IncidentVPS and app hostingNews from
Azure West US outage: a widened network repair cut traffic for five hours
On July 23, 2026, traffic into and out of Azure's West US region failed or slowed from 14:44 to 19:41 UTC. Microsoft's review blames a defect that widened an optical repair to a whole datacenter.
Microsoft Azure’s West US region had a network outage on July 23, 2026. Between 14:44 and 19:41 UTC, some customers saw connection failures and higher latency for traffic entering or leaving the region. Microsoft’s Post Incident Review (tracking ID ZJV6-SGG) says traffic that stayed inside the region was not affected.
What was affected
The review lists many services, among them App Service, Application Insights, Azure AD B2C, Azure AI Bot Service, Azure Cosmos DB, Azure Kubernetes Service, Azure Firewall, ExpressRoute, VPN Gateway, Microsoft Sentinel and Power BI Embedded. Because only traffic crossing the region’s edge was hit, workloads talking to each other inside West US kept working.
What went wrong
A routine break-fix repair started on one optical network device. A defect in Microsoft’s blast radius analysis system widened the repair to cover every optical device in one datacenter. The safety check looked at each device on its own, not at the combined effect, so network routes were withdrawn from many devices at once. Automated recovery could not put them back, because it relied on the same connectivity that had just been cut.
Timeline (UTC, July 23, 2026)
| Time | Event |
|---|---|
| 14:44 | Repair starts, routes are withdrawn, impact begins |
| 15:26 | A routing alert is triaged as a wide-area network issue |
| 16:58 | Automated recovery hands over to engineers after failed retries |
| 17:19 | Engineers link the routing problem to the repair |
| 17:45 | Manual rollback starts |
| 18:26 | Rollback finishes on all affected devices |
| 19:41 | Microsoft confirms the impact is mitigated |
What Microsoft is changing
Microsoft says it has fixed the blast radius defect, added checks so both diverse network paths of a datacenter can’t be under repair at the same time, and scanned existing repair requests for the same problem. Further steps were scheduled for August to October 2026: logging repairs in its change ledger, better traffic monitoring, tooling guardrails, and automated recovery that fails faster and escalates sooner.
What to do
If your app runs only in West US, this kind of outage blocks your users even while the servers stay up. Keep a copy of critical services in a second region, and point health checks and DNS failover at something outside the region.
Sources
Read on . Prices are in US dollars. How we check facts