Apps stopped working and so did websites. Alexa was down and the internet seemed broken. Was it a disaster, and could it have been prevented?
Where were you on the evening of October 20, 2025? If you were finishing dinner at around 8pm, getting the kids or yourself ready for bed, or enjoying a nice bath or book (who has the time), you might not have seen it happening.
But the moment you checked your phone or tried to browse a specific app, you would have seen the internet on fire. The world may have return the classic “this is fine” gif to the fore, because the internet was burning down, and nobody knew why.
What was happening was this: one of the biggest datacentres in the world had problems, and that had dramatic repercussions for some of the biggest websites and apps.
What happened, and could it have been stopped?
The AWS outage of 2025
If the internet seemed like it was on fire, you weren’t alone. Apps went offline, website requests were empty, Amazon’s own Alexa system became a bit of an inattentive mute, and the website DownDetector became everyone’s friend, particularly for journalists compiling lists of everything going down. A website indexing the problems everyone was having, the entire world took notice of problems and chimed in almost immediately.
The result was clear: something wasn’t working.
That “something” was traced back to an entire datacentre: Amazon’s US-East-1 region.
Errors rates suddenly started going up, skyrocketing to the point where things stopped working. It seemed like systems were failing.
To understand what was going on, you need to look at what Amazon is, and more specifically, what Amazon Web Services is.
What is AWS, and why is it so important?
You may know Amazon for its online store and marketplace, and for its Prime TV service, and for the Kindle eReader and its Echo speakers and ownership of Ring home security and a bunch of other things, but it also rents out servers for hosting websites and apps, and servers for generally storing things.
Websites and internet-connected apps and services in the cloud can’t exist without servers, since “the cloud” is just another way of saying “internet-connected services”. That internet doesn’t exist without the connections between servers hosting these services, and simply put, that’s what AWS provides.
Lots of companies use AWS, with the estimated customer count over four million, covering big and small companies. There are other data-centre providers around the world — Amazon isn’t the only place you can get a server from — but it’s one of the big ones, with regions all over the world in places closer to your customers.
If many of your customers are in Australia, you might use a centre and cluster of servers located in Australia, or even somewhere else nearby, such as Singapore. The distance between those users and your data centre is shorter when the servers are nearer, versus say using a data centre in the US or the UK.
But if most of your customers are in the US, you might choose to house in a region closer to those customers.
With the companies using Amazon’s US-East systems, it means the bulk of their data and operations were at that facility, if not all of it.
Weirdly, the culprint wasn’t even the servers themselves, but rather the Domain Name Service controller at Amazon, also known as the DNS.
What’s a DNS?
All servers rely on a DNS, with the Domain Name System acting a little like a phone book or directory of services for where browsers and apps find the data they’re looking for.
Think of DNS as a website’s and app’s way of telling a device where to go.
A phone book is an apt analogy because like a phone book, the result is a number. Your computer and device is contacting that number for you, and the website and app sends back what it needs for you to get the information to finish the job.
For Amazon, a DNS failure means that websites and apps reliant on services at that datacentre can’t actually communicate properly, and the messages and signals are all lost.
The result was clear.
If you wanted to log on to Roblox, Fortnite, Reddit, or Duolingo, not to mention any number of the millions of customers using Amazon’s Web Services, you were out of luck.
Attempts by your browser and app to talk to the systems housing this data were fruitless, and it can seem like the entire internet had stopped.
Could regular people do anything?
It’s not quite the same as the Crowdstrike outage last year, but like that one, unfortunately there’s not a lot you can do in this instance, beyond get by without the services.
In some instances, that’s not easy, particularly when outages take out important and critical services.
Folks awake and working at the time of the outage saw service failures in the architecture and construction world, affecting their jobs. They would just have to take an extended coffee break, and possibly an early mark.
Gamers missing out on their entertainment and anyone calling on Alexa to play a routine, such as an alarm clock missing its alarm or setting up for bed, just had to go without.
We understand some struggled to pay, as the outage temporarily took down payment services, too.
In these situations, making do is arguably all you can do. There’s very little regular people using the web and apps can do to avoid an outage, beyond putting the phone or computer down, and just doing something else, anything else. Try not to think about it.
Could it have been prevented?
But it’s something companies could probably do better on, and the issue often comes to centralisation.
We don’t know quite how much data is used and stored at Amazon’s data storage and service systems, but we know the number is big. Really big.
Think in the millions and billions of gigabytes, pushing past the terabyte mark and hitting terms you’ve probably never heard of, like petabytes, exabytes, and zettabytes. Lots and lots and lots of storage.
Websites and services typically pay for lots of storage together at a bit of a discount compared to running the services themselves because it gets complex to do that, and can be very expensive. Amazon has data-centres and web storage specifically made for this, and it’s where Amazon makes a lot of money doing so.
AWS is a juggernaut in the server industry, and it’s usually reliable.
But like any outage, when it goes down, it goes down hard, and that’s what happened on October 20, 2025.
For companies hosting much (if not all) of their data at one of Amazon’s massive East facilities in the state of Virginia in the USA, the result was that their data and services were being served from one location.
It’s a form of centralisation that can mean faster performance for sites and apps, but can also come with the caveat of total failure and no way to fix things quickly.
If the service or site hasn’t been mirrored to another location, or the place where the app or site was being controlled couldn’t be switched like the signal on train tracks, the service went down.
It’s pretty easy to see why the internet felt like it was broken.
Moving forward
While it’s entirely possible that the 2025’s AWS outage could have been prevented, the skeptic in this journalist suggests it probably will happen again, and again, and companies won’t likely learn their lesson.
Outages this severe do happen, but they don’t happen all that often, and Amazon started restoring the systems relatively quickly after it worked out what was going on.
After about two hours, fixes were already in the process of being deployed, and the internet was gradually returning. These things take time, and the restoration of the web can take some time to trickle out, but it was on the mend.
The solution for many a company is clearly to house their data and services in more places, reducing the risk of failure when it happens. If something like this happens again, much like a train flicking a signal and switching tracks, a web provider can do much the same.
There is a way forward, and there are things companies and service providers can do to mitigate the drama. It just takes time, money, and a desire to fix. This might be like the imperfect Venn diagram, however, where companies are stuck saying you can have two.
