The History of the Internet

First version 21 October 2026, last changed 24 October 2026 · Andreas Stockter

The Internet is a federation of computer networks that exchange data under common rules, so that every connected computer can reach every other. Its parts have owners: the lines belong to telephone companies and cable operators, the computers to firms, public bodies and private people. The federation as a whole belongs to no one, has no centre and no switch that could turn it off. In 2024, by the estimate of the International Telecommunication Union (ITU), 5.5 billion people used it, more than two thirds of the world’s population.¹ The World Wide Web, often confused with the Internet, is only one of its services, alongside email, file transfer, telephony and video: the system of linked documents that a browser displays.

Like the history of AI and the history of software development, this text follows a chain: each era resolves a bottleneck the previous one left open, and creates the next one in doing so. First the line was the bottleneck, then the island, then the incompatibility of networks, then finding, then access, then taking part, and lastly trust. Every technical term is set in bold at the point where it is explained, and is only used after that. A glossary lists all terms alphabetically.

Contents: 1 Prehistory · 2 Packets and ARPANET · 3 TCP/IP · 4 World Wide Web · 5 Commerce, search, bubble · 6 Broadband, Web 2.0, smartphone · 7 Platforms, encryption, rules · The chain in brief · Timeline · Further reading · Sources

1. Prehistory: wire and computer (1840s–1950s)

Messages on the wire. The idea of sending messages as electrical signals along a wire is a hundred years older than the computer. On 24 May 1844 Samuel Morse transmitted the first public telegraph message from Washington to Baltimore over 60 kilometres. His code formed each letter from short and long pulses of current, and the network of telegraph lines spread across the continents within a few decades. In 1858 the first cable across the Atlantic joined Europe and America; it lasted three weeks. In 1866 a durable one was laid. News that had taken two weeks by ship arrived in minutes. The cable under the sea, today made of glass fibre, remains the basis of worldwide traffic to this day.

In 1876 Alexander Graham Bell was granted the patent on the telephone. The telephone network that grew from it worked on a principle that matters for this story: circuit switching. Before a call, the exchange, at first a person at a plugboard and later a relay, establishes a continuous electrical connection between the two instruments. For the duration of the call that line belongs to the two of them alone, even while they are silent. Afterwards it is released. This is ideal for speech, because the connection works without delay, and it is wasteful for anything that comes in bursts.

The measure of information. In 1948 Claude Shannon at Bell Laboratories, the research arm of the American telephone monopoly, published the mathematical theory of communication. He introduced the bit as the unit of information, showed how many bits per second a channel of given bandwidth and given noise can carry at most, and how messages can be protected against transmission errors by check bits.² Every data connection since, from the modem to the fibre-optic cable, moves within his limits, and every figure such as “50 megabits per second” is a figure in his unit.

The computer as an island. The first computers of the 1940s and 1950s stood alone in a hall, and whoever wanted to compute brought their punched cards there. In the 1950s the United States built the air-defence system SAGE, in which radar stations reported their data to central computers over telephone lines; for this Bell developed the modem in 1958, a device that turns bits into tones a telephone line can carry, and back again at the other end. Around 1960 time-sharing arrived: one computer served several people at teletypes at once by switching between them in fractions of a second. For the first time a person sat at a terminal whose computer stood in the next building or, over a telephone line, in another city.

But the terminals of a computer could only talk to that one computer. Each institute had its own system with its own programs and data, and between the systems there was nothing. J. C. R. Licklider, a psychologist who studied the cooperation of human and computer, described in “Man-Computer Symbiosis” in 1960 the vision of a “network of thinking centres” connected by lines.³ In 1962 he took over the information-technology office of ARPA, the research agency of the American Department of Defense founded after the Sputnik shock of 1957, and in 1963 he wrote his colleagues a half-joking memorandum addressed to the “Members and Affiliates of the Intergalactic Computer Network”, asking how computers of different makes should talk to each other at all.⁴ The question was the bottleneck of the age: the computers were islands, and the lines between them would, on the telephone principle, have occupied a whole line for every conversation, for data that comes in short bursts with minutes of nothing in between.

2. Packets and the ARPANET (1960s–1972)

Packet switching. The answer to the bottleneck was found by three people independently. Leonard Kleinrock at MIT examined mathematically, in his doctoral thesis begun in 1961 and completed in 1962, how messages wait and queue in a network when many participants share the lines, and showed that sharing is far more efficient than reserving.⁵ Paul Baran at the RAND Corporation sought from 1960 to 1964 a communications network that would survive a nuclear attack. His answer: no centre, but a meshed network with many routes, in which every message is cut into blocks of equal size that find their own way individually and are merely stored and passed on at each node; if a node fails, the blocks take another route.⁶ Donald Davies at the British National Physical Laboratory arrived at the same principle in 1965 without knowing Baran’s work, for a different reason: he wanted to share computers efficiently. The word packet for the block is his, and his laboratory built a local network on this principle, which went into operation in 1969.⁷

Packet switching therefore means: a message is cut into packets, each carries the address of the recipient like a letter, each travels through the network on its own, and at the destination the packets are reassembled. Lines are not reserved but used by everyone at the same time; a packet waits briefly at a node if the line is busy. This suits data as well as circuit switching suited speech, and it is the principle on which the Internet works to this day.

