Teams could bypass Microsoft 365’s failing network gateway; Exchange Online could not. That is the central explanation in Microsoft’s preliminary report on the global outage of 12 November 2026: the traffic volume of the service powering business Outlook kept it tied to unavailable infrastructure, while Teams could take another route.

The incident began on Thursday at 17:20 Paris time, with users worldwide losing access to Exchange Online mailboxes. It subsequently affected Copilot, SharePoint, OneDrive and several security tools, including Defender. Microsoft says no customers remained affected by 17:10 the following day, almost 24 hours later.

The company published its preliminary report on 15 November in the Microsoft 365 admin centre, accessible only to administrators at customer organisations. On 16 November, Charles Lamanna, executive vice-president responsible for Copilot, agents and platform, shared an abridged version on X, apologising for the disruption to customers.

Many Microsoft 365 services use a shared network gateway through which users’ connections pass before reaching their emails, files or conversations. At the height of the outage, more than 80% of the infrastructure’s servers were unresponsive.

Teams, OneDrive, SharePoint and Office applications can withdraw from this gateway when it fails and reroute their traffic. That capability allowed them to recover partially at an earlier stage. According to the report, Exchange Online’s traffic volume prevents it from switching elsewhere, leaving the service dependent on the gateway throughout the outage.

That difference explains why Teams calls, meetings and messages broadly continued to work, unlike functions dependent on Exchange. Sending and receiving emails, calendar scheduling, meeting invitations and Copilot’s access to Outlook data were disrupted.

Microsoft traces the incident to a flaw in a Windows process, introduced in July by a patch designed specifically for the gateway’s servers. Under heavy load, multiple tasks competed for the same internal lock, exhausting processor resources and memory until the machines stopped responding.

The flaw went undetected for more than three months. According to the report, Microsoft’s load tests used less powerful hardware than its production servers. On 12 November, additional load from Teams, combined with around 14% of servers having been taken out of service the previous day for an unrelated reason, exposed the problem during peak US traffic.

Microsoft acknowledges that its protective mechanisms then amplified the outage. The traffic management system redirected requests from overloaded servers to machines still operating, which then encountered the same flaw. Within two hours, the entire fleet was affected.

The gateway is divided into regions, with isolation barriers intended to contain this kind of failure. A separate problem, not detailed in the preliminary report, prevented those safeguards from working, allowing the outage to spread worldwide.

According to Charles Lamanna, recovery was slowed by queued requests immediately overwhelming restarted servers. Because the change had been in place for several months, the usual tools could not reverse it either. Microsoft says it found the solution at around 08:20 on Friday by restoring the affected code to its earlier state, before rolling it out in batches of servers.

The final report, expected on 20 November, is due to shed light on why the regional isolation barriers failed. It also remains unclear whether Exchange Online will stay dependent on a gateway it cannot bypass when it breaks down.