Between Thursday 12 and Saturday 14 November, Outlook, Teams and other Microsoft 365 services stopped working normally across much of the world. Here is a summary of what happened in the Microsoft 365 outage, what the company has said about the cause and why recovery dragged on for so long.
What happened during the Microsoft 365 outage
The problems started on Thursday afternoon, around 5 pm Spanish time. Outlook went first: many people couldn't open their email in the browser or on their phone, and the messages that did get through arrived late. Within an hour there were reports from more than seventy countries.
Teams, Copilot and tools from other companies that connect to Microsoft 365 were affected next. Some public administrations had trouble with email, a trial was halted in the UK and several Hong Kong banks reported disruption. Microsoft hasn't given a figure for how many users were affected.
What caused it, according to Microsoft
In the first few hours, Microsoft talked about elevated load on part of its infrastructure. It later clarified that the load did not come from more traffic, ruled out a cyberattack and put the problem down to a latent bug exposed by some network changes.
On Sunday it published a preliminary report with more detail. The bug was in an update to the servers that handle connections to Microsoft 365, rolled out from July onwards, which had run without problems for over three months. Under certain traffic conditions, those servers ran out of processing capacity and stopped accepting connections. Those conditions came together on Thursday, when part of the Teams traffic was moved onto the same servers Outlook email uses while another part of the capacity was out of service because of an unrelated issue.
According to the report, testing didn't catch it because the bug got worse on newer machines, the ones with more processor cores.
Why recovery was so slow
Automated systems moved connections to the servers that were still working, and those failed for the same reason once the load reached them. In under two hours the problem had spread to other regions, and every time a group of servers was restarted, all the traffic that had been waiting came back at once and brought them down again.
The fix was to return that part of the system to how it worked before the update, take large groups of servers out of service, apply the change and give them traffic back gradually. Microsoft places the main impact between Thursday afternoon and Friday afternoon, although it didn't close the incident until Saturday. It also extended the email delivery window so queued messages wouldn't bounce as undeliverable.
In the meantime, anyone with Outlook installed on their computer could still read the email they had already downloaded, and Teams calls mostly kept working.
Not the first outage this year
In January, Microsoft 365 was down for almost ten hours in North America, due to a mix of heavy load and reduced capacity during maintenance. In late summer another incident locked Outlook users out of their mailboxes.
Autumn had already been busy for Microsoft, too. In October it fixed more than 1,400 vulnerabilities, its highest monthly count, as we covered in our piece on security flaws and AI. It isn't directly related to this outage, but it gives a sense of the volume of changes a platform like this handles.
Other large providers have had similar outages in recent years. With so many services concentrated on a few platforms, when one of them fails the effect reaches all kinds of businesses, including ones that don't use that service directly.
What we still don't know
The report Microsoft has published is preliminary. The final version, due on 20 November, should set out what Microsoft will change so that a bug like this doesn't spread to every region again. When it comes out, we'll update this article with anything new.