The Garden Hose Holding the World Together: Inside the Submarine Cables That Carry Almost All of the Internet

A Hair-Thin Thread Beneath the Abyss

Somewhere off the coast of Virginia Beach, at a depth of more than five thousand metres, a bundle of glass strands no thicker than a garden hose lies on the cold abyssal plain, carrying roughly 160 terabits of data every second across the Atlantic Ocean. It does not hum. It does not glow. It does not move. And yet, when you load a foreign website in under a hundred milliseconds, this is what you are depending on.

The cable is called MAREA. Microsoft, Facebook and Telxius finished laying it between Virginia Beach and Bilbao, Spain, in 2017, and it remains one of the more instructive objects in the modern world — not because of what it does, but because of how little of it is actually doing the work. The glass inside is carrying the traffic. Everything else is armour.

This distinction matters, because the scale of what the glass achieves is difficult to hold in the mind. The light-carrying core of a single optical fibre measures approximately nine micrometres across — roughly the width of one red blood cell. The pure silica cladding that wraps it brings the strand to 125 micrometres, about the width of a human hair. Eight pairs of these strands, threaded through a nested cylinder of steel, copper and polyethylene, move enough data to stream 71 million high-definition videos simultaneously. The whole assembly weighs 4.65 million kilograms and stretches across more than thirteen thousand kilometres of ocean floor.

That is the intercontinental internet. Not a cloud. A hose.

What the Hose Is Actually Made Of

To understand why submarine cables are engineered the way they are, it helps to read their cross-section as an argument — each layer a response to a specific threat, arranged from the most vulnerable element outward to the most exposed.

At the centre sits the fibre bundle itself, typically eight to thirty-two strands arranged in pairs, each direction of traffic assigned its own strand. Around the bundle, a steel or copper tube packed with water-blocking gel provides the first line of defence against the ocean. Around that, a copper conductor carries several thousand volts of direct current — the power line that feeds the amplifiers strung along the cable’s length. Then comes a polycarbonate insulator, then stranded galvanised steel wires for tensile strength, then an outer sheath of polyethylene.

In deep water, this entire structure measures perhaps seventeen to twenty millimetres across. It is, genuinely, no fatter than a magic marker.

Nearer the coast, the calculus changes. In water shallow enough for trawl doors and anchor flukes, one or two extra layers of heavy steel armour wire are added, and the shore end of a cable can reach the diameter of a soft drink can. The cable dresses for the danger it is actually in, and that danger is almost entirely human in origin. The deep ocean, counter-intuitively, is a gentle place: no trawlers, no anchors, negligible current, very little temperature variation. The cable is not fighting anything down there. It is simply lying still.

Why Glass at 1,550 Nanometres Changed Everything

The physics underlying all of this rests on a fortunate accident of chemistry. When a laser fires a pulse into one end of a silica fibre, the light travels down the core by total internal reflection — striking the boundary with the slightly less dense cladding at a shallow enough angle that it bounces rather than escapes. The light is not so much guided as trapped.

What makes this worth doing across an ocean is the transparency of the glass at a specific wavelength. Near 1,550 nanometres, in the near-infrared, modern silica absorbs so little light that a signal can run tens of kilometres before it fades to the point of needing amplification. A block of it kilometres thick would still be, in practical terms, transparent.

That window at 1,550 nanometres is not an aesthetic or arbitrary choice. It is the wavelength at which silica happens to absorb least. The entire long-haul telecommunications industry — every undersea cable, every terrestrial trunk line — is built on top of that one accident of molecular structure.

Multiplexing: How One Strand Carries Dozens of Signals at Once

A single laser, modulated as fast as electronics allow, yields tens of gigabits per second. Getting to tens of terabits requires sending many signals down the same strand simultaneously without allowing them to interfere — and this is where wavelength-division multiplexing becomes the central technique of modern submarine communications.

The principle is straightforward even if the engineering is not: each signal is assigned its own colour of infrared light, dozens or even a hundred of them sharing one fibre in the way radio stations share the electromagnetic spectrum. Each of those colours is then modulated coherently, with information encoded not just in the amplitude of the wave but in its phase, and carried simultaneously on two perpendicular polarisations. A receiver capable of tracking phase can extract four or more bits from every symbol rather than one.

On MAREA, this arithmetic produced approximately twenty terabits per second per fibre pair, and 160 terabits across all eight pairs at commissioning. That figure has since grown without anyone touching the glass. Because the fibres are fixed but the terminal equipment on the beach is not, operators upgrade cables by swapping hardware at the landing stations rather than relaying steel across the ocean floor. TeleGeography now lists MAREA’s potential capacity at 224 terabits per second — forty per cent above its original specification, on exactly the same strands.

Newer systems have pursued scale through a different strategy. Google’s Grace Hopper cable, which went live between New York, Bude in Cornwall and Bilbao in September 2022, carries sixteen fibre pairs for a design capacity of approximately 352 terabits per second. Meta’s 2Africa cable, whose core was completed in November 2025, applies the same sixteen-pair spatial-division approach across 45,000 kilometres and 33 countries. Running more moderate lasers down more fibres, the engineering consensus has concluded, outperforms running heroic lasers down fewer — largely because the power budget available at the bottom of the sea is finite, and pushing a single fibre harder eventually runs into the limits of what the glass can absorb before the signal degrades.

