01 — Before the wire
Separating the message from the messenger
1792 — 1900
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 a French engineer 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.
02 — The packet
A network with no centre
1961 — 1969
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.
Worth noting
The persistent myth is that ARPANET was built to survive nuclear war. Baran's research was;
ARPANET itself was funded for a duller and more human reason — expensive computers were scarce,
and researchers wanted to share them remotely.
03 — The protocol
The rules that made networks a network
1971 — 1986
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.
04 — The web
The door opens
1989 — 1999
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.
05 — The pocket
The internet stops being a place you go
2004 — 2016
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.
The shape of the shift
For its first thirty years the network was designed for machines that stayed still. Every
assumption — fixed addresses, stable routes, reliable power — was built around that. The mobile
era inverted it, and much of the last decade of network engineering has been spent catching up.
06 — Now
Six billion, and the two billion missing
2017 — 2026
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.
07 — What's next
The layer that hasn't been built yet
2026 —
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.