The Physical Internet: From Copper to Glass
I originally built the Internet Infrastructure Map because I thought that undersea fiber cables were interesting. Since then, I've been slowly piecing the history of the global Internet together. After adding historical fiber cables, I also started to add old telegraph and coax cables. Then I realized that the picture still wasn't complete because there is a period in the mid-1900s where many old cables were decommissioned without replacements. The last missing piece was point-to-point radio systems that became popular at the time.
The lines spanning the oceans on the map are submarine communications cables. Submarine cables are truly a technological wonder. They carry vast amounts of data across the world's largest oceans. Yes, they just lie at the bottom of our oceans. Yes, it is crazy. From route selection, cable manufacturing, and specialized cable-laying ships to amazing robots that can locate and assist in bringing cables to the surface and in-water fiber optic repair habitats, every bit of submarine cable deployment and operations is extreme and fascinating.
All of this said, I think the map now gives a fairly complete view of how worldwide communications have evolved. But this is just a map. There are many stories behind each technology and each cable that could be told. This document tries to give some context around the communications technology throughout modern history. I hope that this page and the map augment each other in interesting ways and spur even more questions in your mind.
Copper Pulses (1850s to 1920s)
If you dragged the timeline slider back to the early years on the old map, a lot of old cables were either missing or had the wrong activation and retirement years. I went through historical cable records back to 1858 to add the missing systems and fix retirement dates. You can also download the full list of cables from 1858 to the present as a CSV on the map.
In August 1858, the first transatlantic telegraph cable linked Ireland and Newfoundland, only to fail after three weeks. Across nearly 4,000 kilometers of seawater, the gutta-percha insulated wire acted like a giant capacitor that smeared sharp square pulses into slow curves, and Wildman Whitehouse blew out the insulation by trying to force messages through with 2,000-volt induction coils. When the steamship Great Eastern laid a permanent cable in 1866, Lord Kelvin and Fleeming Jenkin solved this by sending short, equal-duration positive and negative pulses (+V for a dot, −V for a dash) instead of long dashes that charged up the line, reading the wavy micro-ampere trace on a mirror galvanometer and siphon recorder tape.
I've been continuing my journey into discovering more about older technology by making silly games, so I also made a website at morse.kmcd.dev that allows you to test your speed skills in writing and interpreting Morse code, and it even gives you your (likely abysmal) words per minute.
Longwave Ether (1910s to 1940s)
Getting Around the Cable Monopoly
In the early 1900s, the British Empire owned most of the world's undersea telegraph cables. When World War I started, British ships cut Germany's cables right away, and any telegrams from neutral countries had to go through British censors in London.
If your messages travel through a wire on the bottom of the ocean, anyone with a grappling hook and a ship can cut you off. That pushed countries like the US, Germany, Sweden, and the Netherlands to build long-distance radio stations instead. In the 1910s, radio engineers thought only Very Low Frequency (VLF, 10 to 30 kHz) and longwave signals—with massive 10 to 20 kilometer wavelengths—could bend around the curve of the Earth as ground waves. Sending waves that long required continent-scale stations: mile-long wire canopies hung from rows of 120-to-250-meter steel masts and driven by hundreds of kilowatts from spark gaps, electric arcs, or spinning 200-kilowatt Alexanderson alternators.
Following Marconi's 1907 transatlantic link between Ireland and Nova Scotia, huge longwave stations popped up around the world: RCA Radio Central on Long Island (1921), Germany's Nauen transmitter, Britain's 16 kHz Rugby GBR (1926), and Sweden's 17.2 kHz Grimeton SAQ (1924, whose mechanical alternator still runs on special occasions). Because VLF waves can even penetrate 20 to 30 meters of seawater, navies later built megawatt VLF stations like Germany's Goliath (1943) and the US Navy's Jim Creek and Cutler transmitters to send messages to submerged submarines—which show up on the map as giant coverage rings.
