Skip to content
10 major tech outages of the 2020s, from Google to CrowdStrike to Cloudflare

10 major tech outages of the 2020s, from Google to CrowdStrike to Cloudflare

Qz •Colleen Cabili • November 13, 2026

These 10 major tech outages of the 2020s, from Google to CrowdStrike to Cloudflare, show how a single fault could disrupt flights, payments and everyday apps

One routine change inside one company can stop planes from taking off, block card payments at checkout and silence messaging apps at once. That is a common thread running through the major tech outages of the 2020s. The failures in this list did not come from one dramatic attack on the internet. Several began with something ordinary: a quota setting, a configuration update, a maintenance job or a software release.

This selection covers 10 outages, ordered by date from December 2020 to November 2025. It is not a ranking. No single source measures these incidents against one another. Some hit consumer apps used for chat and video. Others reached airports, hospitals, banks and emergency phone lines.

The causes repeat. Google $GOOGL traced its December 2020 sign-in failure to an automated quota system. Meta $META said a configuration change on its backbone routers took , Instagram and WhatsApp offline in October 2021. The company also said it was not a malicious attack. Cloudflare said its November 2025 disruption came from an internal change rather than an attack. In the U.S., the Federal Aviation Administration's preliminary finding on its January 2023 Notice to Air Missions failure pointed to a damaged database file. Amazon $AMZN Web Services' post-event summary on its October 2025 cloud outage cited a DNS problem in its Northern Virginia region.

Another pattern is concentration. Several of these incidents started at one provider and spread to the services built on top of it. A software bug at Fastly, a content delivery network, knocked out sites including $RDDT and Gov.uk in 2021. A faulty CrowdStrike $CRWD update in July 2024 crashed Windows machines at airlines, hospitals and banks. Microsoft $MSFT estimated that 8.5 million Windows devices were affected. Telecom failures followed a similar shape on a national scale. Rogers in Canada and Optus in Australia each had network failures that disrupted emergency calling.

Each slide that follows explains what failed, when it happened and what is known its cause. Where a company or agency published an incident report or public explanation, the slide names it and keeps to what that organization said.

Read together, the 10 cases span social and consumer apps, cloud platforms, content delivery networks, telecom carriers, security software and the U.S. flight-notice system.

Google sign-in outage of December 2020 knocks out Gmail and YouTube

On Dec. 14, 2020, Google $GOOGL 's sign-in system failed for roughly 45 minutes, and Gmail, YouTube and Google Drive stopped working for users who were logged in. Google 's incident report traced the failure to its automated quota system, which cut the capacity of its User ID Service.

The outage hit services that look separate to the people using them. Gmail handles email. YouTube streams video. Google Drive stores files. All three depend on Google confirming who the user is before anything else happens. When that check failed, the products built on top of it failed along with it.

That check is called authentication. It is the step in which a service confirms an account's identity before granting access to it. Google runs one shared account system across its consumer products. A single Google account opens Gmail, YouTube, Drive and the company's other services. The design spares people from managing a separate login for each product. It also means a fault in the sign-in layer reaches every product that relies on it.

The User ID Service is part of that sign-in layer. Once the automated quota system reduced its capacity, authentication requests arrived without enough resources to be processed. Requests that could not be verified went unanswered. The email, video and file services waiting on them stopped responding.

Quota systems are a standard tool in large computing operations. They divide capacity among internal services and cap how much any one of them can draw. That keeps a single service from consuming resources that others need. In this case, the automated system responsible for allocating resources was the component that took them away.

The episode shows that a safeguard can become a single point of failure. The quota system governed a service that every signed-in product needed, so its error did not stay in one place. Email, video and file storage failed together during the same short window. Products that look independent shared one hidden dependency, and that dependency decided whether any of them worked.

Fastly CDN outage of June 2021 takes down and Gov.uk

On June 8, 2021, a software bug in Fastly's content delivery network knocked major websites offline, including $RDDT , The New York Times and the U.K. government's Gov.uk. The bug was triggered when a Fastly customer changed a configuration setting.

A content delivery network, or CDN, is a distributed system of servers that stores copies of web content in many locations. When a reader requests a page, the CDN serves it from a server close to that reader instead of from the publisher's origin server. That setup makes pages load faster. It also absorbs traffic spikes that could overwhelm a single site.

The same setup creates a dependency. Publishers that route their traffic through a CDN rely on it to deliver their pages at all. When the CDN fails, readers can lose access to a site even though the site's own servers keep running.

That is the pattern this outage followed. The problem did not start inside , The New York Times or Gov.uk. It started at the shared provider that sat between those sites and their readers.

The change itself came from one customer, not from Fastly. CDN customers typically control the settings for their own content, such as which files are stored and for how long. In this case, one customer's adjustment exposed a flaw in Fastly's software. The effect reached sites far beyond that customer's own.

Fastly said 85% of its network was back to normal within 49 minutes. The company then issued a fix for the bug.

