The Internet: From ARPANET to the Network in Your Pocket

Follows: The Silk Roads: The World's First Internet Ran on Camels and Graph Theory: From the Bridges of Königsberg to Your GPS

In this deep dive
1. The idea before the machine: Licklider, Baran, Davies
2. ARPANET, 1969: "lo"
3. A network of networks: TCP/IP, DNS and the 1983 switch
4. The web and the browser
5. Boom, bust, broadband and the phone
6. How a packet actually travels
7. Platforms and social media
8. The state of affairs in 2026, and the open questions

At 22:30 on 29 October 1969, a graduate student at UCLA named Charley Kline sat down to log in to a computer at the Stanford Research Institute, about 500 kilometres up the California coast. He typed an L. The other end confirmed it. He typed an O. Confirmed. He typed a G, and the system crashed.

So the first message ever sent over the network that became the internet was "lo", as in "lo and behold." They got the full login working about an hour later. The professor supervising, Leonard Kleinrock, still keeps a scan of the handwritten log on his website.

Today about 6 billion people use the descendant of that link. The Silk Roads showed how a web of relays, where almost nothing travelled end to end, could carry religions and inventions across a continent. Graph Theory showed how throwing away everything but "what connects to what" lets you reason about any network. The internet is both ideas made literal: a relay system of dots and lines, where your message is handed from machine to machine and nobody runs the whole thing. This deep dive is the history of how it was built, a plain explanation of how it works, and an honest look at where it stands now.

Part 1 — The idea before the machine: Licklider, Baran, Davies

In 1960 most computers were room-sized machines that worked through stacks of punched cards. You handed in your job and came back hours later. A psychologist named J. C. R. Licklider thought that was the wrong relationship. In a paper that year, "Man-Computer Symbiosis," he imagined people and computers working together in real time, each doing what it was best at.

In August 1962 he went further. In a series of memos he described a "Galactic Network": computers all over the world linked so that anyone could reach data and programs at any site. That October he became the first head of the computer research programme at ARPA, the Pentagon's research agency, and in an April 1963 memo he addressed his far-flung research contractors as "Members and Affiliates of the Intergalactic Computer Network." He didn't build the network himself. What he did was convince the people who came after him at ARPA, including Ivan Sutherland, Bob Taylor and Larry Roberts, that it mattered. In 1968 he and Taylor published "The Computer as a Communication Device," predicting online communities of shared interest that would form regardless of where their members lived.

Licklider's real invention wasn't a machine. It was a reason to build one: computers as a way for people to talk to each other.

The problem with phone lines

The obvious way to connect computers was the telephone network. But the phone system is circuit-switched: when you call someone, a continuous line is reserved for you for the whole conversation, even during the silences. Computer traffic is bursty, a flurry of data and then nothing, so reserving a whole line wastes most of it. And a network built around a few big switching centres has a few big points of failure.

Two men, working separately on different sides of the Atlantic, came up with the same answer.

Paul Baran was an engineer at the RAND Corporation in California. In the early 1960s, with the Cuban missile crisis making nuclear war feel close, the question he worked on was how a military command network could survive an attack that destroyed many of its nodes. His answer, laid out in RAND studies from 1960 to 1962 and then in On Distributed Communications (1964), had two parts. First, build a mesh with many redundant links, not a hub with spokes. Second, chop each message into small "message blocks," send each one separately, let every node pass blocks along toward their destination by whatever route was free (he called it "hot-potato routing"), and reassemble them at the end.

Donald Davies at Britain's National Physical Laboratory (NPL) arrived at almost the same design in 1965, with a different goal: a commercial national data network for Britain. He called the pieces packets, and that's the word that stuck. His team built a working packet network inside the NPL around the end of the decade.

Contested credit, and a myth

Who invented packet switching? The honest answer is "several people, independently." The Internet Society's history, co-written by many of the pioneers, notes that the work at MIT, RAND and the NPL "proceeded in parallel without any of the researchers knowing about the other work." That same history, which Kleinrock co-wrote, credits him with the first paper on packet switching theory, in July 1961. Others disagree. The Computer History Museum's timeline describes his thesis as "a seminal text on queuing theory" and calls Davies "the English inventor of packet switching." Read that way, Kleinrock analysed how messages queue in a network, while Baran and Davies designed the chopping-up. It's a live argument among historians, and you'll see both versions in print.