Shortwave Beams & Transoceanic Voice Calls
In the mid-1920s, amateur radio experiments and ionospheric research showed that shortwave signals (HF, 3 to 30 MHz) refract off the ionosphere (~250 km altitude) and skip across the globe on less than 5% of VLF transmitter power. When Marconi's directional "Beam System" opened for Britain's Imperial Wireless Chain to Canada, Australia, South Africa, and India in 1926–1927, those shortwave beams carried so much traffic at a fraction of the cost of undersea cables that they nearly bankrupted the telegraph cable companies, forcing them to merge with Marconi into Cable & Wireless in 1929.
Even better, radio had enough bandwidth to carry human voice instead of just Morse code—something old gutta-percha telegraph cables couldn't do across oceans because capacitance blurred audio frequencies together. In January 1927, AT&T and the British Post Office opened the first commercial transatlantic phone service at 60 kHz longwave between New York and London, and by the early 1930s shortwave voice trunks connected North America, Europe, South America, Hawaii, and Tokyo. For nearly thirty years before the first coaxial phone cable (TAT-1 in 1956), every overseas phone call went over radio.
The Dutch Megaproject: Radio Kootwijk & Radio Malabar
Even though the Dutch were neutral during World War I, British censors refused to pass encrypted telegrams over subsea cables between Amsterdam and the Dutch East Indies (modern Indonesia). In 1918, the Dutch government started building a direct 12,000-kilometer VLF radio link between the Netherlands and Java. On Java, engineer Cornelis de Groot built Radio Malabar on Mount Puntang, south of Bandung, stringing a two-kilometer antenna cable across a mountain gorge with a 2.4-megawatt arc transmitter.
Two years after Kootwijk and Malabar entered service in May 1923, both stations added shortwave transmitters so they could carry voice calls in addition to telegrams. Queen Mother Emma inaugurated public phone calls between Amsterdam and Java in 1929 with the line "Hallo Bandoeng, hoort u mij?" ("Hello Bandung, do you hear me?"). A three-minute call cost 21 to 33 guilders, which was roughly a full week's wages at the time.
Coaxial Waves (1950s to 1980s)
Transoceanic Telephone Cables
Shortwave radio was much cheaper to build than VLF, but solar flares and changes in the ionosphere could wipe out reception for hours. TAT-1 (Transatlantic No. 1) went into service between Oban (Scotland) and Clarenville (Newfoundland) in 1956 with 36 voice channels, immediately triggering the retirement of point-to-point transatlantic longwave and shortwave radio trunks. By 1983, TAT-7, the final coaxial transatlantic telephone cable, entered service between New Jersey and England with 4,200 voice channels before the transition to optical fiber.
What made coaxial cable such a leap over old gutta-percha telegraph wire was polyethylene insulation (developed for wartime radar) paired with a concentric outer copper return shield. Instead of leaking energy into the seawater, the outer copper tube trapped high-frequency radio waves inside the cable. Using Frequency-Division Multiplexing (FDM), engineers stacked dozens—and eventually thousands—of 4 kHz voice channels side-by-side at different frequencies on a single copper core, along with wideband transoceanic television feeds.
Undersea Vacuum Tubes & High-Voltage DC
Because high-frequency analog signals fade quickly in copper, TAT-1 required two separate cables (one for each direction) with 51 flexible, torpedo-shaped vacuum-tube repeaters spliced directly into each line every 70 kilometers—thin enough to pass smoothly around a cable ship's drum without stopping. Each repeater held three hand-assembled Bell Labs 175HQ vacuum tubes powered by ~4,000 volts of direct current sent down the same central copper conductor carrying the phone calls, engineered to run for 20 years on the seabed without a single filament burning out.
Even normal aspects of cables like having in-line amplifiers become far more difficult when you cannot drive a truck out to fix something. You can't drive a truck to the bottom of the ocean. Well, you can, but you won't be much use once you get there and good luck doing it more than once.
