Table of Contents
I Go Away and What Happens… The Biggest Exchange Online Outage Ever
I guess I shouldn’t take a ferry to France to search for interesting wines. I did last week, and Exchange Online promptly experienced one of the biggest and most public outages in Microsoft 365 history. It was so bad that I was called for comment by NBC Los Angeles, the New York Post, and NDTV Profit, an Indian business and finance news channel. That just doesn’t happen normally.
Now that normal service has been resumed (and the press has stopped calling), it’s time to reflect on the situation.
What Happened to Exchange Online
According to the service history for EX1550543, the problem started at 16:20 GMT on November 12, 2026. Figure 1 shows a start time of 10:05 GMT, but this appears to be an error because the first note in the service history is time stamped at 16:20 and says “We’re investigating a potential issue and checking for impact to your organization.” I noticed that the problem eased for my tenant and email started to arrive again around 10:00 GMT on November 13, so the outage affected me for nearly 18 hours. However, Microsoft determines the exact length of an incident based on service recovery, and the Microsoft 365 service health dashboard shows an end time of 16:10 GMT on November 13. Using 16:20 GMT as the start time, EX1550543 lasted 23 hours and 50 minutes. This time might be adjusted by Microsoft as they develop their post-incident analysis.

MO1550821 is the other incident. This is a catch-all incident covering the knock-on effects of the Exchange Online outage on other Microsoft 365 services. The start time for the incident is the same as for EX1550543. The service health dashboard shows an end time of 04:10 GMT on 13 November, giving an incident length of 11 hours and 50 minutes. I’m unsure why Microsoft regarded MO1550821 as over before EX1550543.
The full answer to what happened to Exchange Online should be revealed when Microsoft publishes its Post-Incident Report (PIR), which should be a very interesting read. I don’t know if Microsoft will issue separate PIRs for EX1550543 and MO1550821. It would seem sensible to have a single PIR covering both.
Not an Outlook Problem
Despite the furore across the internet, this wasn’t an Outlook problem. The issue was with Exchange Online. Some Outlook clients had worse experiences. Outlook classic has great offline capabilities and works better in these circumstances than the new Outlook. Mobile clients are out of luck. And many other non-Outlook clients connect to Exchange Online over protocols like ActiveSync (Apple iOS and Samsung mobile mail clients) and POP3/IMAP4 (Thunderbird and other clients). Outlook was unfairly tagged simply because that’s where most people experienced the problem.
The cause was not a cyberattack. Microsoft has said this consistently, including a statement from Charles Lamanna, Microsoft EVP for Copilot, Agents, and Platform, saying “I can confirm that there is no indication this is the result of malicious or cyber activity.” Later he added that the problems seemed to involve “cascading failures + thundering herds at our network perimeter” (Figure 2).

Based on what’s been shared to date, the failure involved a large number of Exchange Online servers becoming unavailable, which pushed additional load onto the remaining infrastructure. Reuters said that Downdetecter logged more than 7.4 million reports worldwide
Microsoft subsequently said that the incident was introduced by an issue during an earlier network capacity upgrade that surfaced following a recent network change. That’s more information, but not yet enough to explain precisely why so many Exchange Online servers became unavailable or why recovery proved so difficult. We’ll have to wait for the PIR to join the dots.
Another striking aspect was how difficult recovery proved to be. Restarting servers wasn’t enough because some became unresponsive again when traffic returned. Microsoft had to restore capacity gradually, rebalance traffic, and drain accumulated email queues while checking that the infrastructure remained stable. In some respects, restoring a service of this scale can be harder than breaking it. Bringing capacity back too quickly risks creating another thundering herd and starting the cycle all over again.
According to Microsoft speakers at the MEC 2022 conference, at the time, Exchange Online spanned 300,000 physical mailbox servers and handled 9.2 billion messages daily. Microsoft hasn’t published an updated server count. Any way you count the servers, a lot were affected.
Exchange Online is at the Core of Microsoft 365
Exchange Online is the Microsoft 365 email service. It’s used by other applications for many reasons, including SharePoint Online sending sharing links to people, or Teams users scheduling meetings. Exchange Online also hosts compliance records generated to track how people use features like Microsoft Copilot. As Exchange performance declined, other applications experienced their own problems. The list of services mentioned by Microsoft in its service updates (Figure 3) is evidence of just how connected and dependent other services are on Exchange Online.

Despite many posts on X and elsewhere, I know of no evidence that Microsoft’s recent employee redundancy programs meant that engineers didn’t have enough experience or knowledge of core Microsoft 365 services to fix the problem more quickly. It’s true that many experienced and well-respected Microsoft engineers have left the company this year, but whether retaining those engineers would have accelerated diagnosis and recovery is impossible to know.
It’s plausible that having access to people with years of experience with Exchange Online, its connections to the rest of Microsoft 365, and the underlying datacenter infrastructure would have helped drive the problem to a faster conclusion, especially if that experience included similar situations. Those of us who have been around the block a few times like to think that our knowledge is invaluable when problems erupt, but perhaps not when things fall apart at the scale at which this outage unfolded. After all, how many engineers working today have experience of handling a worldwide outage that affected a potential user population of around of around 500 million people (Microsoft 365 470-odd million paid seats based on the growth number reported in its FY27 Q1 results plus free seats used by partners and others).
The Impact on Uptime Percentage
When horrible problems happen to services that we depend on, it’s natural for consumers to think about compensation. In this case, Microsoft documents its commitment for different online services like Exchange Online in its Service Level Agreements for Online Services.
The section for Exchange Online defines downtime as any period of time when users are unable to send or receive email with Outlook Web Access. Choosing OWA is sensible because the client depends on a direct connection to the server. Microsoft’s formula for uptime percentage is calculated using this formula:

