LE PERSPECTIVE / THE STORY

If you run IT, you spent Thursday refreshing a status page. Starting around 11 a.m. Eastern on Nov. 12, Exchange Online stopped letting people into their mailboxes, first in the Americas, then anywhere. Outlook went with it, and Teams, Copilot, SharePoint and Defender caught the knock-on effects.

Downdetector logged around 140,000 reports from 74 countries, and those are only the people who bothered to complain. UK police forces told people not to file crime reports online. Britain's Insolvency Service lost email, and the NHS and Parliament felt it too. In Australia, the federal police and the country's spy agency were hit. Government offices in Japan went back to paper. Even OpenAI said its Microsoft connectors were acting up.

Every few years, one company's bad day becomes everyone's problem. This one earned a spot near the top of the all-time list, so we ranked the whole list.

Four of the Ten Happened in the Last Year
AWS in October 2025, Cloudflare a month later, AWS again in March and now Microsoft. Four of the ten worst outages ever landed in about 13 months. Same handful of vendors, bigger blast radius every time.

What actually broke at Microsoft

Microsoft says it wasn't an attack. Servers started failing their automated health checks and got pulled out of rotation, which dumped more load on the ones still standing. A core process that sets up secure connections got overwhelmed, so those servers stopped answering too. A textbook cascade.

Recovery was the painful part. Engineers brought servers back, traffic hit them and they fell over again. Overnight, Microsoft admitted a fix was "not taking as expected," froze all deployments and config changes, scaled up VMs and started patching region by region, beginning with places heading into business hours. It's also rolling back a network config change that lined up with the start of the outage, though it hasn't called that the cause. Delayed mail is being held for up to 48 hours so it doesn't bounce, and a full post-incident report is promised.

How we ranked them

Most outage lists rank by hours down or by how many people complained. We ranked by real-world damage: flights grounded, 911 lines dead, payments frozen, work stopped.

10. AWS Middle East, 2026: the outage a war caused

Number 10: AWS logo

WHEN
March 1, 2026

DOWN FOR
Months, for some services

WHAT BROKE
Drone strikes on data centers

On March 1, AWS reported that "objects" had hit one of its UAE data centers, causing sparks and a fire, and local authorities cut the power. A day later AWS said what everyone suspected: drones had directly struck two of its UAE facilities, and a strike near a third site in Bahrain had damaged infrastructure there too.

At the peak, more than 100 AWS services in the UAE region were down or degraded, and banks and consumer apps across the Gulf went with them. AWS told customers to migrate to other regions and later waived March charges for the UAE region. You can't roll back a burnt building, so recovery has dragged on for months.

9. Google, 2020: 47 minutes without a login

Number 9: Google logo

WHEN
Dec. 14, 2020

DOWN FOR
47 minutes

WHAT BROKE
The login system lost its quota

On a Monday morning in December 2020, Gmail, YouTube, Drive, Docs, Meet and Classroom all died at once, worldwide. Schools in the middle of pandemic remote learning pushed back classes, and every app using "Sign in with Google" locked people out too.

The cause was almost funny. During a migration to a new quota system, the old one told Google's central identity service it was using zero storage. Automation believed it and cut the quota. No storage, no logins, for 47 minutes.

8. Fastly, 2021: one customer, one setting

Number 8: Fastly logo

WHEN
June 8, 2021

DOWN FOR
Under an hour

WHAT BROKE
A dormant bug, one valid setting

On June 8, 2021, The New York Times, CNN, The Guardian, Reddit, Twitch, Amazon and the UK government's website all started throwing errors. The culprit was Fastly, a CDN most people outside tech had never heard of.

A bug had been sitting in Fastly's code since mid-May. Then one customer pushed a perfectly valid config change, and 85% of Fastly's network started returning errors. Fastly spotted it within a minute and had 95% of the network back in 49 minutes, which is genuinely fast. It still made front pages worldwide.

7. AWS S3, 2017: the typo that broke the internet

Number 7: AWS logo

WHEN
Feb. 28, 2017

DOWN FOR
About 4 hours

WHAT BROKE
One mistyped command

An AWS engineer debugging a slow billing system ran a command to pull a few servers. One input was wrong, and a much bigger chunk disappeared, including two subsystems S3 needs to run. S3 in us-east-1 was down for about four hours, and an estimated 150,000 sites and services went with it, from Slack and Trello to smart home gadgets.

Recovery dragged because, as AWS admitted, those subsystems hadn't been fully restarted in years. And AWS's own status dashboard couldn't show the outage properly, because it depended on S3 too.

6. Cloudflare, 2025: a file that got too big

Number 6: Cloudflare logo

WHEN
Nov. 18, 2025

DOWN FOR
About 3 hours

WHAT BROKE
A config file that doubled in size

Cloudflare sits in front of roughly a fifth of the web, so when it fell over on Nov. 18, 2025, X, ChatGPT, Spotify and a long tail of other sites fell with it. Even Downdetector went down, which is its own special kind of irony.

The trigger was a routine database permissions change. It made a config file for Cloudflare's bot-management system double in size. That file got pushed to the whole network, blew past a hard-coded limit, and the proxy that routes traffic crashed. Core traffic was back in about three hours, everything within six. Cloudflare called it its worst outage since 2019.

5. Rogers, 2022: the day Canada couldn't call 911

Number 5: Rogers logo

WHEN
July 8, 2022

DOWN FOR
About 15 hours for most

WHAT BROKE
A bad change in the network core

On July 8, 2022, Canadians woke up to phones with no signal and homes with no internet. A maintenance update in Rogers' core network made its routers fail, and more than 12 million customers went dark. By one estimate, about a quarter of Canada lost internet.