Spanning the Pacific & Pushing Copper to the Limit
By the 1960s, single-cable bidirectional systems with rigid, solid-state transistor repeaters expanded across the globe: CANTAT-1 (1961) and COMPAC (1963) linked Britain, Canada, Fiji, New Zealand, and Australia, while TPC-1 (1964) carried the first coaxial voice calls across the Pacific from Hawaii, Midway, Wake, and Guam to Japan—just in time for the 1964 Tokyo Olympics. To keep up with surging international phone demand, late-generation coaxial cables used thicker dielectric cores, solid-state repeaters every 9.5 kilometers, and speech-interpolation multiplexing (TASI) that dynamically squeezed callers into the brief silences of other people's conversations.
Satellites (1960s to Present)
Putting Relays in Orbit
In the 1960s, telecommunications companies also started launching radio relays on satellites. If you put a satellite into geostationary orbit 35,786 km above the equator, it orbits at the exact same speed that the Earth spins, so it stays over one spot in the sky and can reach about a third of the globe. I actually tried adding satellites to the main map at first, but between the orbit rings and the ground footprints, it cluttered up the globe too much and made it hard to click on cities. If you want to see live satellite constellations in 3D, check out satellitemap.space or Stuff in Space.
In July 1962, Bell Labs' Telstar 1 relayed the first live transatlantic television pictures between giant horn antennas in Andover, Maine and Pleumeur-Bodou, France. Because Telstar orbited in an elliptical Medium Earth Orbit (952 × 5,632 km), ground antennas could only track it for 20 minutes per pass before it dipped below the horizon. Arthur C. Clarke had proposed the fix back in 1945: park a satellite 35,786 kilometers directly above the equator, where its 24-hour orbital period matches Earth's rotation so it hangs motionless in the sky. After Hughes Aircraft's Syncom 2 and Syncom 3 proved geostationary orbit in 1963–1964, INTELSAT I ("Early Bird") entered commercial service in 1965 with 240 voice circuits—nearly doubling transatlantic telephone capacity overnight.
By July 1969, INTELSAT stationed three INTELSAT III satellites over the Atlantic, Pacific, and Indian Oceans, completing the first global satellite village just days before 650 million people watched Apollo 11 land on the Moon live. Meanwhile, because equatorial geostationary satellites sit too low on the horizon above 70° latitude, the Soviet Union launched its Molniya / Orbita constellation (1965) into steep 63.4° highly elliptical orbits that lingered high over Siberia for eight hours out of every 12-hour pass. Throughout the 1970s and early 1980s, geostationary satellites carried over half of all transoceanic phone calls and virtually 100% of live international television.
SATNET, Early Transatlantic TCP/IP & the 550 ms Speed-of-Light Tax
Satellites also carried the very first intercontinental Internet connections. Starting in 1975, DARPA's Atlantic Packet Satellite Network (SATNET) shared a 64 kbps SPADE transponder channel on an INTELSAT IV spacecraft to link ARPANET in the United States (via Etam, West Virginia) with packet networks in the UK (Goonhilly Downs), Norway/Sweden (Tanum), West Germany (Raisting), and Italy (Fucino). In November 1977, Vint Cerf and Bob Kahn's famous three-network TCP/IP testbed routed packets from a packet radio van in San Francisco across ARPANET and up through SATNET in geostationary orbit to University College London.
What eventually cost geostationary satellites the core voice and Internet trunk market wasn't just bandwidth—it was the speed of light. Traveling 35,786 km up to the Clarke Belt and 35,786 km back down adds ~240 milliseconds of one-way propagation delay (~480 to 550 ms round-trip, or a full second if a call hopped across two satellites). On phone calls, people constantly talked over each other; on TCP/IP connections, waiting half a second for every SYN/ACK handshake and congestion-window acknowledgment throttled throughput. When TAT-8 (1988) brought laser-driven submarine fiber across the Atlantic with ~65 ms round-trip latency and massive capacity, point-to-point voice and Internet backbones rapidly migrated back to the seabed.