The myth is cleaner to correct. You'll often read that the internet was built to survive a nuclear war. Baran's work was about exactly that, but ARPANET was a different project. The Internet Society's history says flatly that the claim "was never true of the ARPANET, only the unrelated RAND study." RAND's own account says ARPANET was "intended for scientists and researchers who wanted to share computers remotely." Computers were extremely expensive, and ARPA was paying for a lot of them. Letting researchers use each other's machines over a network was cheaper than buying everyone their own.

Part 2 — ARPANET, 1969: "lo"

In 1966 Bob Taylor, by then running ARPA's computing office, got money to link the computers of all the research groups ARPA funded, and persuaded Larry Roberts to come from MIT to lead the project. At a conference in 1967, Roberts heard Roger Scantlebury of Davies's team present the NPL design, learned from him about Baran's work too, and adopted the word "packet." After that meeting the plan's line speed went up from 2.4 to 50 kilobits per second.

The key design choice was not to make the big host computers do the networking themselves. Instead, each site would get a small dedicated computer, the Interface Message Processor (IMP), whose only job was to pass packets on. A team at the firm Bolt Beranek and Newman (BBN), led by Frank Heart, won the contract to build them in December 1968, with Bob Kahn playing a major role in the design. IMPs were the ancestors of today's routers.

That first RFC deserves a moment. Crocker set up the series as an informal, fast way for network researchers to share ideas, because academic journals were far too slow. The habit stuck. Proposals were published openly, anyone could argue with them, and what worked became standard. The RFC series is still how internet standards are written. On 31 October 2026 the RFC Editor's index had passed number 10,000.

Email was the surprise. ARPANET was built to share computers, but the application people actually loved was sending each other messages. The Internet Society's history dates Ray Tomlinson's email software to March 1972. Other accounts put his first network email in 1971. Either way, Tomlinson is the reason addresses have an @ in them: he used it to separate a user's name from their machine's. The Internet Society's authors call email the network's first "hot" application. People wanted to talk to each other, just as Licklider had said.

ARPANET was built so researchers could share expensive computers. Within a few years, its most popular use was gossip.

Part 3 — A network of networks: TCP/IP, DNS and the 1983 switch

By the early 1970s ARPANET was not the only packet network. ARPA itself was experimenting with packet radio (networks over the airwaves, for vehicles and soldiers) and packet satellite links. France had CYCLADES, designed by Louis Pouzin, first demonstrated in November 1973. Pouzin coined the word datagram for a packet sent on its own with no guarantee it would arrive, and the word catenet for networks chained together through gateways. Each network worked differently. The question became: how do you connect networks that know nothing about each other?

Bob Kahn, by then at ARPA, and Vint Cerf, then at Stanford, worked on the answer from the spring of 1973 and published it in May 1974 as "A Protocol for Packet Network Intercommunication." Kahn's ground rules, in the Internet Society's paraphrase, are still the internet's philosophy:

1. Each network stands on its own. Nothing inside it has to change to connect.
2. Delivery is best effort. If a packet doesn't arrive, the sender sends it again.
3. The boxes connecting networks (gateways, later called routers) stay simple. They don't remember the packets they pass.
4. There's no global control at the operations level.

That design pushed intelligence to the edges. The network in the middle just moves packets as well as it can. The computers at each end check that everything arrived, put it back in order and ask for anything missing. Over the next few years the protocol was split in two: IP (Internet Protocol) gives every machine an address and gets packets roughly to the right place, and TCP (Transmission Control Protocol) runs on the end computers and makes the conversation reliable. An RFC from December 1974 already used the short word in its title: "Specification of Internet Transmission Control Program." The final specifications, RFC 791 for IP and RFC 793 for TCP, appeared in September 1981.

Flag day

TCP/IP became a US defence standard in 1980. In November 1981 Jon Postel published RFC 801, a plan for every ARPANET host to switch from the old protocol to TCP/IP. The deadline was 1 January 1983, a "flag day": every host had to convert at once, or be left talking to the network through makeshift workarounds. It went more smoothly than people feared, and participants handed out buttons reading "I survived the TCP/IP transition." The switch also let the military split its operational traffic off into a separate network, MILNET, leaving ARPANET for research.

1 January 1983 is often called the internet's real birthday, because it's the day the network of networks started running on one common language.

