ARPA awarded the contract to build the network to Bolt Beranek and Newman. The team there, led by Frank Heart and Bob Kahn, developed the routing, flow control, and network control software, and built the Interface Message Processors, the dedicated computers that would handle packet routing at each node. The host-to-host protocol, the rules that let the computers themselves communicate once connected, was specified largely by graduate students at UCLA, led by Steve Crocker, who also initiated the Request for Comments process in April 1969. RFC 1, entitled "Host Software", was the first in a series that is still used today to propose and refine Internet standards.

The four-node ARPANET was complete by December 1969, connecting UCLA, Stanford Research Institute, the University of California at Santa Barbara, and the University of Utah. It was a government network, funded for research, restricted to universities and military sites. Commercial use was strictly forbidden. But the ideas it embodied were spreading fast, and the engineers who built it were already thinking about something larger.

Through the early 1970s, other packet-switched networks began to appear, each exploring slightly different approaches to the same problem. In France, Louis Pouzin directed a research network called CYCLADES, which he began building in 1972. Pouzin's contribution was philosophical as much as technical: he insisted that the network itself should do as little as possible, leaving reliability entirely to the computers at either end. This was the end-to-end principle taken to its logical conclusion. His term for a linked collection of networks was a "catenet", from concatenated network, and the ideas he implemented in CYCLADES would directly shape the protocols that came to govern the Internet.

In Britain, Peter Kirstein's research group at University College London connected to the ARPANET in 1973, becoming the first international link in what would eventually become a global network. That same year, Bob Metcalfe at Xerox's research centre developed the theory behind Ethernet, the local-area networking technology that would eventually carry most of the world's office traffic. The building blocks were accumulating from multiple directions at once.

The problem was that all these networks could not talk to each other. Each used different protocols, different packet sizes, different conventions. A message from the ARPANET could not reach a machine on the CYCLADES network or a British academic network without a human intermediary. What was needed was a common language for networks, a protocol not of computers but of networks themselves.

Bob Kahn, now working at DARPA, recruited Vint Cerf, then at Stanford University, to work on the problem. By 1973 they had worked out the core idea: a common internetworking protocol that would hide the differences between individual networks behind a shared set of rules. The network would be responsible only for moving packets; the machines at either end would handle everything else. This was the insight borrowed from Pouzin and from Davies, now formalised into something that could run on any hardware, over any connection.

Cerf and Kahn published their ideas in May 1974. The specification of the resulting protocol, the Transmission Control Program, appeared as a formal document in December 1974. That document contains the first recorded use of the word "internet", used as a shorthand for "internetwork". Testing began in 1975 at Stanford, at Bolt Beranek and Newman, and at University College London simultaneously. On November 22, 1977, the team demonstrated a three-network connection: a packet radio van driving through San Francisco, the ARPANET, and a satellite link reaching across the Atlantic to London. Data moved seamlessly between all three. The proof of concept was complete.

The protocol was then split into two distinct layers. The Transmission Control Protocol, TCP, handled reliable delivery and error checking. The Internet Protocol, IP, handled routing and addressing. Together they became TCP/IP. The United States Department of Defense adopted TCP/IP as the standard for all its military computer networking in January 1983, and on that date the ARPANET formally switched over. The old network had become the seed of something that could, in principle, connect every computer on Earth.

For most of the 1970s and early 1980s, the networks built on these protocols were still closed systems, funded by government and restricted to researchers. The National Science Foundation stepped in where ARPA was pulling back, funding a network called NSFNET in 1986 to connect its supercomputing centres across the United States. NSFNET quickly became the backbone of American academic networking, and it grew fast: upgraded to carry more traffic in 1988, expanded to dedicated fibre in 1991, and connecting universities and research institutions across the country.

International connections followed quickly. France's national research network connected to NSFNET in 1988. The Netherlands followed. The Nordic universities gained open access. Japan, which had built its own academic network by 1984, connected to NSFNET in 1989. South Korea had established a domestic TCP/IP network as early as 1982 and formally joined NSFNET in 1990. Australia formed AARNet in 1989, a dedicated IP network for its universities. New Zealand established its first international IP connection to the United States in the same year. Within a few years, the network had ceased to be American. It had become genuinely global.

While governments and universities were building the backbone, a parallel commercial world was developing alongside it. In 1979, CompuServe became the first service to offer commercial electronic mail to personal computer users. By 1980 it offered real-time chat. Bulletin board systems, linked by a network called FidoNet, were popular among hobbyists. Usenet, born in 1979 from a dial-up connection between Duke University and the University of North Carolina, gave ordinary computer users a way to post messages to global discussion groups. These systems were not yet the Internet, but they were training millions of people in the habits the Internet would soon inherit.

The first commercial Internet service providers appeared in 1989, in the United States and Australia. In the United States, a company called The World became the first commercial dial-up provider that year. MCI Mail became the first commercial email service to connect experimentally to the Internet. The walls between the government-funded network and the commercial world were beginning to crack.

