BLOG | Nov 17, 2026

Analyzing the Microsoft Outlook outage: when the failover spreads the failure

On November 12, a software defect that had sat dormant in Microsoft’s networking gateway servers since July took down Outlook for organizations around the world. Here’s what Microsoft’s preliminary report reveals, and what it means for how you think about failover, shared paths, and change in your own network.
Forward Labs
Forward Labs
 
Who should read this post?
  • Network engineers and architects who design for failover and redundancy
  • NetOps and incident response teams who need to triage issues as fast as possible
  • IT leaders accountable for service resilience and outage risk
What is covered in this content?

If your organization runs on Microsoft Outlook, you probably spent last Thursday fielding the same question: “Is email down for everyone, or just us?” It was down across much of the world. Within an hour, outage reports had come in from 74 countries and territories. Microsoft has since published a preliminary account of what went wrong, and it’s a useful case study in how large failures come together.

What happened, and when?

On Thursday, November 12, the networking gateway that handles traffic for many Microsoft 365 services began to fail. Exchange Online, the cloud email service behind Outlook, took the hardest hit, though other services were disrupted too. Microsoft’s monitoring flagged rising connectivity problems at 11:03 a.m. ET, and the first customer support case arrived 30 minutes later.

Microsoft’s preliminary post-incident report, released November 15 and summarized publicly by executive Charles Lamanna, attributes the outage to three factors:

  • A latent defect. In July, Microsoft deployed a fix built specifically for its Microsoft 365 networking gateway servers. The fix introduced lock contention in an internal Windows process, meaning processes stall while waiting on one another. Under certain conditions, that could exhaust a server’s CPU and memory. The fix passed load testing and ran worldwide without impact for more than three months. However, that testing focused on less powerful hardware with fewer processor cores, and the defect was worse on newer machines with more cores.
  • Added load. On the evening of November 11, Teams failed over onto the primary gateway that also serves Exchange Online.
  • Reduced capacity. Nearly 14% of the gateway’s server capacity was already out of service because of an unrelated issue the day before. The service still had enough headroom, so Microsoft said this reduction “had no impact” on its own.

As U.S. traffic peaked on November 12, the three factors came together. Requests piled up until servers stopped responding, so traffic was rerouted to the healthy ones. Those servers then hit the same defect and failed too. The mechanism meant to route around failure ended up spreading it. In less than two hours, the failure had moved across Microsoft’s gateway fleet, and at its worst, more than 80% of the gateway servers were unresponsive.

Three factors lined up
Figure 1. Three factors, each harmless on its own, lined up on November 12.

Who was impacted?

Any organization that relies on Microsoft 365 could have been affected, no matter how healthy its own network was. But the impact varied by service. Services that could fail out of the gateway did, which is why Teams, OneDrive, SharePoint, and Office apps recovered sooner. Exchange Online couldn’t. According to Microsoft’s report, it can’t fail out of that gateway “for scale reasons.” As a result, Outlook stayed down long after other services came back.

The public sector felt it too. Governments around the world put contingency measures in place, and US city governments reported services returning on Friday after a day of disruption.

Is it resolved?

Yes, but recovery was slow. Servers brought back online were immediately overwhelmed by waiting traffic and went unresponsive again. Microsoft’s report says its automated recovery processes weren’t designed to handle a failure that broad. Engineers ultimately built an update that returned the affected code to its previous known-good state, then applied it to groups of servers.

Microsoft reported healthy service at 11:10 a.m. ET on November 13, roughly 24 hours after its monitoring first flagged the problem. It issued its final customer update that night. A final post-incident report is due November 20.

Why does this matter?

When something breaks, the first question is usually “what changed?” Microsoft asked it too, pausing all ongoing and planned deployments and configuration changes while it investigated.

According to the preliminary report, though, the defect shipped in July. The triggers that week weren’t deployments. They were conditions: a failover that added load, and capacity that was already partly out of service. Neither caused problems alone. Together, they exposed a defect that had run without impact for months.

So “what changed?” was still the right question. The answer just wasn’t a recent deployment. It was a shift in where traffic flowed, onto infrastructure already short on capacity. A review of recent deployments wouldn’t have surfaced that.

Microsoft’s own explanation evolved as the picture came together. Early on, it pointed to a traffic surge and a problem with a component that processes server health. By Friday, it had linked the incident to an earlier upgrade and a more recent network change. The preliminary report then laid out all three factors.

