---
title: 'Why Microsoft couldn’t fix its own servers for 36 hours...'
source: 'https://youtube.com/watch?v=ytmawkPTP24'
video_id: 'ytmawkPTP24'
date: 2026-11-17
duration_sec: 382
channel: 'Fireship'
---

# Why Microsoft couldn’t fix its own servers for 36 hours...

> Source: [Why Microsoft couldn’t fix its own servers for 36 hours...](https://youtube.com/watch?v=ytmawkPTP24)

## Summary

This video analyzes the massive Microsoft 365 outage of November 2026, which took down Outlook, SharePoint, and other services for 36 hours. It explains the technical root cause—a cascading load balancer failure—and contrasts it with a separate FAA system outage, debunking any connection between the two events.

### Key Points

- **Outage scope and impact** [00:02] — Exchange Online outage lasted ~36 hours, affecting Outlook, Copilot, SharePoint, and other Microsoft 365 services. Down Detector saw reports from 74 countries within an hour; Wall Street traders reverted to phones and fax machines.
- **Initial failure and misdiagnosis** [01:07] — At 11:17 AM ET, Exchange Online mailboxes stopped opening. Microsoft initially blamed elevated load, then clarified it was fewer servers due to 'unexpectedly unavailable' servers.
- **Root cause: cascading load balancer failure** [01:32] — Load balancers ping servers and pull unhealthy ones out of rotation. When many servers fail at once, traffic lands on survivors, which get overloaded and also fail—a domino effect. Servers failed automated health checks, and the secure-connection process got overwhelmed.
- **Failed fix attempts** [02:12] — Microsoft tried rebooting servers, but they became unresponsive again once traffic resumed. They then bypassed health checks, which is a workaround rather than a true fix. The outage also took down admin tools, complicating remediation.
- **Real-world chaos** [02:52] — Government agencies reverted to paper records, a Colorado jail took bond payments in person, and the entire country of Australia was affected (already Friday there). Microsoft Teams kept working like an 'invincible apocalyptic cockroach.'
- **Separate FAA outage** [03:04] — The FAA grounded all departures for ~5 hours, the second time since 9/11, causing 9,000+ cancellations. Officials said it was not a cyber attack and not connected to Microsoft. Blame was placed on network equipment overload, possibly from an L3 Harris contractor or old Cisco routers.
- **Microsoft's post-mortem** [04:07] — Microsoft started rolling back a change on Friday, apologized, and attributed the issue to a latent bug in the secure-connection setup introduced during an earlier network capacity upgrade. The upgrade meant to add capacity removed it instead.
- **Human error, not AI** [04:36] — Microsoft and the FAA emphasized the issues were not caused by AI or software 'shitification' but by human error. This is notable because Microsoft recently offered voluntary exit packages to ~60,000 employees (a quarter of the company) to build more data centers for AI agents.
- **Financial impact** [05:02] — The outage caused roughly $7 billion in damages, leading to potential compensation claims. The video humorously suggests not to email the claim.

### Conclusion

The video concludes that while the Microsoft and FAA outages share a common pattern—network overload on a critical box—they are not directly related. It highlights the fragility of centralized systems and the irony that a capacity upgrade caused the capacity loss.