LEO Mega-Constellations vs. Subsea Glass
To escape geostationary latency, newer constellations moved much closer to Earth. O3b / O3b mPOWER (2013) dropped down to Medium Earth Orbit at 8,062 km (~130 ms RTT) to deliver multi-gigabit IP trunks to Pacific islands and cruise ships, while Low Earth Orbit (LEO) constellations like Iridium (780 km), OneWeb (1,200 km), and SpaceX's Starlink (550 km) orbit 65 times closer than the Clarke Belt—cutting round-trip latency down to 25 to 40 milliseconds. Modern LEO satellites even use optical inter-satellite lasers to route packets directly through the vacuum of space (where light travels 47% faster than inside silica glass).
Even so, over 99% of intercontinental data still travels through submarine fiber-optic cables. Satellites are unbeatable for ships, aircraft, disaster zones, and remote communities where you can't trench fiber, but shared radio spectrum in orbit cannot match the raw density of glass: a single modern subsea cable like Marea or Grace Hopper pushes 200 to 350+ terabits per second through a pipe the size of a garden hose, and almost every satellite connection still lands at a gateway Earth station that plugs right back into the terrestrial fiber grid.
Photonic Glass (1988 to Present)
Submarine Fiber Cables
In 1988, TAT-8 replaced coaxial copper with laser-driven single-mode optical fiber across the Atlantic, carrying 40,000 simultaneous phone circuits on two pairs of glass fibers. Note how few of these cables exist in total. They connect our content digitally and culturally, but there are only a few hundred active submarine cables right now in the world. That truly is an extremely small amount, though many more terrestrial cables connect on land. There's not good public data on terrestrial fiber links for several reasons, but you can be assured that there are significant backbone fiber optic cables buried nearby almost every major highway and rail line in the US. Imagine lines that nearly mimic the road system and you'll have an idea of what that map would look like.
Total Internal Reflection & Optical Amplification
Instead of one copper wire carrying a few words per minute in Morse code, we now have hundreds of undersea fiber-optic cables carrying terabits of data every second. Each glass fiber has a high-index silica core (n₁ = 1.48) surrounded by lower-index glass cladding (n₂ = 1.46), so laser light entering above the 80.6° critical angle reflects cleanly off the boundary and stays trapped inside the core. Every 70 kilometers or so before the light fades down to the −20 dBm noise floor, undersea repeaters use erbium-doped fiber amplifiers (EDFAs) and backward-pumped Raman lasers to boost the optical signal directly in the glass without converting it back into electricity.
Island Hopping & ROADMs
Fiber links can also visit islands along the way in order to give them internet access. You don't have to bring the whole main trunk cable up onto the beach for every island. An underwater branching unit on the seafloor splits off a smaller spur line to the island, and a device called a ROADM (reconfigurable optical add-drop multiplexer) picks out specific wavelengths of laser light for local traffic and lets the rest of the wavelengths keep going to the next destination.
Places to Spot on the Map
The Maldives: The Maldives is a great place to see island-hopping fiber in action. You might notice that two domestic cables (Dhiraagu and Ooredoo) zigzag down the atolls and cross each other. They were commissioned about 5 years apart, and the second cable wove through islands that the first round skipped. It must have been a long 5 years if you lived on one of the islands that didn't get high-speed internet on the first pass.
Svalbard: Svalbard is a Norwegian archipelago way up in the Arctic Circle that gets 6 months of daylight followed by 6 months of polar night. It's wild to see two redundant undersea fiber cables running all the way up there from mainland Norway. They serve Longyearbyen and the Global Seed Vault, but the real bandwidth user is SvalSat, a ground station up on a mountain plateau where polar-orbiting satellites downlink their data on every pass over the North Pole.
Africa & the Suez Canal: Africa also has massive subsea cables like 2Africa, Equiano, and WACS tracing the coast all the way around the continent. It is much easier to lay fiber in the ocean than to trench long-haul cables across dozens of land borders. There are major hubs with a ton of bandwidth capacity in internet exchanges in Cape Town, Johannesburg, and Pretoria. Also, zoom in on Egypt: it's crazy to see how many Europe-to-Asia optical cables squeeze through the Suez Canal and the Red Sea, making it one of the biggest chokepoints on the internet.