The moment that changed everything for ordinary people happened not in the United States but in Switzerland. Tim Berners-Lee, a British computer scientist working at the particle physics laboratory CERN near Geneva, had been frustrated for years by the difficulty of sharing information between researchers who used different computers and different systems. In 1989 he proposed a solution: a system of documents linked by clickable references, stored on computers connected to the Internet, accessible from anywhere on the network. He called it the World Wide Web.

Berners-Lee's Web was built on three foundations: a language for writing documents, a protocol for requesting and delivering them, and an addressing system that gave every document a unique location. The Web opened to the public in 1991. By 1993 and 1994, websites for everyday use were appearing, and the number of users began to grow not in hundreds but in millions. What had been a network for scientists became, almost overnight, a medium for everyone.

The physical infrastructure that made this possible was undergoing its own transformation. Through the mid-1990s, engineers were deploying fibre-optic cables capable of carrying vastly more data than copper telephone lines, and developing a technique called wave division multiplexing that allowed a single fibre strand to carry many separate streams of data simultaneously. Bell Labs deployed an early version of this system in 1995. The following year, a company called Ciena, co-founded by the laser pioneer Gordon Gould, deployed the world's first dense version of the system on Sprint's fibre network. The capacity of the Internet's backbone expanded enormously in a very short time.

The final barrier between the network and full commercial life fell on April 30, 1995, when the National Science Foundation ended its sponsorship of the NSFNET backbone, removing the last formal restrictions on commercial traffic. Traffic moved onto networks run by Sprint, MCI, and AT&T. The Internet was now, in every meaningful sense, open for business.

The speed of the Internet's expansion into daily life is almost impossible to overstate. In 1993, Internet traffic accounted for about one per cent of all information flowing through two-way telecommunications networks worldwide. By 2000 it was more than half. By 2007 it carried more than ninety-seven per cent. A technology that had been invisible to most of the world's population in the early 1990s had, within fifteen years, become the dominant medium of human communication.

The mid-to-late 1990s brought the first wave of commercial Internet culture. Amazon opened in 1995. eBay in the same year. MapQuest gave people maps they could look at on a screen. Email moved from universities to offices to homes. Discussion forums, personal websites, and the first blogs gave individuals a way to publish to the world for the first time. The Web of this period, later described as "Web 1.0", was largely static: pages were written by publishers and read by audiences. The conversation was mostly one-way.

Between 1997 and 2001, investors poured money into any company with a website and a business plan, however thin. Valuations soared to levels disconnected from any plausible revenue. Then, in 2000, the bubble burst. Hundreds of companies collapsed. Billions of dollars in market value disappeared. But the underlying infrastructure, the cables, the protocols, the habits of use, survived. The crash was financial, not technical, and the Internet continued to grow through it.

Governance of the expanding network became increasingly complex. The basic administrative functions, assigning addresses and managing domain names, had been handled informally by a handful of individuals in ARPANET's early years, most notably Jon Postel at the University of Southern California, who managed the Internet Assigned Numbers Authority until his death in 1998. As the network became global and commercial, more formal institutions took shape. ICANN, a California non-profit, was established in 1998 to coordinate domain names and addresses. Regional registries for address allocation were set up on every continent. The question of who governed the Internet, and in whose interest, would remain contested for decades.

The second wave of Internet culture arrived in the mid-2000s and was qualitatively different from the first. Where Web 1.0 had been about publishing, Web 2.0, a term that gained wide currency after a conference in late 2004, was about participation. The shift was from reading to writing, from audience to community, from static pages to platforms where users generated the content themselves.

Wikipedia launched in 2001 and grew into a collaborative encyclopaedia written by millions of volunteers in hundreds of languages. YouTube, founded in 2005, let anyone upload video. Facebook, also launched in 2004, and Twitter, in 2006, built networks where the content was not articles or products but people themselves, their opinions, their photographs, their daily lives. Reddit gathered communities around shared interests. These platforms did not merely distribute information; they reorganised social life around it.

Underpinning all of this was a revolution in hardware. In the 1990s, most Internet access required sitting at a desktop computer in a fixed location. By the late 2000s, smartphones had become mainstream, and with them came the ability to access the Internet from anywhere, at any time, from a device that fitted in a pocket. The cross-over point arrived in 2008, when the number of Internet-connected mobile phones exceeded the number of connected personal computers. In much of the developing world, the mobile phone was not just the most common way to reach the Internet: it was the only way.

The mobile revolution transformed who used the Internet as much as how they used it. Countries that had never built significant fixed-line telephone infrastructure leapfrogged directly to mobile connectivity. In India, South Africa, Kenya, the Philippines, and Pakistan, the majority of domestic Internet users were accessing the network from a mobile phone rather than a computer. The Internet was no longer a technology for wealthy countries with advanced infrastructure. It was becoming, unevenly and incompletely, a global utility.

The Internet's growth was not without its tensions. The question of net neutrality, the principle that Internet service providers should treat all traffic equally regardless of its source or content, became a major political battleground in the United States through the 2010s. In February 2015, the Federal Communications Commission ruled in favour of net neutrality, classifying broadband as a telecommunications service. The FCC chairman at the time said: "This is no more a plan to regulate the Internet than the First Amendment is a plan to regulate free speech." In December 2017, the FCC reversed that decision by a narrow vote, reigniting the debate.

