Blog / How the internet works /
The internet is more fragile than people think
We talk about the internet as if it’s a cloud, something weightless and everywhere. It isn’t. It’s cables, buildings, power supplies, a few critical systems and some very ordinary human decisions. And much of it depends on fewer things than most people imagine.
My argument is that the internet is robust in design and fragile in practice. It can route around damage in theory, but a surprising amount of the web hangs on a small number of cables, companies and configuration files. If you run a website or a business online, that should change how you plan.
The design is tough
Fairness first. The internet was built to keep working when parts fail. Data is split into packets and routed dynamically, so if one path goes down, traffic can find another. There’s no single central switch to break.
That’s real, and it’s why the internet as a whole doesn’t go down. What fails are pieces of it, and the pieces matter a great deal when they are yours.
Where it breaks
1. Undersea cables
The vast majority of international internet traffic travels through submarine cables laid across the sea floor. They are about as thick as a garden hose, and they can be cut by ship anchors, fishing, earthquakes and landslides.
In March 2024, multiple cables off the west coast of Africa were damaged at around the same time, and internet services slowed or dropped in several countries, including Nigeria. It took a while for repairs, since specialised ships are needed. Traffic was rerouted where possible, but capacity was limited.
For anyone in a region served by few cables, this isn’t theory. It’s a real exposure.
2. A handful of big companies
Large parts of the web run on a small number of cloud providers, content delivery networks and DNS services. That’s efficient and cheap, and it concentrates risk.
- In October 2016, an attack on a major DNS provider, Dyn, made many popular sites unreachable for people in large parts of the world, even though the sites themselves were working.
- In June 2021, an error at the CDN Fastly took down a long list of well-known websites for about an hour.
- On 4 October 2021, the Facebook outage removed Facebook, Instagram and WhatsApp for around six hours because of a configuration change.
None of these were wars or disasters. They were ordinary technical failures in widely shared systems.
3. DNS
If DNS fails, nothing can be found, however healthy the servers are. It’s a single dependency under almost everything, and many sites rely on one DNS provider.
4. Routing mistakes
The internet’s routing system largely runs on trust between networks. A mistake or a malicious announcement can send traffic the wrong way, and it has happened more than once.
5. Power
Data centres, cell towers and routers need electricity. In places with unreliable grids, backup generators and batteries are part of the infrastructure, and when they fail, so does connectivity.
6. Software updates and configuration
A bad update pushed to many machines can do more damage than a physical attack. The common thread in many major outages is a change that was meant to be routine.
7. People
Cables get cut by accident, passwords get lost, and employees make mistakes. Human error is the biggest cause of outages most years.
Why it matters to you
As a user
You may lose payments, messages, work or access to services at the moment you need them. Phone-based payment, messaging and banking all rely on this layer working.
As a business owner
Your site depends on a chain: domain registrar, DNS provider, hosting, CDN, payment provider, email service and the visitor’s own connection. Any one link can fail. Most small businesses have never listed theirs.
As someone running a site
Uptime promises are only as good as the weakest link. A hosting provider that guarantees 99.9% can’t protect you from your DNS provider failing.
The best counter-argument
The strongest response is that outages are rare and usually short. The internet has become more reliable overall: systems are monitored, companies publish post-mortems, and engineers learn from each failure. Most of us use online services all day with few problems.
That’s right, and it’s why fragility gets forgotten. Reliability works until the one day it doesn’t. The question is whether you’ve planned for that day, not whether it will be frequent.
What to do about it
You can’t fix the undersea cables, but you can reduce your own exposure.
- Map your dependencies. Write down every service your site needs: domain, DNS, hosting, CDN, email, payments, analytics. Note who the provider is for each.
- Avoid a single point of failure where it’s cheap to do so. Use a separate DNS provider from your host if you can, and keep domain access in a separate account.
- Keep backups off the same platform as the live site.
- Own a second channel to your customers, such as an email list, so a social outage doesn’t silence you.
- Monitor from outside. An uptime service tells you something is wrong before customers do.
- Write an outage plan. Who decides, who tells customers and where do you post updates? I wrote about what I tell clients when hosting goes down and about preparing a client for a planned outage.
- Make your pages light. If connections are slow or unstable, a small page still loads.
- Learn from incidents, yours and others’. A short post-mortem after an outage is cheap insurance.
Also think about latency and distance. If your visitors are far from your server, a CDN reduces the number of long journeys, as I covered in hosting a WordPress site in Africa.
A calmer way to see it
Fragility isn’t a reason for panic. It’s a reason for humility. Anything built from many parts will fail somewhere, sometimes. The healthy attitude is the one engineers have: assume things will break, and decide ahead of time what happens next.
Treat your website the same way. It’s not a question of whether a dependency will fail. It’s whether you’ll know, and whether you’ll have a plan.
Key takeaways
- The internet routes around damage in design, but depends on cables, big providers, DNS and configuration in practice.
- Major outages are usually ordinary failures, not attacks.
- Every site has a chain of dependencies; list yours.
- Reduce single points of failure, monitor externally and own a second channel.
Related reading
- How undersea cable cuts slow down countries
- Undersea cables: the hidden wires that carry the internet
Which single service, if it vanished tomorrow, would hurt your site most?