## Transcript

boosted global productivity and made the world a better place by breaking Outlook email for most of the planet. Exchange Online, the back end behind it, Coast and stayed dead for roughly 36
hours, dragging Copilot, SharePoint, and about a dozen other Microsoft 365 services into the grave with it. And now legally, they're forced to rename it Microsoft 364. The down detector saw reports from 74 countries in about an
hour. Traders on Wall Street went back to phones and fax machines. And it all out of Nvidia. &gt;&gt; Don't let Outlook control what you do &gt;&gt; What are the chances of that? In fact, he got to be CEO for a whole extra day
with Outlook. But then on Friday the Federal Aviation Administration grounded every departing flight in the declared all these events were connected. In today's video, we'll find
Microsoft. Why turning it on and off again didn't work and whether or not Jensen's Wong is the true cause of this catastrophic outage. It is November 16th, 2026, and you're watching the code report. The dominoes began to fall at
exactly 11:17 Thursday morning, Eastern time, when Exchange Online mailboxes stopped opening. And then just a mere 90 minutes later, Microsoft posted that it Americas, which was accurate in the sense that the Americas are part of the
an elevated load, but possibly from Jensen's Wong. But then they quickly clarified that it wasn't more traffic. It was fewer servers because a large number of them had become quote
unexpectedly unavailable. To understand the root of the issue, we first need to load balancer constantly pings every server fails, it gets pulled out of the rotation, which is great when one server
once, all their traffic lands on the survivors who get overloaded and then get pulled out too, just like a bunch of dominoes toppling each other over. And Microsoft describe. The servers were failing automated health checks and the
process responsible for establishing secure connections got so overwhelmed that the server stopped responding. So Microsoft tried the classic solution that works 99% of the time. They unplug them and then plug them back in again.
words, the servers can become unresponsive again once they receive traffic because every single male client on planet Earth was standing outside announced they had an alternative health check method, which is just a BS way of
saying we're bypassing the health check to see if the servers work despite them being unhealthy. And to make matters worse, the outage also took out the 365 their admin tools to find out why the admin tools were down. And when NBC News
comment, the email bounced. Meanwhile, back in the real world, things were in total chaos. Government agencies had to go back to paper records, a jail in Colorado had to take bond payments in person, and the entire country of
because it was already Friday there. The only downside, though, is Microsoft Teams somehow kept working like an invincible apocalyptic cockroach. But then on Friday afternoon, all hell broke loose when the FAA halted all departures
nationwide for roughly 5 hours. And that's only the second time that's happened since 9/11. It resulted in more than 9,000 flight cancellations on Friday alone. And officials said it was not a cyber attack. And two giant
systems dying within 24 hours looks very suspicious, especially right after Nvidia's Wong just got laid off. Now, although I wish this was a catastrophic for views on this channel. It appears the two events are not related. The
Homeland Security Secretary said, quote, "There is no connection between this and Microsoft." Microsoft said it has no indication they're related. And the of switches and routers that got overloaded with data and then came out
later and blamed it on the L3 Harris defense contractor who swapped out a flooded all the other routers with about a thousand login attempts every four or 5 seconds. Other sources have put the blame on old Cisco routers. But as of
right now, the FAA hasn't released an official root cause. But even though the Outlook outage and the FAA problems aren't directly related, they share the the network overwhelms a box that everything depends on. And that brings
still waiting on the full post-mortem, but we know they started rolling back a on Friday, they said they were deeply sorry and that the incident was introduced by an issue during an earlier network capacity upgrade. Apparently,
latent bug in the system that sets up secure connections and that caused point where all the servers ran out of processing power, which is kind of supposed to add capacity is the thing that removed all of the capacity. But
Microsoft and the FAA have said is that these issues were definitely not caused by AI and shitification of software. It was definitely you stupid humans that definitely had nothing to do with this. that part at least they're 100% certain
about already. And it's very important for Microsoft to say that because less than 3 weeks ago they offered voluntary exit packages which are definitely not layoffs to about 60,000 employees, roughly a quarter of the company. And
build more data centers to run more AI agents to replace even more humans. But even though this outage wasn't caused by AI agents, it caused roughly $7 billion hurt by this, you may be entitled to compensation. You just have to reach out
credit. Although I wouldn't recommend doing that by email. A much better use of your time though would be to check out Open Router, the sponsor of today's video. It gives you a single API to access over 600 different models and can
automatically pick the best one for each request based on intelligence, speed, and cost. But the problem is that many of these models are served by multiple providers who are all competing to show the cheapest price per token, which is a
very gameable metric. So, Open Router just rebuilt their entire routing algorithm to score every provider on what a successful response actually costs you, including the real price after caching, how fast tokens come
back, and how reliably the requests succeed. Those scores refresh every few or starts rate limiting, it automatically routes your traffic somewhere else. So far, the new router is delivering about 20% faster time to
first token and 60% fewer rate limit errors with no code changes, which has made a noticeable difference in my weekly rewrites of Horse Tinder. Try it This has been the code report. Thanks for watching and I will see you in the
for watching and I will see you in the next one.