The ARPANET. In 1966 ARPA brought in Larry Roberts to build a network between the computers of the institutes it funded, above all to share expensive machines. At a conference in Gatlinburg in October 1967 he presented the plan and there learned of Davies’ packet idea.⁸ Wesley Clark contributed the decisive trick: not the big computers of the institutes should do the switching, but at each site a small, identically built switching computer, the IMP (Interface Message Processor). The institutes then had only to connect their computer to the IMP, not to the whole network. The firm Bolt Beranek and Newman in Cambridge near Boston won the contract in January 1969 and delivered the first IMP in September: a Honeywell minicomputer in a steel cabinet the size of a refrigerator.

On 29 October 1969 at half past ten in the evening the student Charley Kline sat at UCLA in Los Angeles and tried to log in to the computer of the Stanford Research Institute (SRI) 560 kilometres to the north. He typed the L, then the O, both arrived; at the G the system at SRI crashed. The first message of the ARPANET was therefore “LO”. An hour later the login worked.⁹ In December 1969 the network had four nodes: UCLA, SRI, the University of California at Santa Barbara and the University of Utah, connected by leased telephone lines at 50 kilobits per second. An IMP cost about 100,000 dollars.

Protocols and RFCs. For two computers of different makes to understand each other requires agreements down to the detail: how a message begins, how it is acknowledged, what happens on an error. Such an agreement is called a protocol. The students who designed the protocols at the first sites in 1969 called themselves the Network Working Group, and Steve Crocker wrote their first note on 7 April 1969. Because the group did not want to dictate anything to anyone, he called it “Request for Comments”.¹⁰ The name stuck. RFC is to this day the name of every document in which the rules of the Internet are written, and there are more than 9,000 of them. That a standard begins as a request for comments which anyone may read and answer shaped the culture of the network more than any single technical decision.

The first protocol between the computers themselves, the Network Control Program of 1970, allowed two things: logging in to a remote computer (Telnet) and transferring files (FTP, File Transfer Protocol). Both had been the purpose of the network. What then really made it big, nobody had planned. Ray Tomlinson at Bolt Beranek and Newman wrote a program at the end of 1971 that sent a message to a user’s mailbox on another computer, and needed a character to separate the user name from the computer name. He took the @, because it occurred in no name. Email was invented, and by 1973 it made up three quarters of the traffic on the ARPANET.¹¹ On 3 May 1978 a salesman at Digital Equipment sent an invitation to a product demonstration to about 400 addresses on the network and reaped indignant replies: the first advertising email, long before it was called spam.

In October 1972 Bob Kahn demonstrated the ARPANET publicly at a conference in Washington, with forty terminals from which visitors could work on computers across the country. The network then had 29 nodes, and the demonstration convinced the field that packet switching works.¹² In the same year Louis Pouzin in France started the network Cyclades, which went a step further: its packets, called datagrams, ran through the network without any guarantee; order and completeness were ensured solely by the computers at the ends.¹³ This principle, keeping the network dumb and the ends clever, the Internet adopted.

Radio and cable. Norman Abramson at the University of Hawaii connected terminals on the islands by radio to the computer in Honolulu from 1970, and because all shared the same channel he invented the rule by which everyone simply transmits and, on a collision, repeats after a random wait.¹⁴ Bob Metcalfe carried this principle over in 1973 at Xerox’s research centre in Palo Alto to a coaxial cable to which all the computers of a building were attached, and called it Ethernet.¹⁵ Ethernet has connected the computers within a building ever since; the ARPANET connected cities. Together they raised the next question: there were now several networks with different rules, and a computer in one could not reach a computer in another.

Try it yourself. Anyone with a terminal can see the basis of the network in two commands. ping stockter.info sends small packets to the server of this site and measures how long the reply takes, usually between 10 and 40 milliseconds. traceroute stockter.info (on Windows tracert) shows the nodes the packets pass through: your own router, a few switching computers of your provider, an exchange point, the computers of the data centre. Each line is an IMP of today, and that the route may look different tomorrow is Baran’s idea of 1964.