That’s not a knock on Microsoft’s engineers. Symptoms surface first and causes later. When the cause is months old, the obvious places to look can point the wrong way.

For network teams, the more useful set of questions are about how your network behaves when conditions shift. What happens when one service fails over onto infrastructure another service depends on? Which services can’t fail over at all? How much headroom remains when part of your capacity is already out?

Has this happened before?

Yes, including twice since late 2025 at hyperscale.

  • Microsoft Azure Front Door, October 2025. Azure Front Door is Microsoft’s global content delivery and traffic-routing service. Microsoft 365 was among the services disrupted, for roughly eight hours. Microsoft’s post-incident review found that a configuration change exposed a latent bug, which its protection systems failed to stop.
  • Cloudflare, November 2025. A database permissions change caused a configuration file used by Cloudflare’s bot-management system to double in size. As the file propagated across the network, it exceeded a limit in Cloudflare’s core proxy software, which then failed. Cloudflare initially suspected an attack.

These incidents involved different companies and different systems, but they share a shape. A routine change or event met a weakness no one had tested for, and the effects spread well beyond where it started.

Change is a common thread in major outages more broadly. In Uptime Institute’s 2026 annual outage analysis, configuration and change management failure was the most commonly cited root cause of major network-related outages, cited by 41% of survey respondents. 

What should operations teams do now?

None of this was yours to cause or fix. If your users told you before Microsoft did, you likely spent the first stretch ruling out your own network. At points, Microsoft pointed admins to its public status page in case the admin center itself was unavailable.

Now that service is restored, consider these steps:

  1. Check what quietly failed. Rerun scheduled jobs and integrations that depend on Exchange Online. Confirm that security and compliance tooling tied to Microsoft 365 is reporting normally.
  2. Document your impact window while it’s fresh. Note when the outage started and ended for your users, and how many were affected. You’ll need this for your internal review and for any service credit discussion with Microsoft.
  3. Read the final post-incident report. It’s due November 20, and it will say more than any early coverage, this post included.

Leadership may also be asking, “Could this happen to us?” It’s a fair question. Start by defining the minimum service your organization needs if a major provider goes down for a day. Then look at your own network:

  • If something broke right now, how quickly could you prove where the problem exists?
  • Which applications share a path or device, and what happens when one fails over onto another?
  • Which services have only one way out?

What this means for your own network

No one outside Microsoft can verify Microsoft’s network. But the way this failure spread has direct parallels in enterprise networks. A failover pushed traffic onto shared infrastructure, and one service was left with nowhere to go. Three practices help:

  • Know what shares a path. Before a failover moves one application’s traffic onto another’s infrastructure, you should know which services depend on the same firewalls, load balancers, or gateways.
  • Know which services have only one way out. If all your Microsoft 365 traffic exits through a single proxy or internet breakout, that’s your version of Exchange’s gateway. Single points of failure are far easier to find in a design review than in the middle of an incident.
  • Be able to prove quickly that it isn’t your network. During a provider outage, the fastest route to resolution is evidence that your traffic leaves your network correctly and that nothing changed on your side.

A network digital twin is built to answer these questions. Forward Enterprise uses a mathematically accurate digital twin of your hybrid network to trace paths end to end, compare network behavior between any two points in time, whether days, weeks, or months apart, and verify that failover paths exist and behave as designed. The answers come from the network’s real configuration and state, not from diagrams or memory.

The bigger pattern

Uptime Institute’s research notes that outages are increasingly caused by interactions between systems rather than single faults, which makes root causes harder to pin down. This outage is close to a textbook case. A months-old defect, a failover that added load, and capacity already out of service each caused no impact on their own. Together, they disrupted email for organizations around the world.

Networks will keep changing, and failovers and maintenance windows are part of running them well. The goal isn’t to avoid them. It’s knowing, before the next one, what shares a path, what can’t fail over, and how traffic actually moves when conditions shift. For network teams, that’s a visibility problem first: seeing how your network actually behaves, not just how it was designed.

For a closer look at what that visibility requires, read The Network Digital Twin Guide.

Industry Recognition

Winner of over 20 industry awards, Forward Enterprise is the best-in-class network modeling software that customers trust

Customers are unanimous:
Forward Enterprise is a game-changer

From Fortune 50 institutions to top level federal agencies, users agree that Forward Enterprise is unlike any other network modeling software

Most Recent

Browse all posts

Subscribe to our newsletter

Make sure you don't miss a post by signing up here for our monthly 'Moving Forward' newsletter

Ready to get started?

Top cross