Downtime is measured in user-minutes. In other words, the number of minutes that users were deprived of the service because they couldn’t send or receive email. Different tenants might have different outage minutes depending on the datacenter region they connect to.
There are 43,200 minutes in November. We don’t know Microsoft’s official outage length yet, but taking the 23 hours and 50 minutes for EX1550543 as calculated above (1,430 minutes) as the figure, if every user in the tenant lost access to Exchange Online for that amount of time, the uptime percentage is:
(43,200 – 1,430) / 43,200 = 96.69%
For tenants where every user lost Exchange Online access throughout the outage, these calculations are a reasonable guide. If only some users were affected, or users recovered at different times, Microsoft’s user-minute calculation produces a different result. The calculation is below 99% and above 95%, resulting in the 50% tier for a service credit.
Microsoft 365 recorded its worst uptime performance since I started tracking the data in 2013 when it achieved 99.526% in the first quarter of 2026. Comparable outages would have left their mark on earlier quarterly results, which is why we can reasonably say that the November 2026 incident ranks among the biggest outages in Microsoft 365 history. The Q1 CY26 result was already a huge surprise because Microsoft 365 had been so consistent over the years. The hypothetical monthly Exchange Online figures generated by this outage are considerably worse, which leads to the question of service credits.
Will Customers Get Some Service Credits?
According to Microsoft’s service level agreement, customers may qualify for service credits when the thresholds for uptime percentage performance aren’t met.
| Uptime Percentage | Service Credit |
| < 99.9% | 25% |
| < 99% | 50% |
| < 95% | 100% |
Service credits are the “sole and exclusive remedy for any performance or availability issues for any Service under the Agreement.” Microsoft doesn’t return cash to compensate customers. A service credit is applied against future fees customers pay to use the affected service. Effectively, a service credit is a partial refund to compensate for loss of service from Microsoft. Explicitly, Microsoft says that service credits “will not be awarded to compensate for any other forms of loss.”
Few customers pay only for Exchange Online. Instead, they buy products such as Microsoft 365 E5 that cover many services. An Exchange Online failure producing an uptime percentage between 95% and 99% doesn’t yield a 50% service credit based on the Microsoft 365 license fee. SLA calculations and service credits depend on the affected service, its applicable monthly service fees, and the terms covering that service.
For comparison, Microsoft charges $8 per user per month in the U.S. for a standalone Exchange Online Plan 2 license. That doesn’t mean Microsoft values the Exchange Online component of Microsoft 365 E5 at $8 for SLA purposes, but it illustrates why a 50% Exchange Online service credit shouldn’t be interpreted as 50% of the full monthly cost of a Microsoft 365 E5 license ($60).
Where problems affected other Microsoft 365 services such as Teams, their uptime percentage must be assessed against the applicable SLA terms for those services. Many affected services continued to function to some extent. For example, users could continue working with files stored in SharePoint Online and OneDrive for Business document libraries but might have been unable to share files because Exchange Online couldn’t accept and deliver sharing notifications. Copilot Chat might have failed because it couldn’t connect to Exchange Online to store compliance records. Each service therefore experienced a different degree and type of disruption. Whether these problems constitute downtime for SLA purposes depends on how Microsoft defines availability for each service. Determining whether a service credit is due, and if so how much, is consequently likely to be a complicated exercise.
The point is that Microsoft 365 is a highly connected ecosystem where the connections between applications are not always obvious. Because Exchange Online provides much more than email, including calendar, mailbox, compliance, and substrate services used by other workloads, it is probably one of the most deeply connected services in Microsoft 365.
The other thing about service credits is that Microsoft will not extend the credit automatically. Customers must make a claim through the Microsoft 365 admin center or their CSP. Claims for November 2026 must be made by the end of December 2026.
A Horrible Outage
The outage was horrible and caused many to doubt the wisdom of using cloud services (perhaps the same people who doubted the wisdom of using on-premises servers?). That said, on-premises Exchange Server deployments are hardly immune to major outages. The difference is scale. An on-premises outage usually ruins the day for one company. When something goes badly wrong in the cloud, thousands of companies can have their day ruined at the same time, and everyone hears about it.
More importantly, the outage raised the question of how organizations should cope when all their eggs are in the proverbial single basket. Different companies experienced different problems in different locations, and I wonder how many will now reconsider their plans for handling the next great outage.
Exchange Online has a solid mailbox database resilency layer (the Database Availability Group or DAG), and spreads tenants across individual forests made up of DAGs. Exchange Online has a solid mailbox database resiliency layer (the Database Availability Group or DAG), and spreads tenants across individual forests made up of DAGs. It seems that the Achilles heel exposed by this incident wasn’t Exchange Online’s mailbox database resiliency layer but the infrastructure responsible for accepting and routing connections into the service.
Things must improve. Consider what might happen if Microsoft’s dream of a proliferation of agents comes to pass without change. In a recent podcast, Satya Nadella Satya Nadella imagined a Microsoft with 200,000 employees and perhaps 2 million or even 20 million agents “all in a loop.” Many of those agents will potentially depend on Exchange Online to access email, calendars, contacts, or other mailbox data, or to act on behalf of users. Imagine what kind of load that environment will generate if this kind of agent deployment spreads outside Microsoft?
Me, I’m just glad things are back to normal.