Intercontinental traffic does not travel by satellite. Very nearly all of it runs through cables on the seabed, and the total number is measured in hundreds rather than thousands. The map of the internet, drawn honestly, is a few hundred lines between a smaller number of landing points, and that concentration explains most of what happens when one fails.
The object itself
In deep water a submarine cable is around seventeen to twenty one millimetres across, roughly the thickness of a garden hose. The fibers sit in a small tube filled with a water blocking compound, wrapped in steel strands for tensile strength, because the cable must survive being paid out from a ship and later hauled back from several kilometres down. Around that is a copper conductor, then insulation and an outer sheath.
Close to shore the construction changes completely, gaining one or more layers of heavy steel armour until the diameter approaches that of a human arm. The deep ocean is a benign environment. The continental shelf, where fishing gear drags and ships anchor, is not, and the engineering follows that risk gradient exactly.
Power, delivered from shore
The copper conductor is not incidental. Amplifiers along the route need power and the only place to source it is the beach. A landing station feeds direct current onto the conductor at up to roughly fifteen kilovolts using the sea as the return path, with the far end fed at opposite polarity so the voltage burden is shared.
The feed is constant current rather than constant voltage, typically around one ampere. Every amplifier sits in series on the same conductor and draws power from that current as it passes, so a constant current supply keeps every device fed regardless of how many are in the chain.
It also means a break in the conductor stops the entire system rather than degrading it. The cable is one long series circuit spanning an ocean.
Amplifiers, not repeaters
Every fifty to a hundred kilometres the cable passes through a housing containing erbium doped amplifiers, one per fiber pair, pumped by laser diodes drawing on the line current. These are amplifiers in the strict sense. They boost the optical signal without detecting it, decoding it, or knowing anything about the data.
The consequence is significant. The seabed equipment is indifferent to modulation format and bit rate, so a system installed years ago can be upgraded to far higher capacity by replacing only the terminal equipment ashore, and several cables in service carry many times the capacity they were sold with without a ship returning to the route.
Noise accumulates rather than being cleaned up, since each amplifier adds spontaneous emission along with gain, and the accumulated noise across dozens of spans is a fundamental limit. Modern coherent receivers exist largely to extract signal from exactly that impairment.
Why the fiber count keeps rising
Capacity was historically increased by pushing more bits through a few fiber pairs using better modulation and more wavelengths. That approach hit a hard limit, because a fiber carrying high optical power becomes nonlinear and beyond an optimum launch power the additional power degrades the signal.
The response redefined the constrained resource. The real constraint on a submarine system is electrical power delivered from shore, not the capacity of any single fiber. Given a fixed power budget, total capacity is higher spread across many pairs at modest power each than concentrated into a few driven hard, because each fiber then sits in a more efficient region of the power against capacity curve.
Systems that once carried four to eight pairs are now built with sixteen, twenty four, or more, and aggregate capacity on modern trans oceanic cables reaches hundreds of terabits per second.
What actually breaks them
Faults are common, something in the region of a hundred to two hundred worldwide each year, and the overwhelming majority happen in shallow water near shore. The dominant causes are fishing gear dragged across the seabed and ships anchoring where cables lie. Natural causes are a much smaller share, with submarine landslides triggered by earthquakes being the significant one, capable of cutting several cables at once where routes run close together.
This is why cables are ploughed into the seabed to a depth of one to three metres wherever the water is shallow enough for trawling. Burial does not make a cable invulnerable, it makes it expensive enough to reach that ordinary fishing gear passes over it.
Repair is a slow physical operation. A ship must be available and must reach the fault, which alone can take days. The cable is grappled from the seabed, cut, and one end buoyed off while the other is recovered, tested, and spliced to a new section, then the buoyed end is recovered and joined and the repaired section laid back with enough slack for the depth. Weeks is normal, and permits, weather, and ship availability routinely extend it.
Geography decides the failure modes
Cables do not spread evenly across the ocean. They converge on a small number of corridors defined by geography, landing rights, and cost, and those corridors are the real fragility. The Red Sea and the approaches to Suez carry a large share of Europe to Asia traffic through a narrow channel.
The Luzon Strait concentrates East Asian capacity in a seismically active region. The approaches to Singapore form another dense convergence.
In each of these a single event can affect multiple cables, because they are close enough to share a hazard, and protection within a cable does not help when the whole corridor is affected. The only effective answer is genuine route diversity on paths that share no corridor, which is expensive precisely because the cheap paths are the ones everybody already uses.
Latency is route length
Latency follows the same geography. Light in silica travels at roughly two thirds of its vacuum speed, close to five microseconds per kilometre, so the round trip between two continents is set almost entirely by how much cable lies between them rather than by the equipment at either end.
Cable routes are longer than the great circle distance because they avoid hazards, follow permitted corridors, and land where landing stations exist, which is why shorter physical routes command a premium and why polar routes are periodically proposed.
Who owns them now
Ownership has shifted alongside this. Submarine cables were historically built by consortia of carriers, each taking capacity in proportion to its investment. Over the past decade the large content and cloud operators have become dominant investors, financing cables outright or taking major stakes, because they move enormous volumes between their own data centres and owning the path is cheaper than leasing it.
A growing share of intercontinental capacity is therefore built to serve the internal requirements of a handful of companies, which is a structural change in who decides where new capacity goes.
Note: a trans oceanic cable is a single series electrical circuit thousands of kilometres long, carrying amplifiers expected to run for a quarter of a century without maintenance, on a seabed nobody inspects. The remarkable thing is not that they fail, it is how rarely.