Two large cloud incidents landed in four days, from November 12 to November 15, 2026: a Microsoft 365 outage centered on Exchange Online, and a three-hour sign-in and connectivity failure in the AWS US-EAST-1 (Northern Virginia) Region. This piece reflects what the two vendors had published as of November 16, 2026. Microsoft's account is preliminary and AWS had stated no cause, so details may change.

The incidents side by side

Incident Start (UTC) End (UTC) Services Stated cause
Microsoft 365 (MO1550821, EX1550543) November 12, 2026, 16:20 November 13, 2026, 16:10 Exchange Online and products that depend on it Preliminary: latent software defect on gateway servers
AWS US-EAST-1 November 15, 2026, 11:05 November 15, 2026, 14:10 or 14:21 11 listed, led by IAM Identity Center None stated as of November 16, 2026

Microsoft 365, November 12 to 14, 2026

Microsoft's full record sits in the Microsoft 365 admin center, behind a tenant sign-in. On November 16, 2026 the Microsoft 365 Status account on X said a Preliminary Post-Incident Report had been released there. What follows comes from that report, dated November 15, 2026, as reproduced by Office Watch and reported by The Register, and from the MO1550821 advisory as mirrored by Cloud Admin Hub and quoted by BleepingComputer.

  • November 12, 16:20 UTC. Microsoft opened EX1550543, its stated start, after monitoring flagged connectivity problems at 16:03 UTC.
  • 18:27 UTC. MO1550821 followed, listing Exchange Online, Microsoft Teams, SharePoint Online, Microsoft Copilot 365 and Microsoft Defender XDR.
  • 19:43 UTC. The advisory said servers were "failing automated health checks, which is causing them to become unavailable", shifting load onto the rest.
  • November 13, 07:20 UTC. Engineers began applying an update that reverted the defective change.
  • 16:10 UTC. Stated end of active customer impact, 23 hours 50 minutes after the start.
  • November 14, 04:15 UTC. Final customer update.

Sources conflict on the end. Microsoft gives 16:10 UTC on November 13, yet its advisory at 20:01 UTC that day said "We do not have an ETA for full recovery yet", and The Register reported service restored to all customers early on November 14, 2026 GMT.

The stated cause also moved. At 17:57 UTC on November 13, 2026 the advisory blamed "an issue during an earlier network capacity upgrade". The preliminary report lists three factors: a latent defect, from a fix whose deployment to the gateway servers began in July 2026, that made servers unresponsive under certain load; an automated failover of Teams traffic onto the same gateway about 15 hours earlier, which added load; and nearly 14% of gateway capacity already out of service for an unrelated reason. Traffic management then redirected users to healthy servers, which hit the same defect: "a cascading failure from server to server". At the lowest point more than 80% of servers were unresponsive.

As of November 16, 2026 the final report, promised within five business days of full resolution, had not been published, and Microsoft had not explained the "separate issue" that let the failure spread past the gateway's internal partitions.

AWS US-EAST-1, November 15, 2026

The AWS Health Dashboard event "Increased Connectivity issues in US-EAST-1" lists 11 services: AWS IAM Identity Center, AWS Sign Up, Amazon Elastic Load Balancing, AWS NAT Gateway, Amazon VPC Lattice, Amazon CloudWatch, Amazon Connect, Amazon OpenSearch Service, AWS Step Functions, AWS IoT Device Management and Kiro. Times are converted from AWS's PST.

  • 11:05 UTC (3:05 AM PST). Sign-up, authentication and connectivity errors began, per AWS's closing summary.
  • 12:38 UTC. First dashboard post, 93 minutes later.
  • 13:15 UTC. AWS confirmed sign-in errors through IAM Identity Center and services relying on it: "There are no workarounds for impacted customers."
  • 13:20 UTC. Authentication mitigated; most services recovered.
  • 14:21 UTC. A subset of Network Load Balancers in one Availability Zone, use1-az1, mitigated.
  • 14:53 UTC. Final update: resolved.

AWS's summary carries two end times: an impact window closing at 14:10 UTC and the load balancer fix at 14:21 UTC. As of November 16, 2026 AWS had stated no cause, and its Post-Event Summaries page listed nothing for this event.

What the two have in common

  • A shared front door failed. Microsoft lost a gateway that fronts many Microsoft 365 services, and products as unrelated as Planner and Universal Print were listed as affected. AWS customers who sign in through IAM Identity Center lost the Console, Kiro and Amazon Connect.
  • Neither account describes lost hardware or data. Microsoft held delayed mail by extending its email timeout window to 48 hours; AWS told customers to retry failed requests.
  • Protection spread the failure. In Microsoft's account, health checks pulled servers out and redirection loaded the survivors until they failed too.
  • The response tools were affected. Microsoft's advisory said some pages of the admin center, where it is published, might not load.

Your own dependency map

A site on a rented server can still stop when a cloud service it leans on fails. List what your services call at run time:

  • Login through a third-party identity provider
  • Transactional email
  • DNS hosting and the registrar
  • Payment processing
  • Object storage, including backups
  • A CDN
  • Monitoring and alerting
  • CI and deployment

For each line:

  1. Write down the provider and region.
  2. Mark it "stops the business" or "degrades it". Checkout without payments stops; a site without its CDN usually only slows.
  3. Note the fallback and the date it was last tested.
  4. Do the same for what your team needs to respond: alerts, chat, the password manager.

What actually helps

  • A second provider or region for the one critical path, not for everything. Duplicate the path that takes money; let the rest fail and recover.
  • DNS with sensible TTLs and a tested failover. Short TTLs on records you would repoint, DNS hosted apart from what it fails away from, and a failover you have actually run.
  • Status checks from outside your provider. A probe on another network, alerting by SMS, does not wait for a vendor's status page.
  • An out-of-band way to reach your team. When email and chat are the outage, you need a phone list or second messaging service, agreed in advance.
  • A written, rehearsed runbook. Who decides to fail over, how, and how to fail back.

The honest counterpoint

Leaving a cloud does not remove single points of failure; it changes whose they are. One dedicated server has one disk controller, one power feed and one upstream network, and when any of them fails the outage is yours. Physical loss is a failure mode too: the October industry roundup covered data center sites physically disabled in October 2026.

Redundancy costs money wherever you buy it, as a second server, a second region or a second provider.

Questions to ask any provider, cloud or not

  • What do you publish during an incident, and how fast? AWS first posted 93 minutes after impact began; Microsoft, 17 minutes after its monitoring flagged trouble. Ask whether the status page depends on the affected platform.
  • Are post-incident reviews public? Microsoft's preliminary report was published inside its admin center; AWS keeps a public Post-Event Summaries page.
  • What does the service credit cover? Is it automatic or claimed, is it capped at the fee for the affected service, and does it exclude lost revenue?
  • Which of your own dependencies sit outside your control? Upstream carriers, power, DNS, identity.

Sources