For about 24 hours in November 2026, email stopped working for a large part of the world. Microsoft 365, the suite behind Outlook, Teams and SharePoint, went down on 12 November and was not declared fully fixed until the early hours of 14 November. Hospitals, courts, government departments and schools all felt it.
Strictly speaking this was an outage, not a power blackout: the servers had power, but a software fault stopped them answering. For the people affected the result was the same. This article sets out what is known about what happened, what such events cost, and how the same pattern has hit other industries. Every number below is attributed to a source, and where sources disagree we say so.
At a glance
- When: 12–14 November 2026. Users started losing mailbox access around 16:17 UTC on 12 November. Microsoft lists 16:10 UTC on 13 November as the end time and declared the incident remediated at 03:51 UTC on 14 November (Microsoft status updates, via azureblog.co.uk).
- What broke: Exchange Online mailboxes were unreachable through Outlook on the web, desktop and mobile. Teams calendars and meeting invites, SharePoint external sharing, Copilot Chat, Microsoft Graph requests and several security and compliance tools were also affected. Teams calls and chats were largely unaffected.
- How many people: about 430 million people use Microsoft 365, and problems were reported in more than 70 countries (BBC, via Wikipedia). Downdetector logged about 130,000 Outlook problem reports at the peak, a third-party figure rather than one from Microsoft.
- Cause, as far as known: Microsoft said it was a technical issue and not a cyberattack or the work of AI agents. A preliminary account points to a dormant software defect in networking gateway servers (details below).
- Tracking IDs: MO1550821 (Microsoft 365) and EX1550543 (Exchange Online) in the Microsoft 365 admin center.
- Mail safety net: Microsoft extended the email timeout window to 48 hours so delayed messages were less likely to bounce.
Timeline
The first thing to notice is the shape of the red bar. Microsoft’s own listed window runs for about 24 hours, but a report on its preliminary findings describes the disruption as lasting more than twelve hours (Petri). Services came back gradually, region by region, and roughly 16 hours in the problem had narrowed mostly to Exchange (The Guardian, via Wikipedia).
What actually failed
Microsoft’s first public explanation, on 13 November, said the incident came from an issue during an earlier network capacity upgrade that surfaced after a recent network change. A fuller preliminary account, summarised by Petri and in a post from Microsoft’s Charles Lamanna, describes several problems stacking up:
- A dormant defect. A software defect introduced in July 2026 sat in the networking gateway servers for months. It did nothing under normal conditions.
- A change in load. Teams traffic had recently been redirected to the gateways that serve Exchange Online, and maintenance reduced the fleet’s capacity by 14%, according to Lamanna.
- A failure that fed itself. Once servers became unresponsive, automatic failover pushed their traffic onto healthy servers, which then hit the same defect. Petri reports that roughly 80% of the gateway fleet was affected.
- No simple rollback. The faulty code was months old, so reverting a recent change could not undo it. Engineers built a targeted fix and restarted servers in groups. As servers returned they were swamped by the backlog of traffic, which slowed recovery further.
Petri also reports that Microsoft’s validation missed the defect because test environments relied largely on older hardware, while the bug was more likely to appear on newer servers with more processor cores. That explanation is the publication’s account of the preliminary report, not a direct Microsoft quote, so treat it as reported rather than confirmed.
Who felt it: industry by industry
Email sounds like a minor tool until it carries the work of a hospital ward or a court. The table lists documented effects from this outage and from other large failures, so you can see that the pattern is not specific to Microsoft or to IT.