3. TCP/IP: the network of networks (1973–1989)

Internetting. Bob Kahn, by now at ARPA, had three networks in front of him in 1973 that did not fit together: the ARPANET with its lines, a packet radio network for vehicles and a satellite network to Europe. Each had its own packet sizes, its own addresses, its own error handling. With Vint Cerf of Stanford University he designed a method that leaves the networks as they are and lays a common layer over them. The paper appeared in May 1974: “A Protocol for Packet Network Intercommunication”.¹⁶ Its ideas: every packet gets a uniform envelope with a network-spanning address. Between the networks stand gateways, today called routers, which read only this envelope and pass the packet into the next network without caring about the contents. And reliability is ensured by the computers at the ends: they number the packets, acknowledge receipt and resend what is lost. A network of networks was called an “internetwork” in the jargon, for short Internet; the name has stood in an RFC since 1974.¹⁷

In 1978 Jon Postel split the method into two protocols. IP (Internet Protocol) knows only the envelope: sender, recipient, onward. It promises nothing; a packet may be lost, arrive twice or in the wrong order. TCP (Transmission Control Protocol) above it turns the unreliable packets into a reliable connection: it sets the connection up with a three-way handshake, numbers every byte, has receipt acknowledged, repeats on loss and ensures that the application sees the data complete and in the right order. Whoever can do without that, say for speech, where a lost packet is better than a late one, takes the lean UDP, which simply sends packets off. The definitive descriptions appeared in September 1981 as RFC 791 and 793.¹⁸

The division of labour between the protocols is called the layer model. Each layer performs one task and relies on the one below for the rest: the application (email, web) hands its data to TCP, TCP divides it into segments and hands them to IP, IP puts them into packets with an address, and the lowest layer, Ethernet or radio, carries them to the next node. On sending, every message is thus put into envelopes layer by layer, called encapsulation; on receipt they are opened again. The routers on the way open only as far as the IP envelope.

An IP address of the first version consists of 32 bits, written as four numbers from 0 to 255, such as 203.0.113.7. That gives 2³², about 4.3 billion possible addresses, which in 1981 seemed inexhaustible and did not remain so. The layer model and the three principles that Jerome Saltzer, David Reed and David Clark formulated in 1984 as the end-to-end principle shape the network to this day: the network carries packets and nothing else; what a service needs, the ends provide; therefore anyone can invent a new service without asking the network.¹⁹ The web, Internet telephony and video streaming arose exactly so, without a change to the network.

The cut-over day. On 1 January 1983 the ARPANET switched to TCP/IP, in one day, after a year of preparation; the old software was switched off, and whoever had not converted was out.²⁰ This date counts as the birthday of the Internet. In the same year the military part was split off as MILNET, and the Department of Defense, which had paid for the development, handed the network over to research.

One problem grew with the network: computers had names people could remember, and the mapping of names to addresses stood in a single file, HOSTS.TXT, which one institute in California maintained and every computer downloaded regularly. With a few hundred computers that worked; with thousands it no longer did. Paul Mockapetris designed the Domain Name System (DNS) in 1983: a distributed, hierarchical database in which every name is read from right to left. For the ending, say .info, particular servers are responsible, which know which servers are responsible for stockter.info, and those know the address. At the very top stand the root servers, which know only who is responsible for which ending.²¹ A computer that needs an address asks a resolver, usually at the access provider, which queries the chain and keeps the result for a while. In 1985 the first endings were set up, .com, .edu, .gov, .mil, .org and .net, and country endings followed. The ending .de was registered in November 1986 and at first administered by the University of Dortmund.

The network grows. In 1984 the Internet counted 1,000 connected computers, in 1987 10,000, in 1989 100,000.²² The driving force was no longer ARPA but science. The American National Science Foundation built the NSFNET from 1985, which connected five supercomputer centres and was open to every university; in 1988 its backbone ran at 1.5 megabits per second, in 1991 at 45. Alongside it, without any central planning, other networks had formed that passed messages from computer to computer over nightly telephone connections: from 1979 the Usenet, a worldwide system of discussion forums in which everything from operating systems to recipes was debated, and the BITNET of the universities. They grew into the Internet. On 3 August 1984 Michael Rotert at the University of Karlsruhe received the first Internet email in Germany, a greeting from the United States; the German research network DFN had been founded in the same year.²³

With growth came the first crises of sharing. In 1986 the traffic between Berkeley and a neighbouring institute collapsed several times to a thousandth: all computers resent immediately on losses and clogged the line all the more. Van Jacobson found the remedy in 1988 that is built into every TCP: a sender feels its way forward at increasing speed, drops to half at the first loss and begins to grow again. Since then millions of connections have shared a line without anyone allocating it.²⁴

On the evening of 2 November 1988 Robert Morris, a student at Cornell, released into the network a program that infected computers through holes in their software and spread onward from there. It was meant to spread unobtrusively, but through a bug it copied itself so often that the infected computers ground to a halt. Within a day about 6,000 of the roughly 60,000 computers on the network were affected, a tenth.²⁵ The Morris worm was the first major security event of the Internet and led to the founding of the first emergency response team for computer security, the CERT at Carnegie Mellon University, in the same month. The network had been built among colleagues who trusted one another; that it would not stay that way was clear from that evening on.

