This page lists the major cloud outages of 2026 through November 13, with the cause each provider gave. Power and cooling failures caused the longest regional outages. Configuration and software changes caused the ones that spread furthest. Azure, Google Cloud and Cloudflare published reports for the incidents below. AWS has published no post-event summary for any 2026 incident, so AWS entries rely on press coverage.
Microsoft Azure
| Date | Where | Duration | Cause, from Microsoft's report |
|---|
| Feb 2-3 | Multiple regions, virtual machines and managed identities | About 6.5 hours, longer for managed identities | A data synchronisation problem in targeting logic |
| Feb 7-8 | West US | About 20 hours 26 minutes | An electrical failure in an onsite transformer cut utility power |
| Apr 24 | East US control plane | About 11 hours 52 minutes | A regression in a newly deployed service version |
| May 29-30 | West US 2 | About 22 hours | A severe thunderstorm disturbed utility power |
| Jul 23 | West US network | 4 hours 57 minutes | A defect widened the scope of a network repair. Microsoft 365 was hit downstream |
| Sep 29 | Sweden Central, including Azure OpenAI | About 5 hours 55 minutes | Report read only in part |
| Sep 30 to Oct 1 | Multiple regions, gateway services | About 5 hours 45 minutes | Request volume and slower processing overwhelmed the control plane |
The Microsoft 365 outage of November 12-13 does not appear on the Azure status history. It was a Microsoft 365 incident.
Google Cloud
| Date | Where | Duration | Cause, from Google's report |
|---|
| Feb 27 | Global endpoint, including Gemini on Agent Platform | 1 hour 58 minutes | Not read |
| Jun 5-26 | Delhi region networking | 21 days 12 hours | A fire at a third-party data centre forced an emergency power shutdown of network equipment |
| Jul 15-16 | Netherlands, one zone | 14 hours 55 minutes | A 3 millisecond voltage drop in the utility power feed |
| Aug 20 | Oregon region, about 25 products | 2 hours 22 minutes | Scheduled fibre maintenance reduced network capacity unexpectedly |
| Sep 1 | Iowa region, two zones | 4 hours 11 minutes | Fibre cables were disconnected by mistake during hardware maintenance |
| Oct 8 | Cloud Storage writes | 1 day 6 hours | Not read |
| Oct 19-20 | London, one zone, about 35 products | 6 hours 8 minutes | A misconfiguration in power control logic |
Amazon Web Services
| Date | Where | What was reported |
|---|
| Mar 1-3 | UAE and Bahrain regions | AWS said objects struck a UAE data centre and caused a fire, and that a drone strike near a Bahrain facility damaged infrastructure, according to The Register |
| May 7-8 | Northern Virginia, one zone | A power loss during what AWS called a thermal event. Coinbase halted trading, according to The Stack. Reported durations range from 12 to 28 hours |
AWS's post-event summaries page had no 2026 entry when we read it. Its most recent was for October 2025.
Cloudflare
| Date | What | Duration | Cause, from Cloudflare |
|---|
| Jan 22 | Route leak in Miami, IPv6 only | 25 minutes | A routing policy change that was too permissive |
| Feb 20 | About 1,100 customer IP prefixes withdrawn, a quarter of those customers bring themselves | About 5 hours by the report's timestamps | A change that unintentionally withdrew customer prefixes. Cloudflare said it was not an attack |
| Oct 20 | Cloudflare One Gateway | About 3 hours 12 minutes | Not stated |
Oracle Cloud
| Date | Where | Duration | Cause |
|---|
| Mar 3-4 | US East (Ashburn) | About 20 hours | Not published. TikTok's US service was affected, according to The Register |
Patterns in 2026
- Power is the slow failure. Five of the incidents above trace to power or cooling, and four of those lasted more than 14 hours.
- Changes are the wide failure. Deployments, configuration changes and maintenance account for most multi-region incidents.
- Physical risk is now on the list. A fire took Google's Delhi networking offline for three weeks, and AWS reported strikes on Gulf facilities.
- Reporting differs by provider. Azure, Google and Cloudflare publish incident reports with causes. For AWS and Oracle in 2026, the public record is mostly press coverage.
For Comparison: Late 2025
| Date | Provider | Duration | Cause |
|---|
| Oct 19-20, 2025 | AWS Northern Virginia | About 14 hours 32 minutes | A race condition in the DynamoDB DNS management system |
| Oct 29-30, 2025 | Azure Front Door, global | 8 hours 24 minutes | Incompatible customer configuration metadata |
| Nov 18, 2025 | Cloudflare, global | About 5 hours 46 minutes | A change to database permissions |
Methodology
Azure entries come from the Azure status history, Google Cloud entries from its incident summary, and Cloudflare entries from the Cloudflare blog and status page. AWS and Oracle entries come from The Register and The Stack, because no provider report was found. The list covers incidents we could confirm and is not complete. Causes are paraphrased from the provider reports. Status as of November 13, 2026.