How the internet works: the journey of data from cable to page

One second, and the page is there

You type an address, press Enter, and in under a second you are looking at a page that physically sits on the other side of the world. At least four distinct things happen in that second, and knowing them in order is the fastest way to understand everything else: your device finds out which number that name corresponds to, it opens a conversation with the machine at that number, the data is broken into packets that cross dozens of different networks, and the browser reassembles the pieces.

Let’s take them one at a time.

A network of networks, not a single network

The internet is not one thing: it is the infrastructure that lets different networks — the one in your house, the one at your school, the one belonging to a Japanese carrier — talk to each other. The word says so: inter net, between networks.

The piece that makes the connection possible is the router. As the MDN documentation explains, it is a specialised computer with a single job: look at the address written on a data packet and pass it to the next network, the one that brings it closer to its destination. No router knows the whole route; each one knows only the next step. Your carrier runs a lot of them, and connects in turn to other carriers: that is how data from your house reaches a server on the far side of the planet.

Practical example: the modem-router in your home does exactly this. On one side it holds together your small network (phone, TV, computer, the robot vacuum cleaner you drive from an app); on the other it connects that network to your carrier’s. Anything leaving the house goes through it.

And where does it physically go? Not through the air. Traffic between continents runs inside optical fibres resting on the seabed: in early 2026 TeleGeography counts over 1.5 million kilometres of them in service, with more than 600 cables active or planned. Wi-Fi is only the last stretch, the one between your sofa and the router.

Addresses: who you are, in numbers

For a packet to reach somewhere, that somewhere needs an address. This is the IP address, and its definition sits in a technical document from September 1981, RFC 791, which describes an address 32 bits long: the four dot-separated numbers you have probably seen, something like 192.0.2.172.

Thirty-two bits looked enormous in 1981. With billions of connected devices they no longer suffice, hence IPv6, defined in RFC 8200 in 2017: 128-bit addresses, designed, the specification says, to support a far greater number of addressable nodes and more levels of addressing hierarchy. The two versions have coexisted for years.

One detail in RFC 791 surprises first-time readers: IP openly states that it does not provide a reliable communication service. No delivery acknowledgements, no error checking on the data itself. It delivers and hopes. This is not an oversight: IP does one job, routing, and leaves the rest to the machines at either end.

DNS: from names to numbers

Nobody remembers numeric addresses, and indeed we do not use them. DNS (Domain Name System) is the phone book that translates a name such as recaplica.com into the matching IP address. When you type an address, the first thing your device does is ask that question.

The book is organised in levels, and at the top sit the root authorities: the DNS is configured with 13 of them, identified by the letters A to M. But they are not thirteen computers locked in a room: each letter stands for a cluster of servers scattered across dozens of countries, all answering at the same address. The organisations running them, as the IANA listing shows, have little in common with each other: NASA, the US Army, the University of Maryland, Europe’s RIPE NCC, ICANN, private companies such as Verisign.

Packets: a message that travels in pieces

Here comes the idea that changed telecommunications. In traditional telephony, talking to someone meant opening a dedicated circuit: a line reserved for the two of you for the whole call. The internet does the opposite. The message is split into packets, each self-contained, each carrying the destination address. The packets leave, cross different routers, may even follow different paths, and arrive when they arrive.

It is a sturdy method: if a line goes down, later packets take another way without anyone having to place the call again. Which way is decided by routers applying a routing algorithm, a precise procedure that picks the next hop at every step.

That leaves the earlier problem: if IP guarantees nothing, how does a photo arrive intact? TCP handles it. Sitting above IP, it numbers packets on departure, checks on arrival what is missing, requests the lost pieces and puts everything back in order before handing the result to the application. The pair is called TCP/IP and has an official birthday: RFC 801, from November 1981, set the goal of moving the whole ARPANET onto these protocols by 1 January 1983, and the deadline held. Earlier still, on 29 October 1969, two ARPANET computers exchanged the first message ever sent over the network. It was meant to be the word login: the “l” and the “o” got through, then the system crashed. The network’s first message was, in effect, lo.

Practical example: sending a photo by message is like posting a book one page at a time in separate envelopes. IP is the postal service that delivers the envelopes; TCP is the person at the other end checking the numbering, asking again for the pages that never showed up and putting the book back together in the right order.

The internet is not the web

This is the most widespread confusion of all, and four lines clear it up.

	Internet	Web
What it is	The infrastructure connecting networks	A service running on top of it
Made of	Cables, routers, modems, IP/TCP protocols	Pages, links, browsers, the HTTP protocol
Born	First message in 1969	Early 1990s
Also used for	Mail, messaging, video calls, updates, browser or not	Pages and applications you open in a browser

When an app uses a generative AI model or a machine learning service, the computation happens on remote servers reached over this same infrastructure: packets, routers, IP addresses. The service on top changes, the transport underneath does not.

Why it holds up

Put the pieces together and the founding choice becomes clear: keep the centre simple and dumb, and make the edges smart. Routers do not know what they are carrying and promise nothing; the machines at the ends check what arrived and ask again for what is missing. It sounds like a surrender. It is actually the reason why the same infrastructure that in 1969 connected two ARPANET computers now carries video calls, maps and streaming without ever having been rebuilt from scratch.