Who decides. The rules of the network continued to arise on the pattern of the first RFCs. In 1986 the IETF (Internet Engineering Task Force) met for the first time, an open assembly without membership in which anyone who turns up can take part. David Clark summed up its way of working in 1992 in a sentence that became its motto: “We reject: kings, presidents and voting. We believe in: rough consensus and running code.”²⁶ A standard is what demonstrably works in several independent implementations. This way of working prevailed over the alternative: the International Organization for Standardization had adopted in 1984, with the OSI model, a formally complete set of rules for computer networks in seven layers, backed by governments and the telephone companies of Europe. OSI was more careful, but TCP/IP ran, was included free in every Unix and had users. By the early 1990s the matter was settled, and what remained of OSI was the vocabulary of the seven layers.²⁷

Try it yourself. Name resolution can be watched with one command. nslookup stockter.info (on Linux and macOS also dig stockter.info) asks the resolver of your own connection and shows the address it returns. dig +trace stockter.info performs the whole chain, from the root server through the .info servers to the responsible server. And ip addr or ipconfig shows your own address in the home network, usually one beginning with 192.168: a private range that the router hides behind a single public address, because the 4.3 billion were not enough.

4. The World Wide Web (1989–1994)

Finding as the bottleneck. In 1989 a scientist at any university in the Western world could fetch files from any other computer, if he knew on which computer they lay, in which directory, under which name, and what the program was called that could read them. The network was there; the information in it stayed hidden. First aids appeared: in 1990 Archie at McGill University in Montreal began to search the file lists of FTP servers, in 1991 Gopher at the University of Minnesota offered documents in menus through which one could click from computer to computer. Both were catalogues. What was missing was a way to point from inside one document directly to another, wherever it lay.

Hypertext. The idea for that was old. Vannevar Bush described in “As We May Think” in 1945 the Memex, a desk with microfilms in which the user can lay and store trails from one document to the next.²⁸ Ted Nelson coined the word hypertext in 1965 for text that is not linear but consists of pieces connected by links, and worked for decades on a system called Xanadu that was never finished.²⁹ Douglas Engelbart demonstrated on 9 December 1968 in San Francisco, in ninety minutes, a system that contained links, windows, video conferencing and a new pointing device, the mouse; the demonstration has been called the “mother of all demos” ever since.³⁰ There were several hypertext systems in 1989, but all for one computer or one database. None reached across the network.

The proposal. Tim Berners-Lee, a British physicist and programmer at the research centre CERN in Geneva, had a practical problem: thousands of scientists from all over the world came for months, brought their own computers, programs and documentation and took their knowledge away with them again. In March 1989 he handed his superior a paper entitled “Information Management: A Proposal”, which wanted to join hypertext with the Internet: every document gets an address, every link contains the address of its target, and a program fetches the target on a click from any computer in the world. His boss Mike Sendall wrote on it: “Vague but exciting”, and let him go ahead.³¹ With Robert Cailliau he worked the proposal out in 1990, and by Christmas 1990 he had written all three building blocks on a NeXT computer.

The three building blocks are the web to this day. The URL (Uniform Resource Locator) is the address of a document: a scheme that says how to fetch it, the name of the computer and the path on it, such as https://stockter.info/en/glossary.html. HTML (Hypertext Markup Language) is the language the document is written in: text with markup in angle brackets for headings, paragraphs and above all for links, <a href="…">, which point to a URL. And HTTP (Hypertext Transfer Protocol) is the protocol with which a program requests the document: the browser opens a TCP connection to the computer, sends a line such as GET /en/glossary.html, and the web server answers with a status code and the document.³² The first website was info.cern.ch and explained what the web is. On 6 August 1991 Berners-Lee presented the project publicly in a Usenet forum.

This is what one of the first pages looked like in source, and a browser of today still displays it just the same:

<html>
<head><title>Example</title></head>
<body>
<h1>A page</h1>
<p>A paragraph with a
<a href="https://stockter.info/en/glossary.html">link</a>.</p>
</body>
</html>

Mosaic and the release. Two decisions turned the tool of a physics laboratory into a world medium. On 30 April 1993 CERN declared in writing that it waived all rights to the web software; anyone could use, change and pass it on freely.³³ Berners-Lee had pressed for this because he had seen how Gopher lost ground after the University of Minnesota announced licence fees in 1993. And in the same year Marc Andreessen and Eric Bina at the supercomputer centre NCSA of the University of Illinois released the browser Mosaic: easy to install, for Windows, Macintosh and Unix, and with images in the middle of the text instead of in a separate window.³⁴ Overnight the web was something one could look at. In mid-1993 there were about 130 websites, at the end of 1994 more than 10,000, in 1997 over a million.²²

Andreessen went to California and founded the company Netscape with Jim Clark in 1994, whose browser Navigator appeared at the end of the year and conquered the field in months. Berners-Lee went to MIT and founded the W3C (World Wide Web Consortium) in October 1994, which has developed the standards of the web openly and without fees ever since. Finding was solved, as far as one had a link, and the next question already arose: who pays for all this, and who may take part? The NSFNET was a research network, and its rules of use excluded commercial traffic.

