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

- Channel: [Fireship](https://www.youtube.com/channel/UCsBjURrPoezykLs9EqgamOA)
- Published: 2026-11-16
- Source video: [Watch on YouTube](https://youtube.com/watch?v=ytmawkPTP24)
- Read on ReadTube: https://www.read.tube/p/videos/ytmawkPTP24
- Language: Original (source language)

## Summary

**Microsoft’s 36-Hour Outage Explained**

Microsoft’s 36-hour Exchange Online outage was caused by a network capacity upgrade that exposed a latent bug, overwhelming secure-connection processing and triggering a cascading server failure. Restarting the systems did not help because reconnecting clients immediately overloaded them again, while the outage also disabled Microsoft’s administrative tools. The disruption affected services and organizations worldwide, but the later FAA flight-grounding event was a separate network failure with a similar underlying pattern. Neither incident was blamed on AI, underscoring how routine infrastructure changes and human-made software defects can still cripple critical systems.

Microsoft’s Exchange Online outage began when a network capacity upgrade exposed a latent bug in the system that establishes secure connections. Connection requests overwhelmed the available processing power, causing servers to fail automated health checks and disappear from the load-balancing pool. The remaining servers then inherited more traffic, failed in turn, and created a cascading outage across Outlook, Copilot, SharePoint, and other Microsoft 365 services.

Restarting the servers was ineffective because the entire global client population immediately tried to reconnect, overwhelming them again. Microsoft also lost access to its 365 admin center, leaving administrators unable to use the tools needed to investigate the failure. The company eventually rolled back the network configuration and introduced an alternative health-check method to restore service, but the outage lasted roughly 36 hours.

The failure had broad operational consequences, affecting users in 74 countries and forcing some organizations to return to paper records or handle payments in person. Microsoft Teams continued operating, but the disruption still caused an estimated $7 billion in lost productivity. Microsoft said affected customers could seek service credits through support.

The Federal Aviation Administration’s nationwide flight-grounding event occurred soon afterward but was not connected to Microsoft’s outage. Officials attributed the FAA disruption to a network problem involving a router replacement, repeated login attempts, and possibly older Cisco equipment, although the agency had not released a final root cause. Both incidents nevertheless illustrate how a network change can overwhelm a critical system on which many other services depend.

The video emphasizes that neither incident was attributed to artificial intelligence, despite Microsoft’s recent workforce reductions and expanded investment in AI infrastructure. The central lesson is that routine infrastructure changes and hidden software defects can still produce massive failures when systems are highly centralized and their safeguards fail together.