Network Peering (Present)
Internet Exchange Points
The blue cities on the map mark locations with buildings called Internet Exchange Points (IXPs), while purple hexagons show routed IP address space and red hexagons indicate active outages. These IXP buildings (and sometimes closets) are specialized data centers that have a uniquely large number of fiber connections from surrounding data centers, other internet exchanges, and long-haul cables. These buildings are "meeting points" for Internet service providers (ISPs), cloud providers, content delivery networks (CDNs), and large enterprises like IBM and Apple, and many could say that it is "where" the Internet lives. This is where networks interconnect or "peer."
Many of these Internet exchanges charge a monthly fee to participate in the exchange, but who everyone peers with and the financial details around the arrangement are up to each company. Some companies like Apple or Google will peer with anyone because it brings their users' devices closer to their services and reduces latency and cost while improving reliability. Traffic served through peering, for Google, is traffic that they do not have to pay backbone providers for. These exchanges usually have a "gravity" to them. It's more valuable to be in an exchange where there are already many other networks present. However, we need to remember to avoid putting all of our eggs into a single basket. Exchanges that are too large create a single point of failure, so large internet-connected cities need to have several of these exchanges.
| How they connect | Paid IP Transit | Private Cross-Connect (PNI) | Public IXP Peering |
|---|---|---|---|
| Hardware setup | Port on a Tier-1 carrier router | Direct fiber patch between two racks | One port on a shared IXP switch |
| Who pays whom | Customer pays carrier per Mbps (95th percentile) | No traffic fees; split the cross-connect cost | No traffic fees; pay the IXP a monthly port fee |
| What routes you get | Full global routing table (~1M+ IPv4 prefixes) | Just that one neighbor's ASN prefixes | Whatever prefixes other IXP members announce |
| Typical path | Can trombone through another city/country | Direct patch cable (sub-millisecond) | One Layer-2 switch hop in the same metro |
Tracing the Path to an Exchange
In a traceroute output, you can sometimes infer the location of an internet exchange based on the names of the intermediate routers. Many internet exchanges are named explicitly to indicate their location.
For example, a router with the name apple-1.equinix-am1.nl-ix.net suggests that it is associated with the Equinix Amsterdam (AM1) data center connected through the NL-IX internet exchange. So my request from my laptop in Copenhagen was probably handled inside of Luttenbergweg 4, Amsterdam, Netherlands. However, not all routers or nodes in a traceroute will have such descriptive names. Some routers might be labeled with IP addresses or generic names that do not reveal their location. Because of all of this, we can see the path that is taken when connecting to a server, which I think is really cool. With my example, I can know that it takes around 15.5 milliseconds for my network traffic to reach the Netherlands and I can tell that Apple is peering directly in this facility. You can verify this fact by looking at PeeringDB. Essentially, if you want apple.com to load faster for your users who are near(ish) to Amsterdam, you may want to run some fiber optic cable to this internet exchange. Put another way, the closer you are to the blue cities, the closer you are to the internet.
To see the difference an exchange makes in practice, compare two paths between KPN and TransIP in Amsterdam, one tromboning through London over paid transit, and one peering directly across AMS-IX:
1 100.64.12.1 bng01.ams01.kpn.net [AS1136] 2.1 ms
2 62.115.40.9 adm-b3-link.ip.twelve99.net [AS1299] 2.6 ms ← Handed to Arelion transit
3 62.115.112.24 ldn-bb1-link.ip.twelve99.net [AS1299] 9.4 ms ← Hauls across North Sea to London
4 62.115.140.18 adm-bb2-link.ip.twelve99.net [AS1299] 14.2 ms ← Right back to Amsterdam
5 198.51.100.42 edge01.ams.transip.net [AS20857] 14.8 ms1 100.64.12.1 bng01.ams01.kpn.net [AS1136] 1.1 ms
2 80.249.208.50 ams-ix.transip.net [AS20857] 1.3 ms ← TransIP router on AMS-IX L2 LAN
3 198.51.100.42 edge01.ams.transip.net [AS20857] 1.4 msExplore the Live Global Network
Explore the map, step through the timeline slider from 1858 to the present, or toggle Artistic Mode to see the cable landings against global city lights.