Try it yourself. Every browser shows, with the command “View page source” (usually Ctrl+U), the HTML of the page it is currently displaying, this one included. Whoever wants to see the request itself types curl -v https://stockter.info/ in the terminal and sees the lines the browser sends and the header lines of the reply with the status code 200 OK. And a page of your own needs no server: save the source above as a file page.html and open it in the browser. That was the whole technology in 1991, and it still is.

5. Commerce, search and bubble (1995–2001)

The opening. Until 1991 the Internet in the United States was a research network; whoever did business over it broke the rules of the National Science Foundation. Commercial providers had built their own networks, UUNET since 1987, PSINet since 1989, and in 1991 they connected these with each other to get round the research network and its ban. In the same year the NSF relaxed its usage rules and for the first time admitted commercial traffic on its backbone, and in 1992 Congress expressly allowed it to do so by law.³⁵ On 30 April 1995 the NSF switched off its backbone; the Internet now consisted of the networks of private Internet service providers (ISPs), which hand traffic to one another at exchange points, mostly without money, on the principle that it benefits both. In Germany access had first arisen at universities, then through small firms from their surroundings, and in 1995 Telekom opened its videotext service to the Internet under the name T-Online; by 1997 about four million Germans were online.

For most people the Internet began with a sound: the whistling and hissing of the modem dialling in to a provider over the telephone line. In 1994 a modem transferred 28.8 kilobits per second, in 1998 56; a page with images took half a minute, and as long as the connection stood the telephone was engaged and the meter ran. Billing was by the minute.

Commerce and encryption. In July 1995 Amazon sold its first book, in September eBay began as the auction site of a programmer in San José, and Netscape went public on 9 August 1995: the share, issued at 28 dollars, closed the first day at 58. That one could send a merchant a credit-card number over the network was owed to an invention by Netscape in the same year: SSL (Secure Sockets Layer), a protocol that encrypts the connection between browser and server and identifies the server by a certificate, an attestation signed by a trusted body that this key belongs to this name. SSL became the standard TLS in 1999, and http:// became https:// for the protected case.³⁶ But encryption remained the exception for checkouts and login pages; everything else ran through the network in plain text for two decades.

The web learns to walk. Netscape added a programming language to its browser in 1995, JavaScript, designed in ten days by Brendan Eich, with which a page could respond to input without asking the server; in 1996 came CSS (Cascading Style Sheets), a language for type, colour and layout that separates appearance from content. The three languages HTML, CSS and JavaScript have been the material of every web page ever since. Added to them were programs on the server that generated pages only on request, from a database, differently for each visitor: the web turned from a collection of documents into a surface for applications.

The browser war. Microsoft had underestimated the Internet in 1995 and caught up with force: Internet Explorer was shipped free with Windows and from 1997 built deep into it. Netscape’s share fell from over 80 per cent in 1996 to under 20 per cent in 2000. The American Department of Justice sued in 1998 for abuse of a monopoly, a court ordered a break-up in 2000, which an appeals court overturned in 2001; it ended in a settlement with conditions.³⁷ Netscape opened the source code of its browser in 1998; from it came Mozilla and, in 2004, Firefox. That browsers are free and their makers earn their money elsewhere dates from this war.

Search. With millions of pages the link alone was no longer a way to find things. Yahoo began in 1994 as the hand-tended catalogue of two students at Stanford, AltaVista showed at the end of 1995 that a computer could search the full text of the whole web, and dozens of search engines followed, all with the same problem: for one word there were ten thousand pages, and which belonged at the top they did not know. Larry Page and Sergey Brin, also students at Stanford, solved it in 1998 with an idea they called PageRank: a page is important if many important pages link to it. The links Berners-Lee had conceived as paths were read as votes. In April 1998 they presented the method at a conference, their prototype had indexed 24 million pages, and in September they founded Google.³⁸ The search engine became the most important door into the web, and from 2000 Google sold, next to the results, small advertisements matching the search words. That was the business model that paid for the web for two decades: advertising based on what the user wants right now.