Censorship and surveillance became equally contested ground. China built a nationwide content filter that blocked access to large portions of the global Internet, creating what critics called a domestic digital divide. Governments around the world discovered that the same network that enabled free expression also enabled surveillance, recruitment by extremist groups, and the industrial-scale spreading of disinformation. The Internet that had been imagined by Licklider as a tool for human collaboration was also, it turned out, a tool for human manipulation.

Tim Berners-Lee, who had given the world the Web, grew increasingly alarmed by what it was becoming. In November 2009 he launched the World Wide Web Foundation to campaign for a web that was safe and accessible to all. A decade later, in November 2019, he and the foundation launched what they called the Contract for the Web, calling on governments, companies, and citizens to commit to principles protecting the web from misuse. "If we don't act now," the contract warned, "we are at risk of squandering" its potential for good.

File sharing had already tested the Internet's relationship with existing industries. Napster, launched in 1999, let users share music files directly with each other, bypassing record labels entirely. Within two years it had tens of millions of users. The music industry sued, and Napster was shut down in 2001. But the successors came quickly: Kazaa, LimeWire, BitTorrent, and others. The Pirate Bay, founded in Sweden in 2003, survived legal action that resulted in prison sentences and heavy fines for its founders. The battle between free distribution and intellectual property rights reshaped the music, film, and publishing industries, forcing each to find new models for a world where copying had become essentially free.

The network eventually extended beyond the planet. On January 22, 2010, an astronaut aboard the International Space Station posted an update to his Twitter account directly and without assistance from the ground, the first unassisted Internet communication from low Earth orbit. NASA had been working since the late 1990s with Google on a protocol designed for deep space communication, one that could cope with the long delays and intermittent connections involved in reaching spacecraft behind the Moon or distant planets. The first field test of what NASA called the "deep space internet" was conducted in November 2008. The Internet, which had begun as a way to connect three terminals in one building, was reaching for the solar system.

Back on Earth, the network continued its expansion into every corner of life. Voice calls moved from telephone networks to the Internet through a technology called Voice over IP, which converts speech into digital packets and sends them alongside everything else. Telephone systems began adopting it commercially from 2003 onward. Text messaging, which had begun on December 3, 1992, when the first message, "Merry Christmas", was sent over a commercial mobile network, became one of the most widely used communication tools in human history: more than two trillion messages were logged in 2020 alone. The first mobile phone with Internet connectivity appeared in Finland in 1996. By 2020, smartphones accounted for the vast majority of all mobile data traffic worldwide.

The address system that underpinned the entire network faced its own crisis. The original Internet Protocol used thirty-two-bit addresses, which allowed for roughly four billion unique locations on the network. As the Internet expanded to connect not just computers but phones, cameras, thermostats, cars, and industrial sensors, those four billion addresses ran out. The last block was assigned in January 2011. The successor protocol, using a vastly larger address space, was already being deployed, but the transition was expected to take many years. The Internet was, even as it reached global maturity, still mid-construction.

By 2021, about four and a half billion people, roughly half the global population, were active Internet users. The fibre-optic cables that carried their traffic stretched across five billion kilometres of installed lines, crossing oceans on the seabed and threading through cities underground. The speeds at which data moved had grown from the fifty-kilobit-per-second connections of the early ARPANET to backbone links running at hundreds of gigabits per second. The network that had crashed after two letters in 1969 was now carrying the weight of human civilisation.

The people who built the early Internet did not agree on exactly what they were building or who deserved credit for it. Historians noted the difficulty: so much was decided in conversations, in telephone calls, in trust between colleagues who had worked together for years. The physical records existed in people's basements and closets. The pioneers were alive and eager to shape how their contributions were remembered. The story of the Internet resisted the tidy narrative of a single inventor and a single eureka moment because it was genuinely the product of dozens of people across three continents arguing, collaborating, and occasionally failing over the course of three decades.

What they agreed on, at the level of architecture, was the principle that had been independently reached by Baran in California, Davies in London, and Pouzin in Paris: that the network should be dumb, and the intelligence should live at the edges. No central authority should control what could be sent or received. No single node should be essential to the whole. The network should route around damage, whether that damage was a nuclear strike or a failed router or a censorious government. That principle, more than any particular technology, is what the Internet is.

The governance question, who owns the Internet, who decides its rules, who speaks for its users, remained unresolved and perhaps unresolvable. ICANN ended its formal contract with the United States Department of Commerce in October 2016, passing oversight to the global Internet community. The Internet Engineering Task Force, a loosely organised body of international volunteers, continued to develop and refine the technical standards. No single country, company, or institution controlled the network. It remained, as it had been designed to be, a distributed system, answerable to no single point, surviving by the principle that no single point of failure should be able to bring the whole thing down. That was the vision Licklider wrote in his memos in 1963. It was the network Baran sketched for a military that never built it, and Davies proved in a British laboratory that most of the world ignored. It was, in the end, the only kind of network that could have become what the Internet became: not because it was the most efficient design, but because it was the most resilient one, and resilience, it turned out, was the property that let everything else follow.