Names instead of numbers

Every machine on the internet has a numeric address. People are bad at numbers, so early on every computer kept a single shared file listing every host's name and address. That stopped working once thousands of independently run networks were joining. Paul Mockapetris at the University of Southern California designed the Domain Name System (DNS), described in RFCs 882 and 883 in November 1983 and revised in RFCs 1034 and 1035 in November 1987. DNS spreads the job out. No one computer knows every name. Instead, servers for ".org" know who's responsible for each ".org" name, those servers know about the names under them, and so on. It's a distributed phone book with no central copy, which is why it scaled.

From research to everybody

In the mid-1980s the US National Science Foundation built NSFNET to connect universities and supercomputer centres, spending about $200 million between 1986 and 1995. It became the internet's backbone. In 1990 the original ARPANET was switched off. The NSF lifted restrictions on commercial use in 1991, and in April 1995 it stopped funding the backbone altogether, handing the job to private long-haul networks. The routing protocol that lets those independent networks tell each other which addresses they can reach, the Border Gateway Protocol (BGP), was first published in June 1989 and, in later versions, still holds the internet together.

Contested credit, again

In a CNN interview on 9 March 1999, Vice President Al Gore said: "During my service in the United States Congress, I took the initiative in creating the Internet." It was widely mocked and misquoted as "I invented the internet." He didn't say that. He did, as a senator, push the 1991 legislation that the Computer History Museum's timeline calls the "Gore Bill," which created a National Research and Education Network programme. Whether "took the initiative in creating" was fair is a reasonable thing to argue about. The engineers who built TCP/IP were Cerf and Kahn and many others.

Part 4 — The web and the browser

Here's the most common confusion in this whole subject: the internet and the web are not the same thing. The internet is the network: the cables, routers and protocols that move packets. The web is one service that runs on it, like email or video calls.

The web was invented by Tim Berners-Lee, a British software engineer at CERN, the European particle-physics laboratory near Geneva. While consulting at CERN in 1980 he had written a personal program called ENQUIRE for linking notes together. In March 1989 he wrote a proposal titled "Information Management: A Proposal" for a hypertext system to help CERN's thousands of scientists find each other's documents. His boss, Mike Sendall, approved the purchase of a NeXT computer in September 1990 and let him build it.

By Christmas 1990 Berners-Lee had three inventions working together:

1. HTML, a simple way to mark up a document with headings and links.
2. HTTP, the rules a browser uses to ask a server for a document.
3. URLs, an address format that can point to any document on any server.

He also wrote the first browser, called WorldWideWeb, and the first web server. The first website, at info.cern.ch, explained the project itself, and CERN still keeps a copy online. In August 1991 he posted the software publicly on Usenet newsgroups. On 30 April 1993 CERN's directors signed a statement that the web's technology could be used by anyone, with no fees payable to CERN. The W3C's history calls it "a milestone document." Timing mattered: that February, the University of Minnesota had announced it would start charging licence fees for Gopher, a competing system, and many users moved to the web.

The browser that changed everything

The early web was text-heavy and ran mostly on university machines. In February 1993 the National Center for Supercomputing Applications at the University of Illinois released the first alpha version of Mosaic, a graphical browser by Marc Andreessen and colleagues. Web traffic, which had been 0.1% of the NSF backbone in March 1993, reached 1% by September. In 1994 Andreessen and others left to start the company that became Netscape. Berners-Lee founded the World Wide Web Consortium (W3C) to steward the web's standards; it was founded on 1 October 1994.

On 9 August 1995, Netscape went public, only sixteen months after the company was formed. Its shares were priced at $28, traded as high as $75 on the first day, and closed at $58.25, valuing the company at about $2.9 billion. Many people date the dot-com boom from that day.

The web won because it was given away. A royalty-free standard beat systems that charged.

Part 5 — Boom, bust, broadband and the phone

The late 1990s were a gold rush. Investors poured money into any company with ".com" in its plan, many of them selling things online at a loss to grab market share. The technology-heavy Nasdaq Composite index peaked at 5,048.62 on 10 March 2000. By 9 October 2002 it had fallen to 1,114.11, a drop of 78%. Many companies failed, among them telecom firms such as WorldCom and Global Crossing.

The bust taught a lesson the crypto piece in this series repeats: a technology can be real and still be wildly overpriced. The fibre laid in the boom didn't vanish; it stayed in the ground for whoever came next. Google, incorporated in 1998, grew through the bust.

