Is the internet down?
Every so often it genuinely feels like the whole internet has stopped. It almost never has. Four specific failures produce that sensation, and each one has a different tell.
The short version
- A DNS resolver failure is the most convincing impersonation of a total internet outage.
- A single CDN or cloud region can break a large fraction of popular sites simultaneously.
- If a site loads by IP address but not by name, the connection is fine and DNS is not.
- Genuine global internet outages do not happen; what happens is a shared layer failing.
The internet has no off switch, no central server and no single point that can be broken to stop it. It is tens of thousands of independent networks that agree to carry each other's traffic. A genuine global outage is not a thing that can happen.
And yet the experience is real and familiar: Nothing loads, every app spins, and the obvious conclusion is that the whole thing has fallen over. Four specific failures produce that sensation, and they are easy to tell apart once you know what to look for.
1. Your DNS resolver has failed
This is the most convincing impersonation of a total internet outage, and it is the most common.
Every connection begins by turning a name into an address. Your device asks a resolver - Your ISP's, or a public one like 1.1.1.1 or 8.8.8.8. If that resolver stops answering, nothing resolves. Every site, every app, every game. Your connection is perfectly healthy and carrying no traffic, because nothing knows where to send it.
Because a handful of public resolvers are configured on an enormous number of home routers, an incident at one produces a sudden global flood of unrelated "everything is down" reports from people whose internet is working fine.
The ten-second test: Open https://1.1.1.1 or https://8.8.8.8 in a browser. Those are IP addresses and need no name resolution. If a page loads, your connection is fine and the problem is DNS. The DNS guide has the fixes.
2. A CDN or edge network is having trouble
A small number of content delivery networks sit in front of an enormous share of the web. Cloudflare alone proxies something like a fifth of all websites. When one of those has a bad configuration push, thousands of unrelated sites return errors within seconds of each other.
The tell is the error page itself. Cloudflare's carries a ray ID and a numbered error in the 520–527 range; other providers have their own branded error pages. If half the sites you try are showing the same third party's error page, you have identified the layer. Each of those numbers means something specific.
The June 2021 Fastly incident is the reference case: A single customer configuration change triggered a latent bug, and a large part of the web - Including Amazon, Reddit, the UK government and most of the major news sites - Returned errors for about an hour.
3. A cloud region is failing
An enormous share of the consumer internet runs on a handful of cloud regions, and one in particular: AWS us-east-1 in Northern Virginia. When it has trouble, the casualty list is bizarre precisely because the affected services have nothing to do with each other - Smart doorbells, airline check-in, project trackers, streaming apps and a bank's mobile app all failing within the same few minutes.
The tell is breadth plus randomness. A DNS problem breaks everything uniformly. A cloud region breaks a strange, specific subset: Some sites work perfectly, others are completely dead, and there is no pattern a user can see. Why us-east-1 in particular is its own story.
4. Your ISP has a problem
Less dramatic, more local. If your provider has a routing fault or a fibre cut, your connection may be technically up - The router shows a link, the Wi-Fi works - While nothing beyond your provider's network is reachable.
The tell is geography. ISP faults cluster tightly: Reports come from one city or a handful of neighbourhoods rather than from everywhere. Cable providers in particular fail at node level, which means the street next to yours can be completely unaffected. Switching a device to mobile data settles it instantly.
Working out which one in under a minute
- Open an IP address directly (
https://1.1.1.1). Loads? Your connection is fine, DNS is the problem. Does not load? Continue. - Try mobile data with Wi-Fi off. Everything works? Your home network or ISP. Same failures? The problem is further out.
- Try five unrelated sites. All dead? DNS or your ISP. Some fine, some dead? A shared provider - A CDN or a cloud region.
- Look at the error pages. A third party's branded error on multiple sites names the layer for you.
- Check what else is failing. A list of what is currently down across many services tells you immediately whether this is you or everyone.
The things that could, in principle, be worse
Three failures would be genuinely internet-wide, and all three are guarded carefully enough that they essentially do not happen.
The DNS root. Thirteen logical root server addresses, served by hundreds of anycast instances across the world, operated by twelve independent organisations. It has never gone down.
A BGP catastrophe. Routing mistakes happen constantly and are usually contained within minutes. A large accidental leak can affect a lot of traffic briefly - Rogers in Canada demonstrated in 2022 what it looks like when it happens to one large network's entire customer base - But the damage is bounded by the fact that other networks quickly stop believing the bad announcement.
Submarine cables. Cut regularly, usually by anchors. Traffic reroutes automatically, sometimes with noticeable latency. It takes several simultaneous cuts to isolate a region, which has happened to a few countries and never to the network as a whole.
The internet was designed during the Cold War to survive the removal of arbitrary pieces of itself, and that design has held up remarkably well. What it was not designed for is everybody voluntarily concentrating on four cloud providers and two CDNs - Which is why the outages that feel like the internet breaking are, every single time, a shared layer rather than the network.
Questions people ask
Can the whole internet go down at once?
Not realistically. The internet is tens of thousands of independent networks with no central point of control, and it was explicitly designed to route around damage. What feels like a total outage is always a shared layer failing - A DNS resolver, a CDN, a cloud region or your own ISP.
Why does it feel like every website is down at the same time?
Most often a DNS resolver failure, which stops every name resolving while your connection stays perfectly healthy. The other common cause is a CDN or cloud region incident, which breaks a large but arbitrary-looking subset of popular sites simultaneously.
How do I check if my internet is working when nothing loads?
Open an IP address directly in a browser, such as https://1.1.1.1. That needs no name resolution, so if it loads your connection is fine and DNS is the problem. If it does not load, try mobile data with Wi-Fi off to separate your network from everything beyond it.
What happens if a submarine cable is cut?
Traffic reroutes over other paths automatically, usually with added latency rather than an outage. Cables are cut regularly, mostly by ships' anchors. Isolating a region takes several simultaneous cuts, which has happened to individual countries but never to the network as a whole.
Read next
Why websites go down
The real causes of website outages - Configuration pushes, BGP withdrawals, expired certificates and dependency cascades.
When DNS stops working
How DNS resolution fails, how to confirm it is the cause, and how to fix it - Including which resolver to switch to and why.
The us-east-1 problem
Why AWS us-east-1 has an outsized blast radius, the major incidents it has caused, and what makes a regional failure global.
Is it down for everyone, or just you?
A practical order of tests for working out whether a website is genuinely down or the problem is on your side - And what each result rules out.