Microsoft 365 outage: what broke, and why it lasted so long

Hello humans.

On Thursday night in India, Outlook stopped loading for a lot of people. By Friday morning it was clear this was no small blip. Microsoft 365 was down or flaky in the US, India, Australia, Japan and beyond, and it stayed that way for hours. As I write this, late on Friday morning in India, Microsoft says it still isn't fixed.

Here's what happened, in plain words, and what I think businesses should take from it.

What happened

It started in the Americas. At 9:47 pm IST on Thursday 12 November (16:17 UTC), Microsoft began looking into reports that people couldn't reach their Exchange Online mailboxes. Just over two hours later, at 11:56 pm IST, it opened a wider incident, MO1550821, warning that users might not be able to connect to several Microsoft 365 services.

Downdetector, which counts problem reports from users, logged more than 130,000 Outlook reports in the US around 1 pm Eastern time. In India, Moneycontrol counted about 1,960 Outlook reports on Downdetector by 8:42 am IST on Friday, up from under 500 earlier. These are reports, not a count of everyone affected, so the real number is unknown.

The list of who was hit is long. Reuters reported that the US House of Representatives, Australia's spy agency and its federal police were caught up in it. Service NSW in Australia asked people to put off visits because some transactions weren't working. The electric truck maker Workhorse Group moved its quarterly results by four days. Even AI assistants felt it: OpenAI and Anthropic both reported errors in their Microsoft 365 connectors, the part that lets ChatGPT and Claude read your work email and calendar.

What broke, and what kept working

The pattern is the interesting part. Microsoft says the features that broke are the ones that talk to Exchange Online, its email service. Parts that don't depend on email kept working. Teams calls, meetings, group chats and channel messages mostly worked, and files stored directly in SharePoint stayed reachable.

One detail stood out to me. Microsoft posted its updates on a public status page, because admins might not be able to open the admin centre where those updates normally go. When your status updates live inside the system that's down, you need a second place to post them.

Why it happened, as far as we know

Microsoft hasn't named a root cause yet, and its explanation has changed as it learned more:

How Microsoft's explanation changed

1. 1

   Heavy loadThursday

   At first, Microsoft said its servers were under heavy load in several regions.

2. 2

   Not more traffic

   Then it said the load wasn't coming from more users. A large number of servers had dropped out without warning, so the rest carried more.

3. 3

   Failing health checks

   A core process on the affected servers was getting overwhelmed, so they failed their health checks, the automatic tests that decide whether a server gets traffic. A different way of checking health brought some of them back.

4. 4

   Fixes that didn't hold

   Restarted servers came back, then stopped answering once traffic reached them again. Microsoft paused deployments and looked at undoing a recent network change.

5. 5

   A surge of traffic insideFriday, 10:27 am IST

   Microsoft said its routing services are sending a surge of calls to the servers behind them. Microsoft is adding virtual machines and forcing the network change to sync, but some machines are stuck in a bad state.

6. 6

   Traffic controlsFriday, 11:47 am IST

   Its latest update says it is weighing traffic controls, such as throttling, protocol changes and failover routing, to steady the servers that face users.

Mahmoud Elkhodr, a senior lecturer at CQUniversity, explained the likely chain in The Conversation, while making clear it's an inference from Microsoft's own updates and not a settled account. Think of a call centre where a supervisor sends callers only to agents who look ready. If the supervisor's check goes wrong and marks good agents as busy, the few agents left get every call. They slow down, fail the check too, and the queue keeps growing. Bringing an agent back doesn't help if the flood of calls knocks them straight out again. That matches what Microsoft saw when restarted servers fell over as soon as traffic returned.

It doesn't look like an attack. Microsoft said there was no sign of one, its latest update says it knows of no malicious activity behind it, and Australia's cyber security minister, Tony Burke, said Microsoft believes it wasn't a cyberattack. An attack that floods servers usually brings more traffic, and Microsoft says this didn't.

What I'd take from it

I build AI agents and automations that sit inside the tools teams already use, so outages like this are what I plan for. These are my own takeaways, not Microsoft's.

Know what depends on your email. Calendars, Copilot, security tools and even third-party AI assistants failed here because email failed. Most teams don't know how much of their day runs through one service until it stops. Write it down.

Have a second way to talk. When Outlook is down and Teams can't schedule a meeting, people still need to reach each other. For most Indian teams that's a WhatsApp group or a phone list. Agree on it before you need it.

Post your status somewhere else. Microsoft did this right. If your customers or staff check a status page, it shouldn't run on the same system it reports on.

Make AI agents fail out loud. An agent that reads a mailbox should say clearly when it can't reach it, wait, and try again later. It should never guess or carry on with half the data. This is the part I'd test first in any agent built on Microsoft 365 or Google Workspace.

Plan for outages instead of running from them. Microsoft 365 also had long outages in January and August this year, and every big provider has bad days. Moving everything to another provider just swaps one single point of failure for another. A fallback for the few things you can't live without works better.

Where it stands now

As of Microsoft's update at 11:47 am IST (06:17 UTC) on Friday 13 November, the outage isn't fixed, and Microsoft hasn't given a time for full recovery. Once it's mitigated, Microsoft has promised a post-incident report with the root cause and the repairs it plans. That report should say what really caused it and why the first fixes didn't hold.

Want your systems and AI agents to keep working when a provider goes down? See what I build, read what a forward deployed engineer does, or get in touch.

Sources

Microsoft service health status (Microsoft's updates of 04:57 and 06:17 UTC on 13 November), and the Microsoft 365 Status post on Exchange Online (EX1550543). Massive Microsoft 365 outage disrupts Outlook, Teams, Copilot worldwide, BleepingComputer. Microsoft has suffered a global outage. A tech expert explains what went wrong, Mahmoud Elkhodr, The Conversation. Microsoft outage drags on with services still unstable, Reuters via RTÉ, and Microsoft outage drags on for hours, Reuters via The Korea Times. Microsoft Outlook outage hits 10 hours and why even ChatGPT felt it, Open. Microsoft outage: what the company said, Digit. Global Microsoft outage disrupts Outlook, Teams and cloud services, Australian Cyber Security Magazine.