Microsoft 365 disrupted worldwide for more than 20 hours: mail, Teams and Copilot hit
Since Thursday afternoon, Outlook, Teams and the rest of Microsoft 365 have been faltering in more than 70 countries. Restarting servers did not help, and the cause is not yet known.
Update · 16 November 2026
After nearly 24 hours it was over: a bug that had lain dormant since July took mail down.
By 16:10 UTC on Friday 13 November, Microsoft says, customers were no longer affected, nearly 24 hours after the outage began. It then watched the service for another night and worked through the mail that had piled up, closing both incidents at 04:15 UTC on Saturday. Because Microsoft held mail for longer, everything queued during the outage was still processed, the company says, rather than bouncing back as undeliverable.
In a first analysis dated Sunday, Microsoft explains what went wrong. The fault lay in a change to Windows that it rolled out in July to its networking gateway, the servers where traffic to Microsoft 365 comes in. For more than three months all was well. Under heavy load, though, processes on those servers ended up waiting on each other until processor and memory ran out and the server stopped responding. Load testing missed it because it ran mostly on lighter machines with fewer cores, and the fault is worst on newer hardware with many. Ordinary Windows PCs and servers are not affected, according to Microsoft.
On Thursday two things came together. Overnight, Teams had automatically switched to the gateway that also serves Exchange Online, which raised the load, and an unrelated problem had already taken almost 14 per cent of those servers out of service. During peak hours in the United States, the first servers locked up. Their traffic was automatically sent to healthy servers, which then took on too much themselves: within two hours more than 80 per cent had stopped responding, and because of yet another problem, the partitions between regions failed to contain the chain reaction.
Teams, OneDrive and SharePoint could move to a second gateway and came back sooner. Exchange Online cannot do so because of its scale, which is why mail in particular stayed down. A restarted server immediately received all the waiting traffic and fell over again, exactly as we describe in the background below. And a bug that is months old cannot be rolled back at the push of a button. Only at 07:20 UTC on Friday did Microsoft have an update that restored the old, working code, and by 11:35 UTC Europe was healthy again.
Microsoft has added capacity and put limits on how traffic is spread, so a failure is less likely to cascade. The final report is due on 20 November. Meanwhile, watch out for phishing: Microsoft’s own security researchers spotted a campaign quoting the real incident number EX1550543, claiming your mail has been held and that you must click within 24 hours to get it back. Check an outage only in the Microsoft 365 admin centre, never through a link in an email.
What strikes us: every step that made the outage bigger was meant to keep the service reliable. Teams moves away when there is trouble, traffic shifts to healthy servers by itself, and the bug sat in a carefully rolled-out fix. Those very safety nets carried the fault around the world. So our advice stands: write down, outside mail and Teams, how you reach each other and which processes depend on mail.
What happened
Since Thursday afternoon, Microsoft 365 has been stumbling around the world. Mail did not arrive, Teams meetings dropped out, and Copilot, security tools and sign-in faltered too, in more than 70 countries.
It started at 16:17 UTC on Thursday 12 November in Exchange Online, the mail service behind Outlook (incident EX1550543). At 18:26 UTC Microsoft opened a second, wider incident (MO1550821): Teams, SharePoint Online, Microsoft 365 Copilot and the security service Defender XDR were affected as well. According to TechCrunch, administrators could not reach Intune, which manages their laptops and phones, and some cloud-based Windows PCs stopped working. Microsoft 365 has more than 430 million users, according to the Dutch broadcaster NOS.
Microsoft first spoke of elevated load, but it was not driven by more traffic: the cause was “a number of servers being unexpectedly unavailable”, and later Microsoft found that “a core process on affected servers was becoming overwhelmed”. It paused all deployments, spread traffic across its worldwide infrastructure and restarted servers, which failed again as soon as they took traffic. Extra virtual machines got “stuck in a bad state”.
At 04:11 UTC on Friday Microsoft said it had “no indication this is the result of malicious activity”. At 09:03 UTC it extended its email timeout window to 48 hours so that delayed messages would not be lost. By 13:00 UTC it reported “sustained improvements across much of Europe” and progress in North America and Asia Pacific, although some users still had intermittent access problems and late mail. At 16:02 UTC Microsoft said availability was improving across a growing part of the affected systems, but it gave no end time. Microsoft has not yet published a root cause; according to NOS it will do so once the outage is over.
