International internet traffic runs overwhelmingly through cables lying on the seabed. Satellites carry a small fraction, and the reasons are physical rather than commercial preference.
Capacity is not comparable
A single submarine cable contains multiple fibre pairs, each carrying enormous quantities of data through different wavelengths of light transmitted simultaneously.
Radio links, including satellites, share a limited allocation of spectrum among all users in a coverage area, and that allocation cannot be expanded by adding hardware.
The gap is large enough that satellite capacity is best understood as serving places cables do not reach, rather than as competing for the same traffic.
Distance creates unavoidable delay
A signal to a geostationary satellite travels a very long way up and back, and the round trip introduces a delay noticeable in conversation and disruptive to many protocols.
Low-orbit constellations reduce that distance substantially and therefore the delay, which is why they are usable for interactive applications where earlier systems were not.
Cables still win on the point-to-point route, since light in fibre follows close to the shortest path between landing stations without leaving the planet.
Landing points concentrate risk
Cables come ashore at a limited number of stations, and those locations are determined by geography, permits and existing terrestrial networks.
Concentration means a small number of sites carry a large share of a region's connectivity, and damage near shore is more likely than damage in deep water.
Most faults are caused by fishing gear and anchors in shallow water, which is why protected zones and burial depth requirements exist along approaches.
Repair is a scheduled maritime operation
Fixing a break requires a specialised cable ship to locate the fault, raise the cable, splice in a replacement section and lay it back down.
A limited fleet of such ships exists worldwide, and one may be far from the fault or already committed, which is why repairs take weeks rather than days.
Networks handle this through diversity, routing traffic across alternative cables automatically. Regions served by few cables have no such option and experience visible degradation.
Ownership has shifted
Cables were historically built by consortia of telecommunications operators sharing cost and capacity across many partners.
Large content and cloud companies now fund cables directly, because their own traffic justifies the investment and control over routing serves their service requirements.
That changes where cables are built, since routes follow data centre locations and customer concentrations rather than the traditional pattern of national carrier relationships.