Broadband, then the phone

In the early years most homes reached the internet by dial-up modem, the screech of a computer calling another computer over a phone line. Broadband replaced it, first over the copper of phone lines (DSL) and cable-TV lines, then increasingly over fibre to the home. The change was more than speed. Always-on connections made it normal to leave the internet running, which is what made streaming video, online gaming and constant messaging possible.

Then the internet left the desk. On 9 January 2007 Apple introduced the iPhone, describing it as a phone, an iPod and "a breakthrough Internet communications device" in one. Smartphones running iOS and Android put a full browser, maps, a camera and thousands of apps in a pocket. The ITU calls mobile phones "the most common gateway to the Internet," and in many low- and middle-income countries a phone is the main way people get online.

The International Telecommunication Union (ITU), the UN agency for telecoms, published the latest global figures in October 2025:

That last pair surprises people. Even in the age of the phone, fixed lines carry about five times as much data as mobile networks, partly because phones at home switch to Wi-Fi, which runs over fixed broadband.

Under the sea

"Wireless" is mostly an illusion. As the research firm TeleGeography puts it, your phone's signal only travels by radio to the nearest cell tower. From there, it goes into fibre. Between continents, that fibre lies on the seabed. As of 2026 TeleGeography tracks more than 600 active and planned submarine cables, and as of early 2026 it estimated more than 1.5 million kilometres of them in service. They range from the 131-kilometre CeltixConnect between Ireland and the United Kingdom to the 20,000-kilometre Asia America Gateway. Satellites carry only a sliver of international traffic: figures from the US Federal Communications Commission put them at 0.37% of US international capacity. Cables used to be built by consortia of phone companies. Today the biggest investors include content companies such as Google, Meta, Microsoft and Amazon, which now lay their own.

Part 6 — How a packet actually travels

Here's what happens, step by step, when you tap a link in London to a news site whose server is in New York. It takes a fraction of a second.

1. Find the address. Your phone asks a DNS resolver (usually run by your provider or a public service) for the site's numeric IP address. If the answer isn't already cached, the resolver asks a root server which servers handle ".com," asks those which server handles the site's name, and asks that one for the address. Answers are cached, so most lookups take milliseconds.
2. Open a conversation. Your phone and the server exchange three short packets, a "handshake," to agree to talk (in TCP: SYN, SYN-ACK, ACK). Newer protocols fold this step into the next one to save time.
3. Agree on a secret. For an https address, your phone and the server run a TLS handshake, using the public-key maths from Cryptography: The Math That Keeps Your Secrets, so that everything after this is encrypted. Anyone in between can see which server you're talking to, but not what you say.
4. Ask. Your browser sends an HTTP request: "GET this page." The request, like everything else, is chopped into packets. Each packet carries a source address, a destination address and a sequence number, like an envelope with a page number.
5. Leave the house. Your packets go by Wi-Fi to your home router, then over your provider's line into its network.
6. Hop, hop, hop. Each router reads only the destination address, checks its routing table and passes the packet to the next router that's closer. Between independent networks, those tables are built by BGP, in which each network announces which addresses it can reach. This is the shortest-path problem from Graph Theory, solved continuously by thousands of machines that each see only their neighbours.
7. Cross the ocean. Somewhere along the way, your packets enter a submarine cable. London to New York is about 5,570 km in a straight line over the globe. Light in glass fibre travels at about two-thirds of its speed in a vacuum, about 204,000 km per second, so even a perfectly straight cable needs about 27 milliseconds one way, 55 there and back. Real routes are longer. Physics sets a floor no engineering can beat.
8. Arrive in pieces. Packets may take different routes and arrive out of order, and a few may be lost. The server's TCP software puts them back in order and asks for anything missing. Each packet also carries a "time to live" counter that every router lowers by one, so a packet lost in a loop eventually dies instead of circling forever.
9. Answer, many times over. The server sends back the page, and your browser discovers it needs dozens more files (images, styles, scripts, ads), often from other servers. Popular sites keep copies on servers near you, in networks called content delivery networks, so many of those files never cross the ocean at all.

Nobody plans your packet's route. Each router makes one local decision, and the global path emerges from millions of them.