The affected sites serve very different audiences. Gov.uk is the U.K. government's central website for public services such as tax, benefits and passports. hosts user-run discussion communities. The New York Times publishes news and opinion. A government portal, a discussion platform and a news publisher all went offline because they depended on the same provider.

, Instagram and WhatsApp go dark in the October 2021 Meta outage

$META , Instagram and WhatsApp went offline together for six hours on Oct. 4, 2021. Meta traced the outage to a configuration change on its backbone routers.

A backbone is the internal network that carries traffic between a company's data centers. Routers on that network direct data from one location to another. Meta said the faulty change cut its data centers off from one another.

The effect reached beyond the company's own facilities. Meta said the same change made its network unreachable from the wider internet. People who opened any of the three apps during the outage could not connect to the servers behind them.

The three services failed at once because they depended on the same company network. One change to that shared infrastructure took a social network, a photo-sharing app and a messaging service offline together.

Meta placed the cause inside its own operations. The company said the outage was not a malicious attack. In its account, the trigger was a change it made to its own routers rather than an intrusion from outside.

Rogers network outage of July 2022 disrupts Canadian wireless service and debit payments

Rogers Communications lost wireless and internet service across Canada on July 8, 2022. The nationwide failure followed a maintenance update to the company's core network. It disrupted debit payments and some emergency calling.

Rogers is one of Canada's major telecommunications carriers. It sells mobile phone plans and internet service. Businesses also depend on its network to move data. A failure at that scale reached well beyond people trying to place a call or load a web page.

A carrier's core network is the central part of its system. It directs traffic between cell towers, connections and the wider internet. A problem at that level can take down wireless and wired service at the same time. That is what happened in July 2022, when both failed together across the country.

Debit payments were among the most visible effects. Card terminals at checkouts need a working network connection to approve a transaction. With that connection gone, debit purchases at affected businesses could not go through.

Some emergency calling was also disrupted. For some people in Canada, the outage cut off a route to emergency services as well as everyday phone and internet use.

Service did not return all at once. Rogers' wireless and internet services came back over the following day.

FAA NOTAM system failure grounds U.S. domestic departures in January 2023

The Federal Aviation Administration's Notice to Air Missions system failed on Jan. 11, 2023, and the failure led to a nationwide ground stop on departing domestic flights in the U.S. The FAA's preliminary finding pointed to a damaged database file.

The outage did not take down a website or a messaging app. It held aircraft at airports across the country. The system that failed is one of the ways pilots learn conditions that could affect a flight before departure.

A NOTAM alerts pilots and other flight personnel to conditions at an airport or along a route. Common examples include closed runways, temporary airspace restrictions and aids that are out of service. Each notice is short and written with standard abbreviations. Flight crews and dispatchers review the relevant notices while planning a trip.

NOTAMs are not only an American tool. The International Civil Aviation Organization sets the standards that member countries use to issue them.

Preflight review is a legal duty in the U.S. Federal Aviation Regulations require the pilot in command to become familiar with all available information concerning a flight before it begins. NOTAMs are one part of that information.

The FAA changed the name of these notices in December 2021. They had been known as Notices to Airmen. The agency adopted the gender-neutral Notice to Air Missions and kept the NOTAM abbreviation.

A ground stop holds aircraft at their departure airports instead of letting them take off. Flights already in the air are outside its scope. The FAA uses ground stops at individual airports for reasons such as weather or congestion.

The January 2023 order was different in scope. It applied to the whole country rather than a single airport. It was widely reported as the first nationwide halt of its kind since the September 2001 attacks.

Microsoft 365 outage of January 2023 hits Teams, Outlook and Azure

Microsoft $MSFT 365 services, including Teams and Outlook, went down on Jan. 25, 2023, along with parts of the Azure cloud platform. Microsoft said a change to its wide-area network disrupted connectivity.

That explanation places the trigger in Microsoft 's network rather than in any one app. A fault at the network level can affect every service whose traffic crosses that network. Workplace tools and cloud infrastructure went down on the same day.

A wide-area network, or WAN, links computers and data centers spread over long distances. A local network typically covers a single building or campus. For a cloud provider, the WAN connects its facilities to one another. It also connects them to the people and businesses using its services.

Microsoft 365 is the company's subscription bundle of productivity software and online services. Teams is its workplace chat and video meeting app. Outlook is its email and calendar service.

Azure is Microsoft 's cloud computing platform. Businesses use it to run their own software and store data. A disruption to Azure can reach the customers of companies that build on it, not only the people who use Microsoft 's own apps.

Optus outage in Australia cuts service for 10 million customers in November 2023

10 million Optus customers in Australia lost service on Nov. 8, 2023, in an outage that lasted roughly 12 hours. Optus attributed it to a software upgrade at one of its international peering networks.

That account places the starting point outside Optus' own network, on a network connected to it. A peering network is a separate network that links directly with a carrier's network so the two can exchange traffic. Carriers use these direct connections to pass data to and from other parts of the internet.