For a decade the link, not the operator, decided what stood at the top. An example of the method in numbers: if three pages point to a fourth, and each of the three has a value of 1, the fourth receives 3 divided by the number of links each of the three sends out. Whoever is named by a page with value 1 and ten outgoing links inherits a tenth; whoever is the only one named by it inherits everything. The whole is repeated until the values no longer change.

Administration. Jon Postel had administered the numbers and names of the network single-handedly since the 1970s, as a service to the community. In 1998 the American government handed this task to ICANN, a non-profit corporation in California that assigns the endings, coordinates the root servers and passes the address blocks on to five regional registries; Postel died in October of the same year. The oversight of the American government ended in 2016. The question of who owns the network has had a practical answer since: the lines belong to the providers, the names and numbers are administered by a foundation, the rules are written by the IETF, and nobody has everything.

The bubble. That the web was worth money, everyone knew in 1999; how much, nobody. Firms without revenue were valued at billions because they were on the net, and whoever asked too long about a business model counted as backward. The Nasdaq, the exchange of technology firms, reached its peak of 5,048 points on 10 March 2000 and lost three quarters of its value by October 2002.³⁹ Hundreds of firms disappeared, with them tens of thousands of jobs and the fibre-optic lines that had been laid in reserve. The lines stayed in the ground, and whoever used them later paid little. The bubble had pre-financed the next era.

The network reached the music market in 1999 through Napster, a program with which users fetched music files from the computers of other users, without a server in the middle. Two years later it had 80 million users, and a court closed it; but the principle that every computer can be provider and customer at once, peer-to-peer, lived on in BitTorrent, and the music industry found an answer only in 2003 with the iTunes Store and in 2008 with Spotify.⁴⁰ The bottleneck of this era was access: the network was for most people a telephone connection billed by the minute, and the pages in it were made for reading, not for writing.

6. Broadband, Web 2.0 and the smartphone (2001–2012)

Always on. Two technologies ended the modem. DSL sends data over the existing copper line of the telephone in a frequency range above speech, so that one can telephone and be online at once, at first at 768 kilobits per second, in 2006 at 16 megabits; in Germany Telekom offered it from 1999, and in June 2000 came the first flat rate, which abolished billing by the minute.⁴¹ The television cable brought comparable rates. Broadband was more than speed: the connection was always there, and the network turned from a place one dialled into to a condition. The number of users kept pace with the growth of the lines: in 2005 a billion people were online, in 2010 two.

The web for writing. The ruins of the bubble were the material of the next wave. On 15 January 2001 Wikipedia went online, an encyclopedia anyone may edit; by 2005 the English edition had 750,000 articles, by 2025 almost seven million, and the project proved that people without pay and without an editorial board write a reference work that keeps up with printed ones.⁴² Weblogs, blogs for short, made everyone a publisher. A web that felt like a program became technically possible through a technique that Jesse James Garrett christened Ajax in 2005: JavaScript fetches data from the server in the background and inserts it into the page without reloading it.⁴³ Google Mail of 2004 and Google Maps of 2005, with its map that could be dragged with the mouse, showed what could be done with it. Tim O’Reilly summed up the new wave in 2005 under the name Web 2.0: the web as a platform, content from the users, services that get better the more people use them.⁴⁴

The services that Web 2.0 brought forth are the platforms of today. Facebook began in February 2004 in a student room at Harvard, YouTube in February 2005 above a pizzeria in California, Twitter in March 2006. The number of websites kept rising, but the time of the users gathered on a few. The social network gave the net something it never had: an identity that holds across pages, and a stream of messages that an algorithm orders by what keeps the user longest. The advertising that Google had attached to search words now hung on persons.

The cloud. In 2006 Amazon began to rent out by the hour the computers it ran for its shop: first storage, then whole virtual machines. What the history of software development describes as the cloud is for the Internet the return of time-sharing on another scale: the computer on which a service runs stands somewhere, is paid for by consumption and can become a hundredfold in minutes. A student with a credit card had access to computing power for which, ten years earlier, a data centre would have been needed. From 2000 came content delivery networks, which keep copies of popular content in hundreds of data centres close to the users; whoever fetches a video today rarely gets it from the provider’s server, but from a cache in their own city.

The net in the pocket. Mobile phones could get online from 2000, over GPRS at the speed of an old modem and on pages written specially for two lines of display. On 9 January 2007 Steve Jobs presented the iPhone, with a full browser on a screen operated with the fingers; in 2008 followed the App Store, in which outsiders could offer programs for the device, and in the same year the first phone with Google’s Android.⁴⁵ The mobile network followed: UMTS brought hundreds of kilobits, LTE from 2010 tens of megabits, 5G from 2019 ten times that. In October 2016, for the first time, more web traffic came from phones and tablets than from desktop computers.⁴⁶ For billions of people in Africa and Asia the phone was the first and only access to the net, and the app, a program that talks to its maker’s server, replaced the browser for many purposes. The web was no longer the Internet; it was one of its services, next to the closed worlds of the apps.

