Microsoft 365 outage traced to three combined gateway failures
Microsoft's preliminary report points to a July code change, a routine traffic switch and a gateway fleet running below capacity. Exchange Online took longest to recover because it could not be moved to other infrastructure.

Key takeaways
- Microsoft's preliminary report blames a gateway defect from July, a routine Teams traffic shift and a fleet running 14 percent below optimal capacity.
- The outage ran from 16:20 UTC on 12 November 2026 to 16:10 UTC on 13 November 2026, just under 24 hours.
- Over 80 percent of affected gateway servers froze, and restarts and rollbacks did not clear the backlog of requests.
- Exchange Online recovered last because scaling limits stopped it moving elsewhere, and a final report is due on 20 November 2026.
In this article
Three problems combined at the gateway
Microsoft has published a preliminary account of the Microsoft 365 outage that struck on 12 November 2026 and ran into the following day. According to Ars Technica, which reviewed the company's Post-Incident Report, three separate problems lined up at once: a hidden software defect, a shift in network traffic and less spare server capacity than usual.
The report puts the outage window at 16:20 UTC on 12 November 2026 to 16:10 UTC on 13 November 2026, just under 24 hours. Extra monitoring continued on affected systems through 14 November 2026. Microsoft classes the event as a technical fault with no connection to a cyber attack or to AI agents, Ars Technica reports.
The disruption reached businesses and public offices that depend on Microsoft's cloud services and its Exchange Online email systems. Ars Technica describes it as among the biggest failures in the company's history.
A gateway bug dormant since July

The root cause was a change to Microsoft's cloud gateway. This is the layer that sits in front of services such as Exchange Online and manages incoming connections. Ars Technica reports that the change was rolled out from July and created a defect that could, in certain conditions, drain a server's CPU and memory until the machine stopped responding. A CPU is the processor that carries out a server's instructions.
Microsoft says the flaw was never caught in pre-release testing. Its test machines were less powerful than some of the hardware already running in production, according to the report. Newer servers with higher CPU core counts appear to have made the lock-up more likely, Ars Technica reports.
Traffic shift and missing capacity
Server load was the second factor. Microsoft runs two independent gateway systems so that people can still reach Microsoft 365 if one of them struggles. Ars Technica reports that about 15 hours before the outage, Teams traffic was moved from the standby gateway back to the primary one, which also carries Exchange Online access.
The move was a routine fail-safe, meaning a standard precaution. It left one gateway handling more traffic and a different mix of requests, and servers began turning away new connections as demand built up.
Capacity was the third factor. An existing fault meant the gateway fleet was running 14 percent below its best operating level, Ars Technica reports. That missing headroom would normally have soaked up the extra demand. With the traffic change and the July defect added on top, requests piled up and the gateways stalled. The report describes the result as lock contention, where requests wait on each other and progress stops.
How the freeze spread and why fixes failed
As US businesses hit their working peaks, gateway servers began to freeze. Backup systems pushed requests away from unresponsive machines, but those requests met the same lock-ups on the standby gateways. Ars Technica reports that over 80 percent of the affected gateway servers ended up frozen.
Controls designed to contain a fault within one region do not appear to have worked. The preliminary report does not explain why, according to Ars Technica.
Recovery was slow because restarts did not help. When a server came back online, the backlog of waiting requests was still there and drained its resources again. Rolling back the July change was not an option either, since the faulty update had never been identified. Ars Technica reports that engineers had to write a targeted update instead.
Teams, OneDrive and other Office apps could be shifted onto different infrastructure. Exchange Online could not do the same because of scaling limits, so email was the last service to return. Microsoft also extended delivery timeouts for affected messages, so delayed mail would not bounce back as undeliverable.
What Microsoft has changed and what comes next
Microsoft says servers were patched and brought back into service gradually. The company has added extra capacity and new guardrails on how traffic is spread across gateways, Ars Technica reports. Its cloud services all appeared to be working normally by the morning of 16 November 2026.
A full final report is due on 20 November 2026. It may explain why the regional isolation controls failed and set out further corrective steps. The preliminary version is not fully public: Microsoft's status pages require Microsoft 365 administrator sign-in, under incident IDs EX1550543 and MO1550821.
For IT teams, the reported sequence shows how a small latent defect can turn into a wide outage when routine traffic changes and thin capacity margins arrive together. Ars Technica notes that Office Watch has reproduced the preliminary report, and that the University of Pennsylvania keeps an archive of Microsoft's service-health updates for readers who want the primary detail.
Sources
Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.