Optus said that after the upgrade, routing information exceeded preset safety limits on its core routers. Routing information is the data routers use to decide where traffic should go. A preset safety limit caps how much of that data a router will accept. Network operators set these limits in advance as a guard against overload.

Core routers sit at the center of a carrier's network. They carry traffic for the customer base as a whole rather than for a single neighborhood or cell tower. A problem at that layer can reach customers far from the point where it started.

Emergency calling was also affected. Failures in emergency calls were widely reported during the disruption. In Australia, emergency services are reached by dialing 000.

Those failures drew regulatory scrutiny of Optus.

CrowdStrike Falcon update crashes 8.5 million Windows devices in July 2024

On July 19, 2024, a defective content update to CrowdStrike $CRWD 's Falcon sensor crashed computers running Microsoft $MSFT Windows. Microsoft estimated that 8.5 million Windows devices were affected.

Microsoft published that estimate in a blog post on its official site. It also put the number in context. The company said the affected machines made up less than 1% of all Windows machines. That still included computers at airlines, hospitals and banks.

In the airline industry, the effects reached passengers directly. Delta Air Lines canceled flights.

CrowdStrike is a cybersecurity company, and Falcon is its endpoint security platform. An endpoint is an individual device, such as a laptop, desktop or server, that connects to a network. The Falcon sensor is the software installed on those devices. Organizations deploy tools like this to detect and block threats on each machine. Falcon also runs on macOS and Linux, but computers running those systems were not affected by this update.

Microsoft was not the source of the defect. The faulty update came from CrowdStrike , a separate company whose software runs on Windows. Security software typically receives frequent updates so it can respond to new threats. In July 2024, an update meant to help protect machines instead left them unable to run.

Unlike outages that leave a website or cloud service unreachable, this one stopped the computers themselves from running.

AWS us-east-1 DNS failure disrupts Snapchat, Venmo and Fortnite in October 2025

Amazon $AMZN Web Services suffered a major outage on Oct. 20, 2025, in its us-east-1 region in Northern Virginia. AWS's post-event summary traced the failure to a DNS problem affecting the endpoint for DynamoDB, its database service.

The disruption reached well beyond Amazon . Snapchat, Venmo and Fortnite were among the apps and services that reported problems. Each depends on infrastructure that AWS runs.

DNS, short for Domain Name System, works like the internet's address book. It translates a name that software asks for into the numerical address of the machine that answers. An endpoint is the address a program contacts to reach a cloud service. When DNS cannot return a working address for that endpoint, software that calls the service cannot find it. That can happen even if the servers behind it are still running.

DynamoDB is a managed database that AWS runs on behalf of its customers. Applications use it to store and retrieve data without operating their own database servers. A lost connection to it can stop an app from loading accounts, records or settings.

AWS's summary located the root of the problem in DynamoDB's own DNS management. It described a race condition. In software, a race condition occurs when two or more processes act on the same data at nearly the same time. The outcome then depends on which one finishes first. If the timing goes the wrong way, the system can end up in a state its designers did not intend.

A cloud region is a cluster of data centers in one geographic area. Many unrelated companies build their products on the same regional services. A fault in one widely used component can therefore surface at a messaging app, a payments app and a video game at the same time. AWS lets customers spread workloads across several regions. That design can limit exposure to a failure in any one of them.

Cloudflare bot management file crash disrupts X and ChatGPT in November 2025

A database permissions change inside Cloudflare caused the company's outage on Nov. 18, 2025. The change made a feature file used by its bot management system grow beyond its expected size. That oversized file crashed the software that routes traffic across Cloudflare's network. Sites including X $TWTR and OpenAI's ChatGPT were affected.

Cloudflare sits between many websites and the people trying to reach them. Customer sites route their incoming requests through its network, which provides services such as content delivery and protection against attacks. Visitor traffic passes through Cloudflare's systems before it reaches the site itself. A failure in that layer can make a working website unreachable.

Bot management is one of the Cloudflare products that operate inside that layer. It tries to separate automated traffic, such as scrapers and credential-stuffing bots, from requests made by people. The system depends on a feature file, a set of data the software consults when it evaluates incoming requests. On Nov. 18, that file became the point of failure.

The damage did not stay inside one Cloudflare product. The crash hit the component that moves requests through the network, not a single add-on feature. Websites that relied on Cloudflare to carry their traffic lost that path to their visitors. The fault sat inside Cloudflare's systems, not on those sites' own servers.

Cloudflare said the outage was not an attack. The company attributed it to an internal change of its own. The bot filtering at the center of the incident is itself one of the tools Cloudflare offers to screen out unwanted traffic.

Cloudflare later published a post-incident explanation. It traced the outage step by step, from the permissions change to the oversized file to the crash in the traffic-routing software. That sequence shows how a change to database permissions reached all the way to the system that carries visitor requests.