It sounds like an exaggeration, but it is true: when you send a message to someone across the ocean, it most likely travels thousands of kilometres along the seabed. Communications satellites account for only a few percentage points of intercontinental traffic. The rest — more than 99% — travels through fibre-optic cables that cross the oceans at depths of up to 8,000 metres. TeleGeography, the industry’s reference research firm, counts about 1.5 million kilometres of submarine cable deployed: enough to circle the planet almost 40 times.
The physics that makes it possible
The foundation is the same fibre that reaches your home, taken to the extreme. A modern submarine cable contains several pairs of single-mode fibres: strands of ultra-pure silica (silicon dioxide) 125 micrometres in diameter — the thickness of a hair — through whose core, about 9 micrometres wide, an infrared laser beam travels.
Light propagates by total internal reflection: passing from a denser medium (the core) to a less dense one (the cladding), the angle of incidence makes the beam bounce and never escape the strand. The catch is that silica is not perfect: it absorbs and scatters part of the light, a loss that engineers measure in decibels per kilometre. In the 1,550-nanometre window, the most widely used, attenuation sits around 0.2 dB/km: after 100 kilometres, the signal keeps barely 1% of its power. And the Atlantic is more than 5,000 wide.
That is why there is a repeater every 50-80 kilometres. The key component is the erbium-doped fibre amplifier (EDFA): a section of fibre “contaminated” with erbium ions that, excited by a pump laser (980 or 1,480 nm), amplify passing signals without converting them into electricity. It is direct optical amplification, with no data regeneration, and it works for dozens of channels at once.
Many lights, one cable
A single laser beam is not enough. The technique that multiplies capacity is wavelength-division multiplexing (WDM): each fibre pair carries dozens of slightly different wavelengths at the same time, spaced 50 GHz apart in the C-band. And each channel does not encode a simple 0 or 1: modern coherent transmitters modulate both the phase and amplitude of the light (QPSK, 16QAM…), while a digital signal processor (DSP) at the receiver undoes the distortions of the journey. The result: 100, 200 or even 400 Gbps per wavelength, and cable capacities above 400 Tbps. Amitié, which links the US with the UK and France, announced 400 Tbps; 2Africa, 45,000 km around the continent, about 180.
Errors, moreover, are corrected in flight with forward error correction (FEC): the transmitter adds mathematical redundancy and the receiver uses it to rebuild damaged bits. Over 5,000 kilometres, that is not optional.
The anatomy of the cable
A modern cable is not a simple thread. From outside in: a polyethylene jacket, a steel-wire armour (essential in shallow waters, where anchors and trawlers can damage it), a copper or aluminium conductor layer and, at the centre, the fibre pairs wrapped in protective gel. The conductor is not decorative: the repeaters need electricity, and they are fed from shore with high-voltage direct current (on the order of 10,000 V) injected from the landing stations at each end. That is why a cable weighs several tonnes per kilometre.
Deployment is a naval operation: a cable-laying ship follows a pre-surveyed route, buries the cable with a plough at depths down to about 1,500-2,000 metres and lays it on the seabed in deep waters, splicing the repeaters on board with fusions that leave losses of hundredths of a decibel.
Speed has a physical limit
Latency is not set by bandwidth but by the speed of light in fibre: about 200,000 km/s (silica’s refractive index, ~1.47, slows it by a third compared with vacuum). A ping between New York and London is around 65 ms round trip; between Madrid and New York, close to 90. That is the physical floor: no algorithm, cache or protocol will make it faster — only a shorter route. Hence high-frequency trading firms pay a fortune for the straightest cables in the North Atlantic.
Faults and espionage
More than a hundred faults occur every year: anchors, fishing nets, earthquakes. The response is a fleet of repair ships that locates the cut (the cable ends are detected electrically), hauls up the end, splices a new section and sinks it again. That is why major routes are designed with redundancy: if one cable fails, traffic jumps to another within milliseconds.
And security? Tapping a cable at sea is technically possible but detectable: splicing a tap introduces losses and bit errors that monitoring systems notice immediately. The weak point is not the seabed but the landing stations, where traffic arrives concentrated — that is where, for example, the Tempora surveillance programme of Britain’s GCHQ and the US NSA focused. The answer is the same as on land: end-to-end encryption.
The AI rush
The explosion of artificial intelligence has pushed transatlantic capacity demand through the roof, and the big tech firms (Microsoft, Google, Meta, Amazon) are funding dozens of new cables due to enter service between 2025 and 2027. AI data centres, hungry for cross-continental bandwidth, have turned submarine cables into first-class critical infrastructure. The same technology that has been joining continents since 1866 — when a telegraph cable first crossed the Atlantic — is still the invisible skeleton of the internet.
