,

Microsoft 365’s 24-Hour Outage: What Failed, What It Cost and What It Means for Every Industry

Cover for the Microsoft 365 outage of November 2026

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

Timeline of the Microsoft 365 outageTimeline from 16:17 UTC on 12 November to 03:51 UTC on 14 November 2026 with seven key events.Outage timeline, 12–14 November 2026 (all times UTC)12 Nov 12:0013 Nov 00:0013 Nov 12:0014 Nov 00:001234567Impact window per Microsoft: 16:17 on 12 Nov to 16:10 on 13 Nov (about 24 hours)Clean-up until Microsoft declared the incident remediated (03:51 on 14 Nov)112 Nov 16:17: users start losing mailbox access212 Nov about 19:06: Microsoft pauses deployments and configuration changes312 Nov 23:04: Microsoft starts returning some affected servers to service413 Nov 04:36: partial recovery reported in some regions513 Nov 16:10: end time Microsoft lists for the incident613 Nov 17:57: Microsoft apologises, says it was not caused by a cyberattack or AI agents714 Nov 03:51: incident declared remediated
Times are UTC (UK time equals UTC in November). Events 1, 2, 5, 6 and 7 follow Microsoft’s status updates as reported by azureblog.co.uk; events 3 and 4 follow BBC reporting summarised on Wikipedia. Update timestamps in the status history are not always in order, so treat minutes as approximate.

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:

How the Microsoft 365 outage unfoldedFive steps from a dormant July 2026 defect to a fleet-wide failure that could not be rolled back.How one dormant bug became a global outage1July 2026: a defect shipsA bug in the Microsoft 365 networking gateway software goes live.It stays dormant, and reportedly only appears on newer, high-core servers.2November: the conditions line upTeams traffic is redirected to the gateways that serve Exchange Online.Maintenance cuts the gateway fleet’s capacity by 14%.3Servers stop respondingUnder that load the dormant defect makes gateway servers unresponsive.Health checks fail and the servers are pulled out of service.4Automation spreads the failureFailover moves traffic onto the remaining healthy servers.They hit the same defect. Roughly 80% of the gateway fleet is affected.5No quick way backA simple rollback is impossible: the faulty code is months old.Engineers build a fix and restart servers in groups, bringing capacity back slowly.
Based on Microsoft’s preliminary account as reported by Petri (defect, traffic redirection, failover, roughly 80% of the fleet, no rollback) and a post by Microsoft’s Charles Lamanna summarised on Wikipedia (14% capacity reduction). Microsoft’s final report was still pending when this was written.
  • 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.

IndustryOutageWhat was reported
HealthcareMicrosoft 365, Nov 2026Toronto 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.
HealthcareCrowdStrike, Jul 2024Mass 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).
AviationCrowdStrike, Jul 20245,078 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).
AviationHeathrow substation fire, Mar 2025A 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 servicesCrowdStrike, Jul 2024911 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 courtsMicrosoft 365, Nov 2026Japan’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 businessMicrosoft 365, Nov 2026Richemont delayed its quarterly results (Bloomberg). The Anderson Economic Group estimated about $7 billion in lost productivity (via CNN).
EducationMicrosoft 365, Nov 2026Ofqual, 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:

Published loss estimates for two outagesBar chart: Microsoft 365 outage about 7 billion dollars in lost productivity, CrowdStrike outage about 5.4 billion dollars direct losses for top 500 US companies, Delta alone about 0.55 billion dollars.Published loss estimates (US$ billions)Microsoft 365 outage, Nov 2026: lost productivity (Anderson Economic Group, via CNN)about $7 billionCrowdStrike outage, Jul 2024: direct losses of the top 500 US companies (Parametrix)about $5.4 billionCrowdStrike outage, Jul 2024: Delta Air Lines alone (SEC filing estimate)about $0.55 billionDifferent methods and scope. Read each bar on its own; do not add them or rank them against each other.
Sources: Anderson Economic Group estimate quoted by CNN, and Parametrix and Delta SEC filing figures, all as summarised on Wikipedia’s pages for the 2026 Microsoft outage and the 2024 CrowdStrike outage. Delta’s CEO put its cost at $500 million; its SEC filing estimated about $550 million.

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:

Insured versus uninsured losses from the CrowdStrike outageStacked bar: 10 to 20 percent of the 5.4 billion dollar direct loss was insured, 80 to 90 percent was not.Who pays? CrowdStrike outage losses vs insuranceDirect losses of the top 500 US companies by revenue (excluding Microsoft): about $5.4 billionInsured: $0.54–$1.08 billion (10–20%)Uninsured: roughly $4.3–$4.9 billion (80–90%)Insured range is from Parametrix. The uninsured figure is our subtraction from the $5.4 billion total.
Source: Parametrix estimate, as summarised on Wikipedia. Percentages are our arithmetic: $0.54 billion and $1.08 billion divided by $5.4 billion.

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:

Hourly cost of downtime for enterprisesBar chart: over 90 percent of mid-size and large firms say an hour of downtime costs more than 300,000 dollars; 41 percent say 1 million to over 5 million dollars.What an hour of downtime costs a mid-size or large businessMore than $300,000 per hourover 90% of firms$1 million to over $5 million per hour41% of firmsITIC survey of 1,000+ firms worldwide (Nov 2023 to Mar 2024). Excludes litigation and civil or criminal penalties.
Source: ITIC 2024 Hourly Cost of Downtime survey, as summarised in ITIC’s published results. Bars are drawn to scale.

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.

  1. 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.
  2. 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.
  3. 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.
  4. 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.
  5. 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.
  6. 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

Written by

GitHub · About Codeonion · Contact

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

>> Related articles

all articles →