Some cities couldn't take 911 calls. Interac, the national debit network, went down, so shops couldn't take cards and ATMs stopped working. Passport offices were hit. It took Rogers about 14 hours just to find the root cause. Most people were back after about 15 hours, but some waited days. Rogers later committed to splitting its wireless and wireline networks so one failure can't take down both.

4. Facebook, 2021: the company that locked itself out

Number 4: Facebook logo

WHEN
Oct. 4, 2021

DOWN FOR
About 6 hours

WHAT BROKE
A maintenance command cut its backbone

For about six hours on Oct. 4, 2021, Facebook, Instagram, WhatsApp and Messenger simply vanished from the internet. That's about 3.5 billion users, plus millions of small businesses that run their sales on WhatsApp.

A command run during routine maintenance accidentally cut Facebook's data centers off from its backbone. An audit tool built to block exactly that kind of command had a bug and let it through. Facebook's DNS servers then took themselves off the internet, and the internal tools engineers needed for the fix went down too. Reports said staff struggled even to badge into buildings. In the end, people had to fix it physically inside the data centers.

3. AWS, 2025: a quiet database in Virginia

Number 3: AWS logo

WHEN
Oct. 20, 2025

DOWN FOR
About 15 hours

WHAT BROKE
A DNS failure for DynamoDB

In the early hours of Oct. 20, 2025, Snapchat, Fortnite, Roblox, Venmo, Coinbase, Signal, Alexa and Ring doorbells all started failing at once. So did Delta Air Lines, the UK's tax authority and several British banks. Downdetector logged more than 6.5 million reports worldwide.

The root cause sat in us-east-1. A race condition in DynamoDB's automated DNS management left its regional endpoint with an empty DNS record, so nothing could find the database. DynamoDB sits under thousands of services, including plenty of AWS's own, and because some global AWS features lean on us-east-1, companies hosted elsewhere got hit too. Full recovery took about 15 hours.

2. Microsoft 365, 2026: when work stops

Number 2: Microsoft logo

WHEN
Nov. 12, 2026

DOWN FOR
Still recovering days later

WHAT BROKE
Not confirmed yet

This week's outage ranks this high because of what it hit, not how long it lasted. Email is still the nervous system of work and government, and for most of a day it went quiet across 74 countries: police, courts, hospitals, parliaments.

Two things set it apart. The fix kept failing: servers came back, buckled under load and went down again, and on day two Microsoft was still patching region by region. And Copilot went down with everything else. As more companies wire AI assistants into everyday work, an email outage is now an AI outage too. We'll update this entry when Microsoft publishes its root-cause report.

1. CrowdStrike, 2024: the biggest IT failure in history

Number 1: CrowdStrike logo

WHEN
July 19, 2024

DOWN FOR
Days, for many machines

WHAT BROKE
A faulty security update

On July 19, 2024, CrowdStrike pushed a routine content update to its Falcon sensor, and about 8.5 million Windows machines blue-screened into a boot loop. Microsoft said that was under 1% of all Windows devices. Didn't matter. They were the machines running airlines, hospitals, banks and broadcasters.

More than 10,000 flights were canceled or delayed. Hospitals postponed surgeries, some 911 centers had problems and Sky News went off air. Insurer Parametrix estimated US Fortune 500 companies lost $5.4 billion, with only 10% to 20% covered by insurance. Delta alone canceled about 7,000 flights over five days and put its bill at about $550 million. Worst of all, many machines couldn't be fixed remotely. IT teams had to boot each one into safe mode and delete the bad file by hand, often after hunting down BitLocker recovery keys first.

Why this keeps happening

Look down the list and two patterns jump out. First, none of these was a hack. At least seven of the ten started with someone's own change: an update, a config file, a maintenance command, a typo. The internet rarely gets broken from outside. It gets broken by the people running it, doing routine work.

Second, concentration. AWS shows up three times. A handful of providers sit under almost everything, so when one of them sneezes, thousands of companies that never chose to depend on each other catch a cold together.

And 2026 has been busy. Control Plane's cloud outage tracker counts 18 major AWS, Azure, Google Cloud and Cloudflare outages this year through late October, already more than all of 2025, with power and cooling failures the most common cause. Microsoft alone had a nine-hour Microsoft 365 outage in January and a five-hour Azure outage in July that spilled into Microsoft 365, before this one.

What to actually do about it

You can't stop AWS or Microsoft from having a bad day. You can decide how much it costs you.

  • Map your real dependencies. Which provider, which region and which identity system sits behind every critical path. Most teams only learn this mid-outage.

  • Set up a backup channel now. If email, chat and docs all live in one tenant, your incident team can't even talk when it's down. Pick a fallback and make sure people know it exists.

  • Spread only what really matters. Multi-region or multi-cloud for everything is expensive. For the two or three things that truly can't go down, it's cheap insurance.

  • Stage everything, config included. Canary first, then widen. Nearly every outage on this list spread because a change hit 100% at once.

  • Keep your status page off your own stack. Otherwise it dies exactly when customers go looking for it.

  • Claim your SLA credits. Microsoft and others offer them, but you usually have to file for them.

Microsoft will fix this, publish a root-cause report and promise new safeguards. Then it'll be someone else's turn. The real question isn't whether the next one comes. It's whether your team can still talk, ship and serve customers when it does.

Get stories like this in your inbox every week. Subscribe to The Week in Perspective.