A coding defect that sat undetected in Microsoft's cloud gateway since July froze more than 80 percent of the affected gateway servers and knocked Microsoft 365 and Exchange Online offline for just under 24 hours.
The failure ran from 16:20 UTC on November 12 to 16:10 UTC on November 13, with heightened monitoring holding through November 14. Email and productivity access collapsed as US business hours hit their daily peak, hitting companies and public offices that run on Microsoft's cloud.
Three problems stacked into one cascade. The July gateway change could exhaust CPU and memory under certain conditions, a flaw that escaped pre-release testing because the test hardware was weaker than the newer production machines carrying higher CPU core counts. Routine network maintenance then exposed it, shifting Teams traffic off the alternate gateway and onto the primary system that carries Exchange Online, which left servers handling more traffic and a different mix of requests. An existing fault had also left the fleet running 14 percent below optimal capacity, removing the headroom that would normally have absorbed the surge.
Once the servers began locking up, the backup gateways meant to catch displaced requests hit the same wall, and the regional isolation designed to contain the spread did not hold.
Restarts deepened the damage rather than ending it, because queued requests flooded back the instant servers returned and triggered the same resource exhaustion again. With the defect unknown since July, rolling back to a known-good configuration was impossible. Engineers had to ship a targeted update instead. Teams, OneDrive and other Office applications could be steered away from the failing infrastructure, but scaling limits blocked the same escape route for Exchange Online.
Patches went out gradually, additional capacity and traffic-distribution guardrails are now in place, and email delivery timeouts were extended to stop delayed messages from bouncing as undeliverable. All cloud services were operational again Monday morning. The full post-incident report is due November 20 and is expected to address why the regional isolation systems failed.