So last week Microsoft had a bit of an issue, and if you use any of their tools you either had a day off, or you had a horrible weekend compensating for all the accumulated work (or both).
A Microsoft 365 outage began on November 12, 2026, continued into November 13, and restoration updates extended into November 14.
Microsoft initially said a recent network change had exposed a problem introduced during an earlier capacity upgrade. But what does that mean? How does a change behind the scenes stop people around the world from opening their email?
I'll use a supermarket example to make the sequence easier to picture. It's a simplified explanation.
Microsoft's systems are much more complex, especially at this scale.
This article is intended to explain a complex technical issue in the simplest way I can, so please take this into consideration.
Also, please note that this is an evolving story, so it's possible that things will change as more details emerge.
Let's take a look at the problem.
A problem waiting to happen
Microsoft's November 15 preliminary report describes three factors combining:
- a hidden software problem
- a shift in connection requests
- fewer available servers.
Microsoft also explicitly ruled out cyber activity, AI, and agents, and this is incredibly important, since nowadays it's the first thing that everyone thinks, and for good reason.
Gateway servers are computers that handle connections to Microsoft's services. Traffic means the requests passing through them, and capacity is how much work they can handle.
The software problem came from a July change to those servers. Under heavy load, internal tasks started competing for the same resources until the server ran out of processor and memory capacity and froze.
Why didn't Microsoft's testing catch it? Ironically, their lab tests ran on older, weaker computers. This specific bug actually got worse on newer, more powerful servers with more processor cores and because conditions in the real world stayed calm (for Microsoft standards) enough at first, the change ran for over three months without any visible issues.
Think of a supermarket with a problem in its checkout software. It works during an ordinary day. Customers arrive, pay, and leave.
Nothing failed because we didn't hit the issue yet.
The older software problem was already present. What changed later was the work those servers had to handle.
What changed that day?
Teams traffic had automatically switched to a gateway shared with Exchange Online, Microsoft's hosted email and calendar service. Moving traffic between systems is part of normal operations, but this put more work on that gateway.
Meanwhile, nearly 14% of that gateway's server capacity was unavailable for an unrelated reason. That alone hadn't caused problems. The combination exposed the software problem, according to Microsoft.
Back to our supermarket. Imagine customers who normally use a separate set of checkouts are redirected to these lanes. Some lanes are already closed, so the remaining ones now have more work.
The supermarket doesn't need more customers overall for those lanes to become busier. The same customers are now using fewer checkouts.
But the remaining checkouts are now working under different conditions. A software problem that never appeared on a quieter day can become important very quickly.
The useful detail here is the combination. Looking only at the most recent change leaves out the older defect and the capacity that was already unavailable.
How the problem spread
Microsoft says traffic moved from failed servers to healthy ones, which encountered the same defect. At the worst point, over 80% of servers in the affected gateway infrastructure were unresponsive.
In our supermarket example, closing one checkout sends its queue to the others. People can still get served while someone repairs the broken checkout.
Now imagine every checkout has the same software problem. Moving customers makes the remaining lanes busier, which exposes the problem there too. Another lane stops, its customers move again, and the situation repeats.
So we end up with the same customers waiting at fewer working checkouts. Each remaining checkout has a longer queue to deal with, putting more pressure on the lanes that are still working.
That's a cascading failure: one failure creates conditions that cause more failures. Having more machines helps with many problems, but moving work between machines only helps if the destination can handle it successfully.
Normally, Microsoft has regional barriers, like watertight compartments on a ship (last analogy I promise!), to keep a failure in one part of the world from spilling into others. The overload started during peak hours in the United States, but due to a separate issue, those barriers didn't hold, allowing the cascade to spread globally in less than two hours.
Why more than Outlook was affected
Outlook was the most visible part of the disruption for many people. But Microsoft's incident updates also identified effects on Teams, SharePoint, and Microsoft 365 Copilot. The impact varied between services and customers.
These apps can share connection infrastructure. Some features also rely on Exchange Online, so problems there can affect parts of other apps.
A calendar feature needs calendar data, for example. If that data can't be reached, the feature can fail even when the rest of the app opens.
This helps explain why people could have different experiences during the same incident. The app might open, but a feature inside it could still be unavailable.
It also explains why some apps recovered faster than email. Services like Teams, SharePoint, and OneDrive were able to switch away from the broken gateway to a backup one, so core features like Teams chats and meetings kept working for most of the outage. Exchange Online (which powers Outlook) is simply too large to move out of that gateway, so it took the full brunt of the failure.
Why Microsoft couldn't just turn everything back on
Because the faulty code was three months old, Microsoft couldn't just press an automatic "undo" button to roll it back. And when engineers tried simply restarting servers, the waiting traffic immediately overwhelmed them again. Instead, engineers had to write a new fix, take large groups of servers offline (often over half of a region at a time), apply the update, and bring them back together.
In our example, reopening a checkout wouldn't help if the same software problem remained. And even after fixing the software, reopening just one checkout when hundreds of customers are waiting outside would immediately overwhelm that single lane.
To recover safely, the store has to fix the software on a large group of checkouts behind closed doors, then open many lanes at the same time so the waiting crowd is spread out across all of them.
Microsoft said it restored services gradually, checking stability at each stage. Its restoration updates explained that this took longer but helped ensure a stable recovery.
There was also a backlog of email, meaning messages waiting to be processed. Microsoft warned that access could return before email delivery fully caught up.
Opening the supermarket doors again doesn't instantly serve everyone who has been waiting outside. The shop can be working while the queue still takes time to clear.
Imagine the store is serving the customers already waiting while also letting more people in gradually, so the working checkouts aren't overwhelmed again. Someone arriving now might see a functioning store, while a customer who has been waiting through the disruption still hasn't reached the front of the queue.
That's the distinction between being open again and having caught up with all the waiting work.
That's why being able to open Outlook again and receiving a delayed email could happen at different times. Restoring access and finishing the accumulated work were different parts of recovery.
Final Thoughts
The supermarket example helps explain why a system can work for months and then fail when conditions change. More checkouts don't solve every problem if they all have the same weakness.
Probably I went a bit overboard with the example, so if it doesn't make sense please let me know or comment bellow and I'll fix it.
The findings discussed here remain preliminary. As of the writing of this article, this is the most up-to-date information I have, but I'll update the article once I know more.
Sources
- Microsoft's official statement
- Microsoft's preliminary post-incident report, dated November 15, 2026 (public PDF copy hosted by 365Fanatic)
- Gulf Business reporting on Microsoft's explanation, November 16, 2026
- Public reproduction of Microsoft's restoration updates
- Microsoft 365 admin center. Search Service health for EX1550543 and MO1550821. Administrator access required.
Photo by Sushobhan Badhai on Unsplash
No comments yet
Be the first to share your thoughts on this article!