between 01:07 and 16:00 UTC the load climbed gradually, and isolated server failures already showed up in some regions, which automation kept handling.
3. Fewer servers than usual
At the same time, almost 14% of the gateway's servers were out of service because of an unrelated issue from the day before. They were due back through normal maintenance, and there was plenty of headroom, so on its own it was not a problem.
The domino effect
With more load and fewer servers, the gateway started hitting the July defect right at the US peak hour. And this is the most interesting part of the report: what normally makes the service resilient is what took it down.
When a server stops responding, the traffic management system sends its users to other healthy servers. It happens every day without anyone noticing. This time, the redirected traffic loaded up the healthy servers, which then hit the same defect, and their traffic was spread across even fewer servers. In under two hours the failure had run through the whole fleet. At its worst, more than 80% of the servers were unresponsive.
Normally that capacity is split into thousands of units grouped by region, kept apart so a failure in one does not jump to another. This time, because of a separate issue the preliminary report does not detail, that separation did not stop the chain, and the failure went global.
Teams, OneDrive, SharePoint and the Office apps were able to leave the broken gateway and move to the other one. That is why they recovered earlier. Exchange Online cannot do that because of its scale, and it was stuck.
Why it took almost a day to fix
In a normal outage you roll back the last change and you are done. That was not possible here, for three reasons the report itself gives:
- The faulty change was four months old. The usual fast rollback tools (reverting the deployment, switching the feature off) did not work for something deployed in July.
- Recovered servers went down again. As soon as one came back into service, it received all the waiting traffic at once and hung again. Microsoft saw it live: at 23:55 UTC on Thursday it noted that returned servers "initially remained healthy" but stopped responding once they received traffic.
- Automatic recovery was not designed for this. It is built to restart individual servers, not to bring back a fleet that is almost entirely down. In fact, at around 01:53 UTC on Friday Microsoft switched off part of the automatic traffic redirection.
The public updates from those hours show the search. Microsoft first pointed to a recent change, then said it had found no change that lined up with the start, then mentioned a small number of configuration updates, and at around 01:00 UTC on Friday it found the specific point: the process that sets up connections on each server was getting overwhelmed.
The fix was an update that returned that Windows code to its previous version. It was first applied to part of the servers at 07:20 UTC on Friday, and then in groups: take servers out of service, apply the change, restart them and put them back, starting with the regions entering business hours. Asia Pacific recovered around 11:00 UTC, Europe at 11:35, worldwide webmail at 14:17, and at 16:10 UTC Microsoft declared the impact over.
In one of its Friday updates, Microsoft wrote: "We are deeply sorry for the disruption this incident has caused you and your business."
What happened to email
It was the question everyone who manages client email was asking: was anything lost?
According to Microsoft, no. Inbound mail was delayed, but it stayed in a queue. At 09:55 UTC on Friday Microsoft announced it was extending the window before returning a message as undeliverable to 48 hours, precisely so delayed mail would not bounce. The report lists this among the completed actions: the extended window made sure every message queued during the outage was processed.
One detail worth understanding: access came back before mail did. On Friday afternoon, Microsoft warned that users could get into their mailbox before message delivery caught up. If a client told you "I can log in now but I'm missing this morning's email", that was it.
And if you run a hybrid setup, with your own Exchange server connected to Exchange Online, Microsoft warned that mail flow between the two could also be affected.
For mail coming from outside, the standard helps too. RFC 5321, which defines how email is sent, requires a server that cannot deliver a message to keep it and retry, at intervals of at least 30 minutes, and recommends not giving up for at least 4 to 5 days. A destination server that fails for a full day normally does not cost anyone writing from another provider their message.
What a small business or MSP should check
None of this could have been prevented from the customer side. But there is a big difference between finding out on Downdetector and having it under control.
Know where to look, especially when the admin center won't load either
The official source is Service health in the Microsoft 365 admin center (Health > Service health). Every incident shows up there with its ID (EX for Exchange, MO for Microsoft 365 in general, TM for Teams, and so on) and all of its updates.
During this outage, the admin center itself would not load at times. For that case, Microsoft's documentation gives two alternatives: the public page status.cloud.microsoft and the @MSFT365Status account on X.
Turn on alerts before you need them
Under Service health > Customize > Email you can get new incidents by email, up to two addresses per admin. An obvious detail that is easy to miss: if the alert goes to an Exchange Online mailbox, in an outage like this one it will not reach you. Use an external address. The Microsoft 365 Admin phone app also sends notifications.
Download the post-incident report in time
The dashboard history only shows incidents from the last 7 or 30 days. The preliminary report for this outage came out on November 15 and Microsoft says the final one is due on the 20th. If you manage clients, download it and keep it: it is what you will need if someone asks why they had no email for a whole day, or if it is time to file an SLA claim.
Have a plan to communicate without Outlook
That day, Teams kept working for almost everyone. It will not always be that way, but it makes the point: decide ahead of time which channel you will use to reach your users or clients if email goes down, and make sure it does not depend on email.
What this outage leaves behind
For anyone who runs systems, the report is a lesson that applies well beyond Microsoft:
- A change can pass every test and fail months later. The July change passed the staged rollout and three months in production. Testing used weaker hardware than production, and the defect only showed up on machines with many cores.
- Redundancy that shares a dependency is not full redundancy. There were two gateways, but Exchange could only use one.
- Automatic recovery can make an outage worse. Moving traffic to healthy servers is right when one fails. When the failure is caused by load, moving traffic just spreads the problem.
- Rolling back is only easy when the change is recent. Four months later, there was no undo button.
Microsoft says it has already added capacity, more monitoring of traffic redirection and limits so traffic distribution cannot trigger cascading failures. The full list of fixes will come with the final report, and we will update this article when it is out.
Sources
- Microsoft, Preliminary Post Incident Report, EX1550543/MO1550821, November 15, 2026. Available to admins in the Microsoft 365 admin center.
- Updates to incidents EX1550543 and MO1550821 on the Microsoft 365 service health dashboard, as mirrored by the University of Pennsylvania.
- Microsoft Learn: How to check Microsoft 365 service health
- RFC 5321, section 4.5.4.1
- NBC News / AP