The addresses run out. On 3 February 2011 the central administration assigned the last free block of IP addresses to the regional registries; by 2019 the European registry had none left either.⁴⁷ The answer had stood ready since 1998: IPv6, the new version of the Internet Protocol, with 128 instead of 32 bits, that is 2¹²⁸ addresses, a number with 39 digits, written in eight hexadecimal groups, such as 2001:db8::7. The changeover was the second of the network after 1983, but without a cut-over day, because no one could order it, and it is still going on: in 2025 about half of the traffic to Google ran over IPv6, in Germany more than two thirds.⁴⁸ That both versions run side by side and the old one is kept alive by tricks such as sharing one address among many devices is the price of a network without a centre being unable to force anyone.

The bottleneck at the end of this era was no longer technical. The network was everywhere and always there. But the services through which people used it belonged to a handful of firms, and what those did with the data, nobody knew.

7. Platforms, encryption and rules (since 2013)

The disclosure. On 5 and 6 June 2013 the Guardian and the Washington Post published documents of the American intelligence agency NSA that Edward Snowden, an employee of a contractor, had taken with him. They showed that the NSA collected the call records of American telephone customers, tapped traffic at the fibre-optic cables and had access through a programme called PRISM to data of the big Internet firms.⁴⁹ For the engineers of the network the consequence was unambiguous: in 2014 the IETF declared pervasive surveillance an attack against which the network must be protected, and resolved to make encryption the norm.⁵⁰ The lever came in 2015 with Let’s Encrypt, a non-profit certificate authority that issues certificates free and automatically; before, they cost money and work. At the end of 2015 just under four in ten page loads were encrypted, at the end of 2016 for the first time half, in 2026 around nine in ten, and browsers warn about the rest.⁵¹ What can be read along the way since then is only who talks to whom, no longer what.

The platforms. The second lesson concerned not states but firms. In 2018 it became known that the firm Cambridge Analytica had obtained the data of about 87 million Facebook users through a survey app and used it for election advertising.⁵² The platforms, celebrated in the 2000s as places of taking part, now appeared as what they were economically: collectors of behavioural data that sell attention to advertisers and whose algorithms favour whatever provokes reactions. Google and Meta together received about half of worldwide online advertising spending in 2024.⁵³ The link that had ordered search in 1998 counted for less than the stream the algorithm assembled.

The rules. The answer of politics came first from Europe. The General Data Protection Regulation (GDPR), adopted in 2016 and applicable since 25 May 2018, gave every person in the EU the right to learn what a firm stores about them, to have it deleted and to object to processing, with fines of up to four per cent of worldwide turnover.⁵⁴ The cookie banners that every site has shown since are its most visible and most disputed consequence. In 2022 followed the Digital Services Act, which obliges large platforms to procedures against unlawful content, to disclosure of their recommendation algorithms and to audits by third parties, and the Digital Markets Act, which forbids the largest platforms to favour their own services and to lock users in.⁵⁵ Net neutrality, the principle that a provider treats all packets alike and neither slows nor favours any service, has stood in EU law since 2016; in the United States it was introduced in 2015, abolished in 2017, reintroduced in 2024 and finally struck down by a court in January 2025.⁵⁶ The principle that the end-to-end argument had justified technically in 1984 became a political question, and the answers differed by continent.

The divided net. From the beginning the Internet had been a network without borders; that states would not leave it so showed early. China built from 1998 a system of filters at its exchange points, the Great Firewall, which blocks foreign services and searches domestic traffic; Google, Facebook and Wikipedia are unreachable there, and the domestic market brought forth its own platforms, which are among the largest in the world. Russia obliged its providers in 2019 to install equipment for central control of traffic, so as to be able to cut the country off from the rest of the net. The organisation Access Now counted 296 deliberate Internet shutdowns in 54 countries worldwide in 2024, at elections, protests and examinations, more than ever before.⁵⁷ The word splinternet names the worry that the one network will become several that touch only at controlled crossings.