Two numbers explain a lot about today's internet. The original addresses (IPv4) are 32 bits long, which allows about 4.3 billion of them, fewer than there are people. The last free blocks were handed out by the central registry on 31 January 2011, prompting the regional registries to announce on 3 February that the free pool was gone. The replacement, IPv6, uses 128-bit addresses, which allows about 340 undecillion (3.4 × 10³⁸). The world has been slowly moving to it ever since, and both still run side by side.

Part 7 — Platforms and social media

The early web was mostly pages that their owners published and everyone else read. In the 2000s the pattern flipped: sites became platforms where users supplied most of the content. Facebook launched in February 2004, YouTube in 2005. Add the smartphone in 2007 and its app stores, and the web became something you carry, check dozens of times a day, and contribute to without knowing any HTML.

The scale is hard to overstate. Meta reported that an average of 3.64 billion people used at least one of its apps (Facebook, Instagram, WhatsApp, Messenger) every day in September 2026. Its revenue for July to September 2026 was $64.6 billion, almost all from advertising. That's the business model of most of the consumer internet: services are free to use because your attention is sold to advertisers, and the more precisely a platform can predict what will hold your attention, the more that attention is worth.

The legal bargain

In the United States, a short clause passed in 1996 shaped all of this. Section 230 of the Communications Act, added on 8 February 1996, says that no provider of "an interactive computer service shall be treated as the publisher or speaker" of what its users post. In plain terms: if someone posts something defamatory, you sue the person who posted it, not the site. Defenders say this is what let the participatory web exist at all; no platform could host billions of posts if it were legally responsible for each one as a newspaper is for its articles. Critics say it lets the largest companies in history profit from harmful content with little accountability. Both sides have proposed changes for years.

The European Union has taken a more regulatory path. Its Digital Services Act requires platforms to label ads and say who placed them and why you're seeing them, stop targeting ads using sensitive personal data such as religion or sexual orientation, and avoid manipulative "dark pattern" designs. The largest platforms must keep public repositories of the ads they show.

The argument about effects

Few questions about the internet are argued more fiercely than what social media does to people. Here are the strongest versions of each side.

• The optimistic case: social platforms let anyone publish, organise and find their people. They've given voice to groups the old media ignored, helped families stay close across continents, and let small businesses and artists reach audiences directly.
• The worried case: engagement-maximising feeds reward outrage and misinformation, harvest personal data on an industrial scale, and may harm the mental health of some teenagers. Legislators in many countries, and in the US a bipartisan group of state attorneys general (with whom Meta announced an agreement on teen users in August 2026), now focus on platforms' design choices rather than just their content.
• The research reality: studies of social media and wellbeing have produced mixed results. Some find small average associations, others larger effects for particular groups or particular kinds of use. Researchers disagree, sometimes sharply, about how much is cause and how much is correlation. Treat confident claims in either direction with care, and see How to Read the News Without Losing Your Mind in this library.

Part 8 — The state of affairs in 2026, and the open questions

Who's online, and who isn't

The ITU's most recent figures, for 2025 (its 2026 edition was not yet public at the end of October 2026), are the best global snapshot:

Two facts stand out. First, the coverage gap is small and the usage gap is big. Since 96% of people live within reach of at least a 3G signal, most of the 2.2 billion offline people could technically connect. What stops them, the ITU's figures suggest, is mostly cost, skills and access to a device. Second, progress is slowing. The ITU notes that the gender gap hasn't narrowed at all since 2019 (its parity score was 0.92 in both years), and in low-income countries only one in seven rural people is online. Prices are falling: the median cost of a basic mobile data plan fell to 1.4% of average income in 2025, but a fixed broadband subscription in a low-income country still costs more than a quarter of average income.

Satellites come back

Satellites used to be a last resort, because traditional communication satellites sit in geostationary orbit about 35,786 km up. A request and its reply must travel up and down twice, so even at the speed of light a round trip takes about 477 milliseconds before any processing, enough delay to make video calls awkward.

The new constellations fly much lower. SpaceX's Starlink satellites orbit below 600 km, cutting that physics delay to a few milliseconds, at the price of needing thousands of satellites because each one only sees a small patch of Earth for a few minutes. In October 2026 SpaceX said it operated more than 11,000 Starlink satellites and planned for up to 100,000. The astronomer Jonathan McDowell, who tracks every launch, counted 11,190 Starlink satellites in orbit on 29 October 2026. SpaceX's most recent customer figure I could verify was "more than 6 million" active customers, from July 2025. Since January 2024 some Starlink satellites have also connected directly to ordinary phones. Rival constellations are being launched, among them Amazon's Kuiper and China's Qianfan and Guowang (also called Xingwang).