| Industry | Outage | What was reported |
|---|---|---|
| Healthcare | Microsoft 365, Nov 2026 | Toronto hospitals University Health Network, SickKids and Sunnybrook reported Outlook problems, and Unity Health Toronto declared a code grey (CBC). In the UK, NHS disruption was reported at hundreds of hospitals and GP practices, with some operations cancelled and paper records used in places (The Guardian). Both via Wikipedia. |
| Healthcare | CrowdStrike, Jul 2024 | Mass General Brigham cancelled all non-emergency procedures and visits. Memorial Sloan Kettering postponed procedures needing anaesthesia. Kaiser San Jose Medical Center lost access to patient records and newborn vital-sign monitoring. Two German hospitals cancelled non-emergency operations (Wikipedia). |
| Aviation | CrowdStrike, Jul 2024 | 5,078 flights were cancelled worldwide on 19 July, 4.6% of those scheduled. Delta cancelled more than 7,000 flights by 25 July, affected about 1.3 million passengers by its own estimate, and put the cost at $500 million (CEO) or about $550 million (SEC filing: $380 million lost revenue plus $170 million expenses). |
| Aviation | Heathrow substation fire, Mar 2025 | A fire at the North Hyde electrical substation cut power and closed Heathrow for most of 21 March. Estimates put disruption at more than 1,300 flights and roughly 300,000 passengers; figures vary by source and no audited airline cost was found (Airline Ratings). |
| Emergency services | CrowdStrike, Jul 2024 | 911 outages or disruption were reported in at least 15 US states, with 911 down statewide in New Hampshire. Many call centres switched to backup systems. |
| Government and courts | Microsoft 365, Nov 2026 | Japan’s Government Solution Service was disrupted. The Philippine Supreme Court warned of delayed email responses. Singapore’s court systems had issues. US House staff said the outage ended a day of legislative work, and US law enforcement reported problems including processing bail payments (via Wikipedia, citing CNN, BBC and Nippon.com). |
| Finance and business | Microsoft 365, Nov 2026 | Richemont delayed its quarterly results (Bloomberg). The Anderson Economic Group estimated about $7 billion in lost productivity (via CNN). |
| Education | Microsoft 365, Nov 2026 | Ofqual, ASCL, the Department for Education and many schools in England reported major disruption (Schools Week). |
What it costs
Putting a price on an outage is difficult, and every figure depends on who is counting and what they include. Here are the published estimates, each on its own terms:
Two things stand out. First, these are estimates from different organisations using different definitions, so they should not be added together. Second, most of the cost does not land on the company that caused the outage:
For the CrowdStrike outage, Parametrix estimated that only $540 million to $1.08 billion of the roughly $5.4 billion in direct losses was insured. That leaves most of the bill with the businesses themselves. The same pattern shows up in surveys of what downtime costs in general:
Why the same pattern keeps repeating
- One supplier, many dependants. Hospitals, courts and airlines all rely on the same few software platforms. After the November outage, members of the European Parliament pointed to Europe’s dependence on a single company, and politicians in Portugal and Brazil raised questions about open standards and domestic technology (via Wikipedia).
- Fast to break, slow to fix. CrowdStrike’s faulty update went out at 04:09 UTC on 19 July 2024 and was reverted at 05:27 UTC, a window of 78 minutes. Yet affected machines needed manual repair one at a time, and full recovery took days. The Microsoft outage followed the same shape: the failure spread in hours and the cleanup ran into a second day.
- Automation can amplify a fault. Failover is designed to protect a system, but when every server shares the same hidden bug, moving load around just spreads the problem. That is what Petri describes in Microsoft’s case.
- Dormant bugs are the hardest. A defect that ships in July and bites in November cannot be fixed by undoing last week’s change.
- Physical single points of failure too. Heathrow shows the same logic with electricity: one substation fire took out an airport that handled more than 83.9 million passengers in 2024 (Airline Ratings).
What you can do about it
These are our suggestions, not findings from the sources above. They apply whether you run a hospital IT team, a small business or a side project.
- List your single points of failure. Write down what stops if your email, login provider or cloud region goes down. Most teams are surprised by the length of the list.
- Have a second way to talk. If your team chat and email run on the same provider, a phone tree or a separate messaging app lets you coordinate when both are down.
- Keep an offline fallback for critical work. NHS staff reportedly used paper records during the outage. Printed contact lists and manual procedures cost almost nothing and matter on the day.
- Know what your contract and insurance actually cover. Check the SLA credits and ask whether your cyber or business-interruption policy covers a supplier’s accidental outage. The CrowdStrike numbers suggest many firms were not covered.
- If you ship software, make rollback real. Release to a small slice of users first, keep the ability to revert, and test on the newest hardware you run in production. Slow rollouts are a cheaper fix than a global incident.
- Practise the failure. Run a short drill where email is unavailable for an hour. The gaps you find will be small and fixable.
What we still don’t know
- The final cause. Microsoft has promised a post-incident report with more technical detail and prevention steps. When this was written it had not been published, and sources differ on details such as whether the trigger was the network capacity upgrade, the traffic redirection, or both.
- The real duration. Reports range from more than twelve hours to about 24 hours depending on whether you count first impact, Microsoft’s listed end time or the final all-clear.
- The total cost. The $7 billion figure is one organisation’s estimate of lost productivity. No audited figure exists.
We will update this article when Microsoft publishes its final report. Microsoft’s own status pages were not accessible to us, so the timeline above relies on reputable reports of those updates and on the news coverage they cite.
Sources
- 2026 Microsoft outage, Wikipedia (timeline, scale, affected organisations, Lamanna’s account, with citations to BBC, The Guardian, CNN, CBC, Bloomberg and others)
- Microsoft 365 outage, 12 November 2026, azureblog.co.uk (status update timeline, incident IDs, services affected, Downdetector figure)
- Microsoft explains the Microsoft 365 outage, Petri (preliminary post-incident report summary)
- 2024 CrowdStrike-related IT outages, Wikipedia (flights, hospitals, 911, Delta, Parametrix)
- ITIC 2024 Hourly Cost of Downtime survey
- The impact of Heathrow’s closure in numbers, Airline Ratings




Leave a Reply