by Todd Bishop on Nov 16, 2026 at 7:31 amNovember 16, 2026 at 7:31 am.

Microsoft’s own safety systems helped turn a dormant software flaw into last week’s global Outlook outage, rerouting traffic from failing servers onto healthy ones that had the same problem, until more than 80% of the servers in its Microsoft 365 networking gateway had stopped responding.

That phenomenon, known as a cascading failure, is one of the findings in Microsoft’s preliminary report on the outage, released overnight, dissecting the failure that cut off email for people and organizations in more than 70 countries.

According to the report, the attempts to recover were further complicated by a second problem: servers that engineers restarted were immediately overwhelmed by waiting traffic and stopped responding again, a phenomenon known as a “thundering herd.”.

Industry observers note that microsoft says active service disruptions lasted nearly 24 hours. Delayed email continued to be delivered after service returned Friday morning, and the manufacturer posted its public all-clear on X late Friday night, nearly 38 hours after the outage began.

The underlying flaw dates to July, when Microsoft began rolling out a fix to the software that handles incoming connections on its networking gateway, the servers that sit in front of Outlook and many other Microsoft 365 services.

In a recent development, microsoft hasn’t stated what the fix was meant to address. It ran worldwide for more than three months without problems, according to the report.

Under heavy load, however, parts of that software could get stuck waiting on one another, using up the servers’ processing power and memory until they froze.

That’s what happened last week. About 15 hours before the outage, Microsoft’s systems automatically moved Teams traffic onto the same servers that handle Outlook email, adding to their load, according to the report. Microsoft called the switch part of normal operations but didn’t say what prompted it.

Complicating matters, nearly 14% of the gateway’s server capacity was already out of service because of an unrelated problem, which also isn’t detailed in the report.

In a recent development, when traffic peaked during the U.S. workday, that was enough to trigger the outage.

Two things stand out as likely areas of scrutiny:.

Microsoft’s testing didn’t catch the bug. Its labs mostly used less powerful machines with fewer processing cores, running a different mix of traffic, while the problem was worst on the newer servers handling real customer load.

The fix also passed through Microsoft’s standard staged rollout, spreading worldwide over several weeks without any sign of trouble, according to the report.

The organization certainly has the means to test on the hardware it actually runs. It spent $45.2 billion on capital expenditures in its most recent quarter, much of it on AI infrastructure.

The safeguards meant to contain the damage didn’t hold. Walls between regions failed because of what the report calls “a separate concern,” which it doesn’t explain, so the outage went global.

In a recent development, the gateway is made up of thousands of “capacity units” grouped into regions, with walls between them meant to keep a failure in one place. Microsoft also runs a second, independent gateway so services can switch over during a disruption.

It has been revealed that teams, OneDrive, SharePoint and Office apps made that switch, which is why Teams chats and meetings mostly kept working. Exchange couldn’t, “for scale reasons,” as the tech giant put it, so email stayed down.

Industry observers note that microsoft’s own tools added to the problem. Its automated recovery “was not designed” for a fleet that was almost entirely unresponsive, according to the report, and its usual ways of rolling back a bad change couldn’t quickly undo one that was months old.

The organization’s systems automatically shifted traffic from unresponsive servers to healthy ones, and the extra load triggered the same bug on those servers, too. About 10 hours in, engineers turned off some of that automatic rerouting.

Charles Lamanna, Microsoft’s executive vice president of Copilot, Agents and Platform, posted an abridged version of the report on X late Sunday. “We are deeply sorry for the impact to all our customers and recognize the impact that this has on their normal business & operations,” he wrote.

Lamanna had previewed the findings Friday afternoon, blaming “cascading failures + thundering herds at our network perimeter” for the slow recovery, as GeekWire reported when Microsoft declared the outage over.

Microsoft claims it has rolled back the defective code, added server capacity and monitoring, and put upcoming limits on how traffic is redistributed to avoid another cascade. A final report with a full list of repairs is due Nov. 20.

According to recent reports, it may also answer the biggest question the preliminary report leaves open: why the walls between regions didn’t stop the outage from spreading worldwide. Microsoft blames that on “a separate matter,” apart from the July bug, but hasn’t remarked what it was.

→ Read original article