The Repeaters and the Power Line That Sustains Them

However transparent the glass, light attenuates. Every sixty to a hundred kilometres, the cable passes through a repeater — a pressure-rated cylinder roughly a metre long, shaped like a small torpedo, rated to survive indefinitely at abyssal depths.

Inside each repeater is an erbium-doped fibre amplifier: a short coil of fibre seeded with erbium ions which, when excited by a pump laser, release their energy into the passing signal. Nothing is converted to electricity. Nothing is decoded or digitally regenerated. The photons simply get louder and continue, which is why a repeater built in 1998 can still amplify a modulation scheme invented in 2015. The amplifier does not need to understand the signal to strengthen it.

Feeding those repeaters is its own quiet feat of engineering. The copper conductor wrapped around the fibre bundle carries direct current at up to approximately 10,000 volts, pushed from power-feed equipment housed in buildings at both landing points, with seawater itself completing the circuit through sea-earth electrodes. A trans-Pacific system may have well over a hundred repeaters hanging off a single continuous power line thousands of kilometres long. The whole chain must survive twenty-five years without a service call.

Laying it is slow, careful work. Specialised cable ships — such as those operated by Alcatel Submarine Networks — pay cable out over the stern at perhaps 100 to 200 kilometres a day, ploughing it into the seabed where the water is shallow and letting it settle where it is not. A single ocean crossing can occupy a ship for months.

Where the Cables Break, and Why

Roughly two hundred cable faults occur worldwide each year. Data from the International Cable Protection Committee attributes approximately two-thirds of them to fishing gear and dragged anchors. Earthquakes, turbidity currents and component failure account for most of the remainder.

Sharks, despite their recurring role in popular accounts of cable vulnerability, are a statistical non-event. In the ICPC’s records, fish bites of any kind caused zero cable faults between 2007 and 2014. The animal most dangerous to the global internet is a person on a fishing boat.

Repair follows a procedure that would be broadly recognisable to a Victorian cable engineer. A ship steams to the fault location, drops a grapnel, hauls the severed ends to the surface one at a time, splices in a fresh section of cable, and lowers the resulting loop back to the seabed. Two weeks is a normal turnaround; permits and weather routinely extend that considerably.

Redundancy is what makes any single break survivable, and its absence is what makes a break catastrophic. When the Hunga Tonga-Hunga Ha’apai volcanic eruption severed Tonga’s only international cable on 15 January 2022, there was no alternative routing. Cloudflare’s traffic data shows the country effectively offline for 38 days, with a trickle of satellite bandwidth standing in for a nation’s connectivity, until the repair ship Reliance replaced a 92-kilometre section. Countries with a dozen cable landings never notice a cut. Countries with one notice nothing else.

The same web of dependencies explains why a single mistyped command in the wrong network console can take out more of the internet than any fishing trawler ever has. Physical redundancy and institutional competence are not separate problems.

Why Satellites Have Not Won, and Probably Will Not

Every generation asks why the world does not simply put the internet in orbit, and the answer keeps returning to the same two constraints: capacity and latency, in that order.

A geostationary satellite sits approximately 36,000 kilometres above the Earth. Light requires roughly half a second to complete the round trip before any equipment processes the signal, and the throughput available is measured in gigabits rather than terabits. Low-Earth-orbit constellations such as Starlink fly a few hundred kilometres up and reduce the delay to tens of milliseconds — genuinely transformative for a remote village with no alternative, and still nowhere near the bulk capacity of a single fibre pair, let alone sixteen of them. The US Federal Communications Commission puts satellites at 0.37 per cent of American international capacity, a figure TeleGeography cites when explaining why no serious proposal exists to replace the cables.

There is a historical rhythm in this. The first trans-Atlantic telephone cable opened in 1956 and carried a few dozen simultaneous calls. Telstar went up six years later and looked, briefly, like the future. Seventy years on, the cable won — and it won by getting narrower and cleverer rather than larger.

There are more than 600 active and planned submarine systems in service today, representing something over 1.5 million kilometres of cable on the seabed. Some routes are trivially short. Others were monsters: SEA-ME-WE 3 ran roughly 39,000 kilometres from Norden in Germany to Okinawa in Japan by way of the Suez chokepoint, and held the record for the longest cable on Earth until its retirement in December 2024. Long or short, they share the same modest cross-section, the same layered logic, the same fundamental dependence on a nine-micrometre thread of glass.

So the intercontinental internet remains, at bottom, a physical object: pulses of infrared threaded through strands finer than a human hair, sheathed in steel and polymer no thicker than a garden hose, lying in the cold and the dark four kilometres beneath the shipping lanes. It carries very nearly everything humanity says to itself across the oceans. And tonight, as it has every night since it was laid, nobody using it will give it a thought.