When the cloud goes down. What keeps working in the building and what stops
From 12 to 14 November 2026, Outlook, Teams and Copilot were down in more than 70 countries. Problems were reported by public offices, courts, hospitals, banks and airlines1. Microsoft explained that the cause was an error following an earlier network expansion, not a cyberattack or AI2.
It was not an exception. In the last two years all the largest providers have had serious outages.
Microsoft 365: more than 12 hours of disruption to key services, with problems lasting until 14 Nov 2026. Google: main fix approx. 3 h, full recovery approx. 7.5 h.
Source: The New Stack, KUOW, ThousandEyes
Data table
| Outage | Duration |
|---|---|
| Amazon Web Services, Oct 2025 | approx. 15 h |
| Microsoft 365, Nov 2026 | 12+ h |
| Google Cloud and Workspace, Jun 2025 | approx. 7.5 h |
In October 2025, Amazon Web Services' main US region had problems for almost 15 hours, taking apps, banks and games around the world with it3. In June 2025 a faulty change to Google Cloud's permissions system took down Gmail, Google Drive and Meet, among others. The main fix took about 3 hours, and full recovery about 7.5 hours4. Earlier, in July 2024, a faulty CrowdStrike update stopped Windows computers worldwide and disrupted hospitals, airports and banks for several days5.
What keeps working and what stops
| Situation | Everything in the cloud | Own server and backup connection |
|---|---|---|
| Cloud provider outage | email, files and calendar unavailable | files and local tools work, only cloud email waits |
| Internet outage at the operator | work stops | the backup connection takes over, everything in the building works |
| Power cut | depends on the provider, nothing in the building works anyway | a UPS gives time for a safe shutdown |
| Hardware failure in the building | not applicable | a second server or a backup takes over, cloud email unaffected |
No option protects against everything. The point is that a single failure should not stop the whole institution.
What to prepare
- A backup internet connection from another operator or using another technology, such as a radio link.
- A UPS for the server and network equipment.
- Copies of the most important documents in the building, including those normally kept in the cloud.
- Contacts and procedures on paper, so that during an outage you know whom to notify and what to do.
- Spare hardware or a backup that lets you restore work quickly after a server failure.
The wider background to this shift is described in To the cloud and back.
Sources
- Microsoft says services fully restored after hourslong global outage, ABC News
- As Microsoft services return online, company says outage wasn't caused by AI or cyberattack, KUOW, 13 November 2026
- A cascade of failures: a breakdown of the massive AWS outage, The New Stack
- Google Cloud outage analysis: June 12, 2025, ThousandEyes
- AWS outage explained: how it differs from past CrowdStrike or BGP internet outages, Digit