Under the sea and in orbit. The lines themselves remained what they had been since 1866: cables through the sea. In 2025 about 600 submarine cables lay with a combined length of some 1.5 million kilometres, over which almost all traffic between the continents runs; a single modern cable carries several hundred terabits per second, a millionfold the first fibre-optic cable of 1988.⁵⁸ That they are vulnerable was shown by anchor damage in the Baltic Sea in 2023 and 2024, which triggered investigations into sabotage. As a supplement came the net from orbit: Starlink began in 2019 to put thousands of small satellites into low orbits, by 2025 there were over 7,000, and they supply places without cables, ships, aircraft and, since 2022, Ukraine at war, whereby a single company decided where the net is.⁵⁹ In Germany fibre lay in front of just over half of households and businesses by mid-2025, but had been brought into the building for only just over a quarter and was in use on some 6.6 million lines; the larger part of the traffic continued to run over the copper that had been laid for telephone and television.⁶⁰

The protocols renew themselves. Beneath the surface more changed than the users noticed. HTTP, barely altered since 1997, received a second version in 2015 that bundles many requests over one connection, and a third in 2022 that sits on QUIC, a transport protocol over UDP designed by Google that sets the connection up faster and does not break on a switch from Wi-Fi to the mobile network.⁶¹ TLS 1.3 of 2018 struck out the insecure methods of two decades. The DNS, which had asked unencrypted since 1983 and thereby betrayed which sites someone calls up, can run over HTTPS since 2018. And the IETF works, without kings or voting, by the same procedure as in 1986, only that there are now around 10,000 RFCs.

The machines read along. The latest turn comes from the history of AI. The language models that have written texts since 2022 were trained on the web; the project Common Crawl, which has copied the open web regularly since 2008, comprises several hundred billion pages and was the basis of most models.⁶² Since 2023 programs of the AI firms have read websites systematically, and operators resist: the file robots.txt, with which a website has told search engines since 1994 what not to index, became the front line, publishers sued over copyright, and the network operator Cloudflare, through which a fifth of all websites run, has blocked the AI programs by default since July 2025.⁶³ At the same time search engines have answered questions since 2024 with a paragraph written by AI above the results, so that the link on which websites lived for two decades is clicked less often. Then there is the flood: texts and images that machines generate in seconds fill the web faster than people read it, and models that train further on such material learn from themselves. What stands on the web was never reliable; that it was written by a human was, until 2022, at least probable.

The Internet resolves, in six decades, one bottleneck after another: the line, the island, the incompatibility, the finding, the access, the taking part. What remains open is trust: who reads along, who decides what one sees, and whether what one sees comes from somebody. The technology that built the network, packets, protocols, end-to-end, is not responsible for that. It carries packets and nothing else, and precisely that was its success.

Try it yourself. A click on the padlock next to the address in the browser shows the certificate of this site: who issued it, for which name, until when. The page https://stockter.info/robots.txt shows what this website allows search engines and AI programs. The site test-ipv6.com reveals whether your own connection already speaks the new address version. And whoever wants to know what a page looked like in 1996 finds it in the Internet Archive at web.archive.org, which has copied the web since 1996 and reached a trillion pages in October 2025.

The chain in brief

1. Telegraph and telephone (1844–1950s) connect people over lines, but every connection occupies a whole line, and every computer is an island.
2. Packet switching and ARPANET (1961–1972) share the lines and connect the computers. But every network speaks its own language.
3. TCP/IP (1974–1983) lays a common layer over all networks and makes them one. The information in it remains hard to find.
4. The World Wide Web (1989–1994) links documents by references and gives everyone an address. The network is a research network, and access is expensive.
5. Opening, commerce and search (1995–2001) make the network a market and the search engine the door. The network is a telephone connection, and the pages are only for reading.
6. Broadband, Web 2.0 and smartphone (2001–2012) make the network a permanent condition and everyone an author. The services belong to a few firms.
7. Encryption, rules and AI (since 2013) answer surveillance and platform power. What remains open is whom one can trust on the net.

Timeline

Further reading

• Janet Abbate: Inventing the Internet. MIT Press, Cambridge 1999. The standard history of the ARPANET and the Internet up to the web, from the sources.
• Katie Hafner, Matthew Lyon: Where Wizards Stay Up Late. The Origins of the Internet. Simon & Schuster, New York 1996. The making of the ARPANET as narrative, with the participants as witnesses.
• Tim Berners-Lee, Mark Fischetti: Weaving the Web. HarperCollins, San Francisco 1999. The inventor of the web on his idea and what became of it.
• Andrew Blum: Tubes. A Journey to the Center of the Internet. Ecco, New York 2012. A journey to the cables, exchange points and data centres of which the net physically consists.
• Andrew L. Russell: Open Standards and the Digital Age. History, Ideology, and Networks. Cambridge University Press, New York 2014. How standards arise and why TCP/IP beat OSI.
• Jonathan Zittrain: The Future of the Internet and How to Stop It. Yale University Press, New Haven 2008. Why the open net admitted every new service and what it loses through apps and platforms.
• Shoshana Zuboff: The Age of Surveillance Capitalism. PublicAffairs, New York 2019. The critique of the platforms’ business model.
• Barry M. Leiner et al.: A Brief History of the Internet. Internet Society 1997, continually updated. The history told by the participants themselves, Cerf and Kahn included; free online.

Sources

The chronicle rests on the original papers and contemporary reports named in the footnotes.