Separating the message from the messenger

For all of recorded history, information moved at the speed of a horse, a ship, or a running man. The first true break came in 1792, when Claude Chappe strung a chain of semaphore towers across France. Operators read the pivoting arms of the next tower through a telescope and repeated the signal onward. A message crossed the country in minutes instead of days.

It was not electric, but it was the first system built for one purpose only: to move information faster than a person could carry it. Every network since is a refinement of that idea.

Electricity arrived in 1837, twice over. In Britain, Cooke and Wheatstone installed a working telegraph on the Great Western Railway. In the United States, Samuel Morse built a single-wire system and, more importantly, a code — a way of reducing language to two symbols, short and long. Cooke and Wheatstone got to a public installation first. Morse won, because his design needed one wire where theirs needed six.

Reduce the alphabet to two states, and any wire on earth becomes a sentence.

The logic of Morse code, 1844

The telephone followed in 1876, in one of the most bitterly disputed patent races in engineering history — Bell filing hours ahead of Elisha Gray, with Antonio Meucci's earlier working device lost to an unpaid renewal fee. By 1901, Marconi had pushed a signal across the Atlantic without any wire at all.

By the turn of the century the world had voice, distance and wireless. What it did not have was a way for machines to talk to each other.

A network with no centre

Telephone networks were built on circuits: to talk, you opened a dedicated line and held it open. It worked for voice and was catastrophically wasteful for data, which travels in bursts. Worse, a circuit-switched network has a centre — and anything with a centre can be cut.

The answer emerged independently in three places. Leonard Kleinrock at MIT worked out the queueing mathematics. Paul Baran at RAND, thinking about communications surviving a nuclear strike, described a distributed mesh with no central node. And Donald Davies at Britain's National Physical Laboratory arrived at the same structure and gave it the name that stuck: the packet.

Break a message into small labelled pieces. Let each piece find its own route. Reassemble at the other end. If a node dies, the packets simply go around it. It is the single most consequential idea in the history of networking.

• 1966ARPA funds a four-node experimental network to test the theory.
• 29 Oct 1969The first message is sent from UCLA to Stanford Research Institute. The operator types L, then O — and the system crashes before the G. The first word ever sent over the network was, accidentally, "LO".
• Dec 1969ARPANET reaches four nodes: UCLA, SRI, UC Santa Barbara, Utah.

The rules that made networks a network

By the early 1970s several packet networks existed — ARPANET, radio networks, satellite links — and none of them could speak to the others. The problem was no longer building a network. It was building an inter-network.

In 1974, Vint Cerf and Bob Kahn published the design for a protocol that assumed nothing about the networks beneath it. It would not require them to be reliable, fast, or even similar. It only required that they pass packets. That design became TCP/IP, and its indifference to the underlying medium is precisely why it is still running the world half a century later — over copper, fibre, radio, and satellite links its authors never imagined.

• 1971Ray Tomlinson sends the first network email and chooses @ to separate user from machine — a symbol nobody was using for anything else.
• 1973Robert Metcalfe designs Ethernet at Xerox PARC, solving the local half of the problem.
• 1 Jan 1983Flag Day. ARPANET switches from NCP to TCP/IP overnight. Machines that had not been updated simply stopped working. This is the internet's actual birthday.
• 1983–84Paul Mockapetris invents DNS, replacing a single hand-maintained file of every host on the network with a distributed naming system.
• 1986NSFNET builds a national backbone and connects the university system, becoming the spine the public internet would inherit.

For twenty years the network existed and almost nobody outside research institutions could use it. It had addresses, routing and names. It had no front door.

The door opens

In March 1989, a British physicist at CERN named Tim Berners-Lee circulated a proposal for managing the laboratory's tangled documentation. His manager's written response has become famous: "Vague but exciting."

What he built combined three things — a way to address a document (URL), a way to fetch it (HTTP), and a way to write it with links to other documents (HTML). The first website went live at the end of 1990. It explained what a website was.

The decisive moment was not technical. On 30 April 1993, CERN placed the web into the public domain — no licence, no royalty, no owner. Had it been patented, the internet as we know it would almost certainly not exist.

Vague but exciting.

Mike Sendall, on Berners-Lee's proposal, 1989

• 1993Mosaic, the first browser to display images alongside text, makes the web comprehensible to non-specialists.
• 1995NSFNET is decommissioned and the backbone passes to commercial carriers. Amazon and eBay launch. India's public internet opens on 15 August, via VSNL.
• 1998Google is founded, and ICANN takes over the naming system.
• 2000–01The dot-com crash erases enormous paper value — and leaves behind the fibre, the habits and the engineers.

The internet stops being a place you go

The second decade of the web changed who was writing it. Platforms replaced pages: Facebook in 2004, YouTube in 2005. In 2006, Amazon began renting out its own spare infrastructure, and the cost of starting something on the internet collapsed from millions to a credit card.

Then, in 2007, the internet moved into a pocket. The smartphone did not add a new way to access the network — it removed the idea of accessing it at all. Connectivity stopped being a destination and became an ambient condition.

The consequences were largest where fixed-line infrastructure had never arrived. Across South Asia and Africa, hundreds of millions of people came online without ever having owned a computer. In 2016, India's data market was upended almost overnight when the price of mobile data fell through the floor; the country added internet users at a pace no market had seen before.

Six billion, and the two billion missing

In 2026 the internet passed a threshold it had been approaching for a decade: more than six billion people online, roughly 73% of everyone alive. India alone accounts for around a billion of them, with penetration crossing 70%.

The last few years have layered three shifts on top of each other. 5G pushed capacity to the edge. Generative AI, arriving in public hands at the end of 2022, became the fastest-adopted consumer technology in the network's history and began reshaping what the traffic on it actually consists of. And eSIM quietly severed the last physical link between a device and a network operator — a subscription became software.

But the more instructive number is the other one. Roughly 2.1 billion people remain offline, overwhelmingly in South Asia and Sub-Saharan Africa, and the gap is no longer mainly about cables. It is about cost, devices, literacy and — persistently — gender: men remain meaningfully more likely to be online than women, a gap several times wider than the population difference.

Meanwhile the physical substrate stays stubbornly, unglamorously material. Something close to 600 submarine cable systems carry the overwhelming majority of intercontinental traffic. The cloud is, in the end, glass on a seabed.

The layer that hasn't been built yet

Every era in this history resolved a specific incompatibility. Packets reconciled bursty data with rigid circuits. TCP/IP reconciled networks that could not speak to each other. The web reconciled documents with the people who needed to read them.

The unresolved incompatibility today is distribution. Mobile networks are national by construction — licensed, regulated and settled country by country. Travel, commerce and software are not. Between the global supply of mobile data and the channels that actually reach travellers sits no wholesale layer worth the name: no clean way for a network's capacity to reach a distributor in another country at scale, in real time, with settlement that works.

That gap is the reason this history is worth reading carefully. Each previous layer looked like plumbing until it existed, and then looked inevitable.