Supporters point to coverage for ships, aircraft, disaster zones and remote villages where cables will never be economic. Critics raise the brightness of satellites in astronomers' images, collision and debris risk in crowded orbits, and the concentration of a critical piece of infrastructure in one private company. SpaceX itself publishes work on brightness mitigation and collision avoidance, which shows the concerns are taken seriously on both sides.

Net neutrality

Net neutrality is the principle that internet providers should treat all traffic equally: not block, slow down or charge extra for particular sites or apps. The European Union made it law in Regulation (EU) 2015/2120 of 25 November 2015, whose Article 3(3) requires providers to "treat all traffic equally," with exceptions for legal orders, network security and genuine congestion.

In the United States the rules have see-sawed with each administration. The Federal Communications Commission adopted strong rules in 2015, repealed them in 2017, and restored them in 2024. In January 2025 the federal appeals court for the Sixth Circuit struck the 2024 rules down, ruling that the FCC lacked the authority to classify internet access as a regulated telecommunications service. Since that ruling, net neutrality has not been required at the federal level.

Supporters argue that without neutrality, providers could become gatekeepers, favouring their own services and squeezing start-ups that can't pay. Opponents argue that heavy rules discourage investment in networks, that markets and competition law handle abuses better, and that some traffic (a remote surgery feed, say) genuinely deserves priority. Both sides claim to be defending the open internet.

One internet, or several?

The internet was designed to have no centre and no borders. Governments have been adding borders. China has long filtered foreign sites and services through what's often called the Great Firewall. Russia passed a set of amendments in 2019, informally the "sovereign internet" law, that mandate surveillance of internet traffic and give the state powers to cut the country's segment off from the rest of the internet, including a national version of DNS.

The bluntest tool is simply switching the internet off. The digital rights group Access Now and its partners documented at least 313 internet shutdowns in 52 countries in 2025, which they call the highest number since 2016. Governments that impose them usually cite security, exams or unrest. Critics see them as a way to silence protest and hide abuses.

Supporters of national control argue that states have a duty to protect citizens from crime, foreign interference and harmful content, and that a borderless internet mostly benefits the countries that host the biggest platforms. Defenders of the open internet argue that splitting it would damage trade, science and free expression, and hand governments too much power over what people can say and see. This "splinternet" debate is one of the defining arguments about the internet's future.

Open questions

• Can the last 2.2 billion get online? Coverage is nearly there. Affordability, devices, skills and electricity are not.
• Will satellites close the gap or just serve those who can pay? A Starlink terminal and subscription cost far more than a basic mobile plan in most low-income countries.
• Who governs a network nobody owns? Standards still come from open groups like the IETF that publish RFCs. Laws come from nearly 200 governments, each with its own idea of what the internet should be.
• What happens to the web as AI systems answer questions directly? If fewer people click through to the sites that create information, the advertising bargain that pays for much of the web may weaken. That's covered elsewhere in this series.
• How resilient are the cables? Undersea cables are routed around fishing grounds and anchorages, and the industry spends effort educating other marine industries about where they lie, because damage from ships is a real risk. A few cables carry a lot of the world's traffic.

The takeaway

The internet grew from Licklider's idea that computers could be a way for people to talk, Baran's and Davies's independent invention of packet switching, and ARPANET's first garbled "lo" in 1969. It became a network of networks on 1 January 1983, when every host switched to Cerf and Kahn's TCP/IP, a design that keeps the middle simple and puts intelligence at the edges. DNS gave machines names. Berners-Lee's royalty-free web, and the Mosaic browser, made it usable by anyone. A bubble burst, but the fibre stayed, and the smartphone put it in our pockets. Today about 6 billion people are online, carried mostly by undersea glass and by routers that each make one local decision at a time. The big questions now are human ones: who's still left out, who gets to set the rules, and whether one internet survives.

Try this: open a terminal on a computer (Command Prompt on Windows, Terminal on a Mac) and type tracert example.com on Windows or traceroute example.com on a Mac. You'll see every router your packets pass through, with the time each hop takes. Then try a site you know is hosted on another continent and watch the time jump when your packets cross the ocean. That jump is mostly the speed of light.