In the Netherlands, reports on the monitoring site Allestoringen peaked at 08:00 UTC on Friday, the start of the working day. The Dutch courts could not send or receive email and were reachable by phone only. The municipalities of Leeuwarden, Barneveld and Krimpen aan den IJssel had problems; in Krimpen residents could still apply through the website but got no confirmation email. The provinces of Groningen and Noord-Holland were harder to reach, and home workers struggled to sign in because Microsoft Authenticator was down. Meanwhile phishing emails exploiting the outage were circulating, the Dutch pharmacists’ association KNMP warned.
Actions taken thus far have not provided the anticipated relief.
Microsoft on its first attempts at a fix, as reported by The Register
The background
Why restarting did not help
A large cloud service spreads its traffic over thousands of servers. When some drop out, the rest get more work. If that overloads them too, they fail their automated health checks and drop out as well. A restarted server immediately gets its share of all the traffic that has built up, and falls over again.
On top of that, software retries by itself when something fails. AWS engineer Marc Brooker worked out that five layers of software that each retry three times hand the database at the bottom 243 times its normal load. Researchers call this a metastable failure: the trigger is gone, but the system stays stuck, busy with traffic it cannot handle.
What happens to your mail
When a mail server cannot accept a message, it replies with a temporary error code. The sender’s server then keeps the message in a queue and tries again later. How long differs per service: according to DMARCguard, 8 hours at Mailgun, 24 hours at Exchange Online itself, 72 hours at SendGrid and 5 days in Postfix’s default setting. Anything not delivered by then bounces back to the sender.
The 48-hour extension only helps mail already queued inside Microsoft, not mail still waiting on other providers’ servers, such as Gmail’s. And short-lived messages, such as one-time sign-in codes and password reset links, had often expired before they arrived.
Earlier major outages
On 22 January Microsoft 365 was down in North America for nearly ten hours. On 23 July a bug in a maintenance system put more network devices into maintenance than intended, and Microsoft 365 and Azure faltered for about five hours.
Other providers know the same pattern. In October 2025 AWS’s us-east-1 region in Northern Virginia was down for more than fourteen hours. A fault in the DNS management of its DynamoDB database was fixed within a few hours, but a system that keeps track of servers then became so congested that its work could not finish before timing out. According to AWS, the team had no established recovery procedure for that situation.
Our analysis
A fallback plan for mail and Teams should not live in mail and Teams
For most organisations Microsoft 365 is no longer a program but the office itself: one sign-in opens mail, calendars, meetings, files and phone lists, and increasingly device management and security too. So one overwhelmed process hits everything at once. If your emergency procedure sits in SharePoint and your clients’ numbers in Outlook, you cannot reach them when you need them.
Headlines speak of billions in damage, but nobody shows how that figure was worked out. The real damage lies in small things that depend on mail: a sign-in code that never arrives, a confirmation a resident never gets, a message that misses the court. You can know those processes in advance. And the first reflex, resending everything, often makes things worse: whatever sits in a queue will still arrive, and then twice.
Our expectation: days like this will not become rarer. What Microsoft describes fits a system that keeps itself overloaded, as at AWS in 2025. Switching provider does not solve that, because every big service has days like these. What does help is a plan outside the service: who calls whom, which processes stop, and what you do when the mail comes back.
What stands out
- Security went down with it Defender XDR and Intune were hit. Administrators lost their own tools just as they wanted to check whether an attack was behind it. Microsoft saw no sign of one.
- Signing in depended on it too Microsoft Authenticator was down, so home workers struggled to get in. If the second step of your sign-in lives with the same service, an outage takes your key away as well.
- Earlier this week the advice was the opposite Until Monday 9 November, a faulty spam rule made Outlook.com wrongly reject mail with a permanent error: that mail had to be sent again. This time the error is temporary, and waiting is the advice.
What it means for you
When Microsoft 365 goes down, most organisations do not lose one program: they lose the whole office.
- Write down how you reach each other Make sure the numbers of colleagues, clients and suppliers can be found outside Outlook and Teams, and that everyone knows who decides what happens.
- Know which processes depend on mail Sign-in codes, confirmations, invoices and the tools your colleagues build. With that list, you know on the first morning what has stopped and whom to warn.
- Hold off on resending Delayed mail will still arrive, and sending it again creates duplicates. Only resend what expires, such as sign-in codes and reset links, or what genuinely bounces as undeliverable.
Where we come in · Support
We know what your tools depend on
Tools your colleagues build with AI often run on a mailbox, a SharePoint list or a Teams channel. With Support, senior developers continuously check our fixed checklist, in which external services and continuity are standard items. So you know in advance which tool stops when a service goes down, and every risk becomes a task with an owner.
Do you know how you will reach each other when mail and Teams go down?
Tell us what your work runs on. We’ll think it through with you and always reply within one working day.