Almost all international internet traffic moves through physical cables lying on the seabed rather than through satellites. The system is old in concept, enormous in capacity and surprisingly small in the number of people who maintain it.
Light carries the data
Each cable contains a small number of glass fiber pairs, and information travels as pulses of light reflected along the fiber core across thousands of miles.
Multiple wavelengths are transmitted simultaneously through the same fiber, so a single strand carries many independent channels at once.
Capacity has grown largely through improvements in the equipment at each end rather than through new glass, which is why an existing cable can be upgraded without being replaced.
Repeaters keep the signal alive
Light weakens as it travels, so optical amplifiers are spliced into the cable at regular intervals along the route.
These repeaters are powered by electrical current fed down the cable from shore stations, which is why a cable carries a conductor as well as fiber.
They must operate unattended for decades under pressure at depth, so reliability requirements exceed almost anything in consumer electronics.
Armoring is matched to the hazard
In deep water a cable is roughly the diameter of a garden hose and is laid directly on the seabed with minimal protection.
Near shore, where fishing gear and anchors are the main threat, the cable is heavily armored and often buried by a plow towed behind the laying ship.
Most faults occur in shallow water and are caused by human activity rather than by natural events, which is why armoring is concentrated there.
Repair depends on a small specialized fleet
When a fault occurs, operators locate it by measuring the signal reflected back from the break, which narrows the position to within a short stretch of route.
A repair ship then grapples the cable from the seabed, raises both ends, splices in a replacement section and returns it to the bottom.
There are relatively few such vessels worldwide, held under shared maintenance agreements, so repair timelines depend on where the nearest ship happens to be.
Redundancy hides most failures
Traffic is routed across multiple cables, so a single break usually causes higher latency and congestion rather than an outage most users notice.
Regions served by only one or two cables are the exception, and a fault there can disconnect an entire country until repair is complete.
This is why cable landing points and route diversity are treated as infrastructure planning questions rather than as purely commercial decisions by the operators.