Microsoft published on November 15, 2026 the preliminary report of the global outage which deprived its customers of Outlook for almost 24 hours. The document explains why Teams held up while Exchange Online fell.
It’s time to take stock for Microsoft.
On Thursday, November 12, 2026, the American company experienced one of the longest outages in its recent history. Starting at 5:20 p.m. (Paris time), users of Exchange Online, the service that runs Outlook in businesses, began to lose access to their mailboxes around the world. The outage ended up affecting Copilot, SharePoint, OneDrive and several security tools like Defender, and it took until 5:10 p.m. the next day for no customers to be affected.
On November 15, the American company published the preliminary report of the incident in the Microsoft 365 admin center, accessible only to administrators of client companies.
The next day, Charles Lamanna, executive vice president of Copilot, agents and platform, published an abridged version on X.
Black Friday: your digital serenity at -61%
Secure your purchases during Black Friday with the best offer of the year. Find the good deals, Bitdefender takes care of the scams.
The document reviews the technical sequence of the outage and answers a question that many users asked themselves during these 24 hours, that of knowing why Teams continued to work while Outlook and everything that depended on it no longer responded.
Teams could leave the gateway, not Exchange
To understand, you need to know that many Microsoft 365 services pass through the same network gateway. It is through this infrastructure that user connections pass before reaching their emails, files or conversations. It was she who collapsed on November 12, to the point that at the height of the incident more than 80% of its servers no longer responded.
Teams, OneDrive, SharePoint, and Office apps have the ability to opt out of this gateway when it fails and route their traffic through another path, allowing them to partially recover sooner. Exchange Online does not have this option. According to the report, the volume of traffic it represented prevented it from switching elsewhere, and so it remained plugged into a down infrastructure for the duration of the incident.
This explains the discrepancy felt by users. Calls, meetings and Teams messages generally worked, but everything that relied on Exchange fell with it, whether it was sending and receiving emails, scheduling in the calendar, meeting invitations or Copilot’s access to Outlook data.
A bug that remained invisible and protections that spread the outage
The report also details what caused the footbridge to fall. Originally, there was a defect in a Windows process, introduced in July by a patch designed specifically for this gateway’s servers. Under heavy load, multiple tasks competed for the same internal lock until they exhausted the processor and memory of the machines, which then stopped responding.
A defect that remained invisible for more than three months. The fault is the load tests carried out by Microsoft which ran on hardware less powerful than that of production. On November 12, the additional load brought by Teams and the shutdown of about 14% of servers the day before, for an unrelated reason, ended up making it appear at peak US traffic.
Microsoft also admits that the protection mechanisms then made the situation worse. The traffic management system redirected requests from saturated servers to those that were still functioning, and the latter in turn fell into the same fault. In less than two hours, the entire fleet was affected. The gateway is however divided into regions separated by partitions supposed to contain this type of failure, but another problem, which the report does not detail, prevented them from functioning, and the outage became global.
Charles Lamanna explains that the restart took time because the restarted servers were immediately overwhelmed by pending requests, and because a change that had been in place for months could not be undone with the usual tools. Microsoft only found the solution around 8:20 a.m. on Friday, reverting the affected code to its previous state, and then deployed it in batches of servers.
The final report, expected on November 20, will notably have to explain why the partitions between regions did not work, and say whether Exchange Online will remain dependent on a gateway from which it cannot escape when it falls.
All the tech news at a glance
Add Numerama to your home screen and stay connected to the future!

