The 5 Biggest Internet Outages Ever: When Facebook Vanished, 8.5 Million PCs Died and AWS Broke the Web

August 14, 2026

Every few years, a single failure reaches far enough that the internet stops feeling like infrastructure and starts feeling fragile. Apps go quiet, payment terminals stall, and support lines fill with people asking the same question at once. The incidents below are the ones that reached furthest.

There is no official league table of outages, and any ranking involves judgement. Ours weighs three things together: how many people or devices were affected, how long the disruption lasted, and how much real-world harm followed. That last factor matters most. An outage that grounds aircraft or silences emergency calls is not the same as one that interrupts entertainment, even when the user numbers look similar.

1. Meta, October 2021: the day Facebook removed itself from the internet

On 4 October 2021, Facebook, Instagram, WhatsApp and Messenger disappeared together for roughly six hours. The cause was not an attack. During routine maintenance, a configuration change withdrew the Border Gateway Protocol routes that tell the rest of the internet where Meta’s network lives. As Cloudflare described it, Meta’s DNS servers stopped announcing themselves and the company effectively erased its own directions from the global routing table.

The recovery was slowed by the same dependency. Internal tooling and building access systems relied on the network that had vanished, so engineers struggled to reach the hardware they needed to fix. Network analysis firm Kentik calculated the cost in human terms at around 1.2 trillion person-minutes of lost access, a figure that reflects the sheer size of Meta’s user base rather than any single dramatic consequence.

What makes this the largest outage on our measure is reach. No other single incident has removed four services used by billions of people at the same moment.

2. CrowdStrike, July 2024: 8.5 million machines, and 46,000 grounded flights

On 19 July 2024, a faulty update to CrowdStrike’s Falcon security software sent Windows machines worldwide into boot loops. Microsoft estimated that about 8.5 million devices were affected, a figure reported by Reuters and drawn from Microsoft’s own assessment.

The consequences were unusually physical. Airlines, hospitals, broadcasters, banks and retailers found their staff staring at unbootable terminals. Writing for BBC Future, reporting cited flight-tracking data from FlightAware showing around 46,000 flights delayed or cancelled in the days that followed.

Recovery was slow because many machines needed hands-on attention, in some cases booting into safe mode to delete a single file. An outage that requires a person to touch every affected device does not scale down quickly, whatever the size of the response team.

3. Amazon Web Services, October 2025: when one region carries the web

On 20 October 2025, a failure in Amazon Web Services’ US-EAST-1 region in Northern Virginia disrupted services for more than fifteen hours. Network monitoring firm ThousandEyes traced the trigger to DNS resolution failures affecting the DynamoDB database service, which then cascaded through systems that depended on it.

Because US-EAST-1 is the oldest and most heavily used AWS region, the blast radius extended well beyond Amazon. Reporting by TIME described disruption across consumer apps, games, financial services and workplace tools, including services with no obvious connection to one another beyond a shared cloud region.

This incident illustrates a structural point rather than a novel failure. Concentration creates correlation: when a large share of the web depends on the same region, a regional problem behaves like a global one.

4. Rogers, July 2022: a national network, and emergency calls

On 8 July 2022, Canadian carrier Rogers Communications suffered a nationwide failure of its wireless and wireline networks lasting most of a day. In its subsequent review, the Canadian Radio-television and Telecommunications Commission recorded that roughly 12 million subscribers lost service after a maintenance change to the core network.

The severity came from what stopped working alongside the phones. Interac payment services failed, leaving retailers unable to take card payments, and some customers could not reach 911. Coverage by CBC News documented both effects, and the emergency-calling failure prompted regulatory scrutiny and formal commitments on network resilience.

Ranked on population, this is a smaller event than the three above it. Ranked on harm per affected person, it is arguably the most serious on the list.

5. Dyn, October 2016: the attack that took the front page offline

On 21 October 2016, DNS provider Dyn was hit by a distributed denial-of-service attack that peaked at around 1.2 terabits per second, generated by the Mirai botnet running on compromised consumer devices such as cameras and routers.

Because Dyn resolved domain names for a long list of well-known sites, the attack made much of the web appear to fail at once. As The Guardian reported at the time, services including Twitter, Spotify, Reddit and Netflix became unreachable for users in North America and Europe. Security journalist Brian Krebs, whose own site had been targeted by the same botnet weeks earlier, traced the attack traffic to compromised cameras and digital video recorders in his account at KrebsOnSecurity.

It earns its place for a different reason from the others. It was the moment the industry learned that insecure consumer hardware could be assembled into infrastructure-scale force.

What these five have in common

The pattern is uncomfortable. Four of the five were self-inflicted: a routing change, a software update, a DNS dependency, a maintenance error. Only the Dyn attack was hostile. The largest disruptions in internet history have mostly come from routine work going wrong inside trusted systems, not from adversaries breaking in.

The second pattern is dependency. In each case the damage spread through something shared, whether a routing table, a security agent, a cloud region, a carrier core, or a DNS provider. Resilience failed not because a component broke, but because too much depended on that component at once.

Honourable mentions belong to incidents that fell just outside: the June 2021 Fastly configuration bug that briefly took down a large share of major news and government sites, and the February 2017 Amazon S3 disruption that broke storage-dependent services across the web. Both were severe. Neither matched the reach or duration of the five above.

Knowing sooner does not prevent an outage. It does change your response.

None of these failures could have been prevented by monitoring. What monitoring changes is the gap between something breaking and you knowing about it, which is usually where avoidable cost accumulates: staff troubleshooting the wrong thing, customers left uninformed, decisions delayed while the picture is unclear.

Status Is Down tracks the availability of thousands of services and websites, so when a shared dependency fails you can see whether the problem is yours or theirs. You can check the current picture on our real-time outage view, browse reported service disruptions or the website directory, and look up the categories these five incidents belonged to: social media and networking, web and cloud hosting, telecommunications and internet and online payment and wallets.

Our outage reporting covers incidents as they happen. If you would rather be told than have to check, our monitoring plans add alerting for the services you depend on, and you can create an account to start tracking them. We cannot stop the next large outage. We can make sure you are not the last to hear about it.

Find a service or website