When a page loads in Dubai and the server sits in Frankfurt, the request almost certainly does not travel through a satellite. It travels through a cable on the ocean floor, roughly the diameter of a garden hose, carrying the overwhelming majority of the world’s internet traffic between continents. Estimates from cable industry groups put that share above 95 percent. The satellite constellations get the headlines, but the physical internet is still mostly wet.
The system is bigger, stranger and more fragile than most people assume. Here is how it actually works.
One cable, a few pairs of glass, insane capacity
A modern transatlantic cable looks unimpressive in cross-section. Inside the polyethylene and steel armoring sit a handful of fiber pairs, often just 8 to 16 fibers total. Each pair carries light in dozens of wavelength channels, and each channel moves 100 to 400 Gbps depending on the cable’s generation. Multiply it out and flagship systems commissioned in the last few years talk about capacities in the hundreds of terabits per second across the whole cable. A single fiber pair in the newest cables can carry more data per second than the entire global internet carried in a month in the mid-1990s.
The light does not get amplified along the way for free. Every 60 to 100 kilometers, the cable passes through a repeater, a housing packed with erbium-doped fiber amplifiers that boost the signal. These repeaters have no brains and no software. They are powered by a constant electrical current sent down the copper conductor inside the cable itself, around 10,000 volts DC fed from both landing stations.
Cable landing stations: where the internet touches land
Every cable ends in a landing station, usually an anonymous building near a beach with no signage. The station houses the power feed equipment, the optical line terminals, and the staff who watch the system. From there, capacity is handed to carriers inside carrier hotels, the same kind of facility our piece on how LLM API traffic gets routed deals with at the software layer.
Landing rights are political. A cable connecting two countries needs permits in both, and routes are chosen as much for diplomacy as for distance. That is why so many Europe-Asia cables hug the same corridors through Egypt and the Red Sea, and why a single anchor drag there can degrade latency from London to Mumbai within minutes.
What actually breaks cables
The failure statistics are not secret; the International Cable Protection Committee publishes them. The leading causes, year after year:
- Fishing gear, especially bottom trawlers dragging equipment across the shelf. Around a third to two fifths of faults.
- Anchors, both dragged and dropped, from vessels anchoring where they should not.
- Natural events: earthquakes, undersea landslides, abrasion on rock. The 2006 Hengchun earthquake off Taiwan cut several cables at once and took weeks to fully repair.
- Theft and sabotage, rarer but real, including documented cases of cables being cut for the scrap metal value of their armoring.
Repairs are done by a small global fleet of cable ships, fewer than 50 of them in service. A ship lifts the damaged section, splices in new cable on deck, and lays it back down. Splicing fiber on a pitching deck is exactly as delicate as it sounds, and a single repair typically costs one to three million dollars and takes days to weeks depending on permits, weather and where the fault sits.
Why your Netflix still works when a cable dies
Because the internet was designed by people who assumed things would break. Traffic reroutes at the IP layer within seconds. Latency suffers, since a broken Egypt route pushes Europe-Asia traffic around the Cape of Good Sea or across the Pacific, adding tens of milliseconds, but packets arrive. Content providers also cache aggressively: most of what you stream comes from a server inside your own country or your own ISP’s network, which is why a severed Atlantic cable can be headline news while your evening viewing continues untouched.
Capacity planning is where the redundancy math gets done. Major routes have multiple cables with diverse landings, and carriers buy capacity across several of them precisely so no single cut is fatal. It is the same reasoning behind running multiple providers at the software layer, something we looked at when dissecting the infrastructure behind live sports betting, where a stalled feed means voided bets and angry customers.
The geopolitics nobody can ignore anymore
Undersea cables spent decades as background infrastructure. They are now treated as strategic assets. Governments block cable landings over espionage concerns, as happened with proposed China-linked routes to the US. The Baltic cable cuts of recent years, including deliberate damage to data and power links, pushed NATO to treat cable protection as a naval mission. And satellite backhaul is finally real competition for the first time, though laser-linked constellations still carry a rounding error of global capacity compared with the wet fiber backbone.
The takeaway for anyone who runs anything online: your “cloud” application is physically anchored to a set of tubes on the seabed owned by consortia of telecoms and, increasingly, big tech companies funding their own private cables. Google, Meta, Amazon and Microsoft now bankroll a large share of new transoceanic builds. The map of who owns the cables is slowly becoming a map of who owns the internet’s carrying capacity.
Next time a page loads fast from another continent, spare a thought for a hose of glass under four kilometers of cold water, pumped with ten thousand volts, getting dragged past by a trawler somewhere. It holds, almost every day, and the days it does not, you probably never notice.