Sources & Further Reading

Sources
- Barry Leiner, Vint Cerf, David Clark, Robert Kahn, Leonard Kleinrock and others, A Brief History of the Internet, Internet Society (Licklider's 1962 memos and ARPA role, parallel packet-switching work, BBN and the IMPs, NCP, the 1972 demo and email, Kahn's ground rules, the 1974 paper, the 1983 flag day and MILNET, DNS, NSFNET funding and its 1995 defunding, ARPANET's 1990 decommissioning, and the footnote on the nuclear-war myth).
- Computer History Museum, Internet History timeline, 1960s–1990s (Licklider and IPTO, Taylor and Roberts, Davies and the word "packet," the line-speed change, NSF lifting commercial restrictions in 1991, the Gore Bill, Mosaic).
- RAND Corporation, "Paul Baran and the Origins of the Internet" (2018).
- Leonard Kleinrock, "The Day the Infant Internet Uttered its First Words," UCLA (the IMP log of 29 October 1969).
- RFC Editor: RFC 1 (7 April 1969), RFC 675 (December 1974), RFCs 791 and 793 (September 1981), RFC 801 (November 1981), RFC 882 (November 1983), RFC 1034 (November 1987), RFC 1105 (BGP, June 1989), and the RFC index of 31 October 2026.
- W3C, A Little History of the World Wide Web (ENQUIRE, the 1989 proposal, the NeXT, the 1991 release, Gopher fees, the 30 April 1993 declaration, Mosaic, backbone traffic shares, the founding of W3C). Apple press release, 9 January 2007. Google, "Our story."
- Nasdaq Composite closes: FRED series NASDAQCOM. Netscape IPO figures: contemporary reports as summarised in Wikipedia's Netscape article.
- ITU, Measuring Digital Development: Facts and Figures 2025 (October 2025): internet use, urban–rural and gender gaps, youth, coverage, subscriptions, traffic and affordability.
- TeleGeography, Submarine Cable Frequently Asked Questions (2026).
- Meta, Third Quarter 2026 Results (21 October 2026).
- 47 U.S.C. § 230, via Cornell Law School's Legal Information Institute. European Commission, Digital Services Act policy page. Regulation (EU) 2015/2120, EUR-Lex. Sixth Circuit opinion 25a0002p (January 2025), Ohio Telecom Association v. FCC.
- Number Resource Organization, "Free Pool of IPv4 Address Space Depleted" (3 February 2011).
- Starlink updates page (October 2026 constellation size; July 2025 customer count; January 2024 direct-to-cell texts); Jonathan McDowell, Jonathan's Space Report, Starlink statistics (29 October 2026).
- Access Now, Rising repression meets global resistance: internet shutdowns in 2025 (31 March 2026).
- CNN, Late Edition transcript, 9 March 1999 (Gore). Wikipedia articles on CYCLADES, Donald Davies, packet switching and the 2019 Russian internet laws, used as leads and checked against the sources above where possible.
- Fibre and satellite delays computed from the great-circle distance (London–New York about 5,570 km), a fibre refractive index of about 1.47, and a geostationary altitude of 35,786 km.

Go deeper
- 🌐 A Brief History of the Internet, Internet Society. Written by the people who built it. The best single source, with the bias that implies on questions of credit.
- 📘 Inventing the Internet, Janet Abbate (MIT Press, 1999). The historian's version: careful on contested credit, and on how users, not planners, decided what the network was for.
- 📘 Where Wizards Stay Up Late, Katie Hafner and Matthew Lyon (1996). The ARPANET story as a narrative, full of the people behind Part 2.
- 📘 Weaving the Web, Tim Berners-Lee (1999). The web's inventor on why he gave it away.
- 📘 Tubes: A Journey to the Center of the Internet, Andrew Blum (2012). A reporter visits the buildings, cables and landing stations. Makes Part 6 physical.
- 🌐 ITU Facts and Figures 2025. The official numbers on who's connected, updated every year.
- 🌐 TeleGeography Submarine Cable FAQs, with a link to its interactive cable map. Find the cable your packets probably use.
- 🌐 Companion pieces in this library: Cryptography: The Math That Keeps Your Secrets (what TLS does), Graph Theory (routing), Blockchain, Explained (Technology 3), and The 1990s in Decades of Leadership.