07480 257117 Talk to us

The Microsoft outage: what stopped and what kept working

From Thursday 12 to Saturday 14 November, Outlook and much of Microsoft 365 failed, with reports from 74 countries and territories. Email is built to be held and retried. What a business could not do was tell anyone, and in our view that was avoidable.

74 set large on a black background beside a terminal listing Microsoft 365 on 12 November: SharePoint files and cached mail worked, Teams chat mostly worked, and Exchange Online, Outlook and Copilot stopped, above the words: Countries and territories with reports of problems
74countries and territories with reports of problems on Downdetector
130,000reports of Outlook problems in the United States at the peak, roughly
48 hourshow long Microsoft said it would keep trying to deliver held mail
4 to 5 daysthe least the email standard says a sending server should keep trying

At 17:54 UTC on Thursday 12 November, Microsoft said that people in the Americas could not open their Exchange Online mailboxes. About 13 minutes later it warned that the problem was global. On Saturday it said service was back for the last customers and regions. In between, a judge in Manchester sent a jury home, the Metropolitan Police said some online crime reporting was hindered, and parts of the NHS went back to paper.

Microsoft has promised a full report and had not published one when we wrote this. So this post sticks to what was reported, says who reported it, and marks the places where the accounts disagree.

What stopped, in order

The clearest record we found is a running log kept by an IT support firm, which says its times come from Microsoft’s public status posts and cites Microsoft’s service health records. It is one source and we have not seen the originals, so read the times as that log’s times. It credits the 13:26 entry to BleepingComputer, not to Microsoft. They are in US Eastern time, five hours behind the UK.

Thursday 12 November, hour by hour

When each service was listed as affected. US Eastern time.

09:0010:0011:0012:0013:0014:0015:0016:0017:0018:0009:45Teams problem in North America, cleared at 10:5311:17Exchange Online incident opens: mailboxes will not open12:54Microsoft says so in public. Global warning 13 minutes later13:26Start of a wider Microsoft 365 incident15:32Planner and To Do16:41Intune, Defender for Office 365, Purview and Power BI17:11The admin centre for Microsoft 365 apps

Source: On-Site Technology's log, drawn from Microsoft's status posts, read on 14 November 2026

Look at the gap between 11:17 and 12:54. The log has an Exchange incident open in the admin centre at 11:17 and Microsoft’s first public post 97 minutes later. CNBC put the start of widespread problems shortly before 13:00 Eastern, so we cannot say how many businesses had lost mail in that gap. For any that had, it was the time in which people restart routers and ring their IT person.

The cause, in Microsoft’s words as reported by Reuters and AFP, was an issue introduced during an earlier network capacity upgrade, which surfaced after a recent network change. Microsoft said it was not a cyber attack and had nothing to do with AI agents.

74countries and territories with reports
130,000Outlook reports in the US at the peak, roughly
3calendar days it touched, Thursday to Saturday

One number we are not putting in that row: an estimate of about seven billion dollars in losses, attributed to Anderson Economic Group. We have only seen it quoted second hand, so treat it as an estimate of an estimate.

What kept working

An outage of “Microsoft 365” sounds total. It was not, and the difference mattered to anyone who knew where their work actually lived.

Files in SharePointWorked, per the log

The same log reports files stored directly in SharePoint as still opening. CNBC described SharePoint as largely down on Friday morning, so this was not true for everyone.

Mail already on your computerWorked

The log says cached mail may still be readable, so a phone number in an old message could still be found. You could not send or receive.

Teams chat and meetingsMostly worked

Reported as largely working by Thursday evening. It had its own shorter incident that morning.

Sending and receiving emailStopped

Exchange Online is where the main incident opened, and its backlog was the last thing still clearing.

Copilot and connectors to Microsoft 365Stopped

Copilot was reported as not working. OpenAI said it was investigating problems with its Microsoft connectors, and Anthropic said an upstream issue was affecting its Microsoft 365 connector.

Aviva is the example worth copying. Reuters reported that it switched to backup arrangements for its earnings calls. In our view that only works if the second way of holding the call was chosen on a quiet day, before it was needed.

Why held mail is not lost mail

This part is easy to misjudge, and it changes what is worth worrying about.

When your mail server does not answer, the sender’s server does not throw the message away. The standard that email runs on says it should try again, waiting at least 30 minutes between attempts, and should not give up for at least four to five days. So where the sending server was told to try later, a customer who emailed you on Thursday afternoon was holding a message that their own provider was still trying to hand over on Friday.

Microsoft did something similar from its side. CNBC reported that it extended its own delivery window to 48 hours so that held mail would not expire while the repair went on, and on Friday it said the queues were approaching normal levels. Nobody has published how much mail was held or whether any was lost, and we will not guess.

What this means in practice: after an outage like this, expect a flood of old mail, out of order, some of it already overtaken by a phone call. Read the dates before you reply.

What was actually lost

As far as anyone has said, mostly not the email. The thing a business lost on Thursday was the ability to say anything. No “we are open, ring us on this number”. No reply to the customer waiting on a quote. No message to staff who were not in the room. For many businesses the mail system was the only way of reaching people, and it was the thing that had failed.

That is a choice, usually one nobody remembers making. A page on your own website, a list of customer phone numbers that does not live in Outlook, and an enquiry form that keeps what it is sent would each have worked on Thursday, as long as none of them is hosted with Microsoft. Whether customers can find that page is a separate question, and most of the UK business homepages we checked do not yet pass every basic check.

We have written that afternoon up as a list: what to set up before your email provider goes down. And if you want to know whether an incident is Microsoft’s or yours next time, the public status page at status.cloud.microsoft is there for when you cannot sign in.

Where the numbers come from

  1. Associated Press report carried by ABC News and The Globe and Mail, 14 November 2026: 74 countries and territories, the Manchester trial, the pharmacy warning and Microsoft's statement that service was restored.
  2. Reuters and AFP report carried by Dawn, 13 November 2026: the NHS, the Metropolitan Police, Aviva, the Downdetector counts including more than 130,000 Outlook reports in the US, the OpenAI and Anthropic connector notices, and Microsoft's statements of the cause and that it was not related to cyber or AI agents.
  3. CNBC, 12 and 13 November 2026: Microsoft's first acknowledgement, its apology, the 48-hour delivery window, and that problems began shortly before 1pm Eastern.
  4. Time, 13 November 2026: a peak of nearly 130,000 Downdetector reports for Outlook.
  5. Wikipedia, 2026 Microsoft outage, read on 14 November 2026: the 17:54 UTC acknowledgement and the estimate of losses attributed to Anderson Economic Group, which we have not read at its source.
  6. On-Site Technology, an IT firm's running log that says its times come from Microsoft's status posts and cites its service health records, read on 14 November 2026: the order in which services were listed as affected, and what still worked.
  7. RFC 5321, Simple Mail Transfer Protocol, section 4.5.4.1: wait at least 30 minutes between attempts, give up after at least 4 to 5 days.
  8. Microsoft Learn, How to check Microsoft 365 service health, read on 14 November 2026: the public status page for when you cannot sign in to the admin centre.

Edward Clarke, Director. A decade of website development, testing and customer success. Builds our client sites.