Ask someone how a message gets from Stockholm to Tokyo, and most will describe something vertical — up to a satellite, across, back down. It is a reasonable guess. It is also almost entirely wrong. The message goes sideways, through glass, along the seabed, in a tube about as thick as a garden hose.
Satellites carry a rounding error. The overwhelming majority of intercontinental traffic — more than 99% of it, by the usual industry estimate — travels through submarine cables. Everything people describe as living in the cloud is, at some point in its journey, lying in the dark at the bottom of an ocean.
The Scale of the Thing
TeleGeography’s 2026 cable map tracks 694 cable systems connected by 1,893 landing stations, with over 1.5 million kilometers of cable in service. That is enough to circle the equator around thirty-seven times, laid by a specialized industry most people have never heard of, for an amount of money that would be unremarkable on a national infrastructure budget.
The cables themselves are underwhelming to look at. The fibers inside are roughly the diameter of a human hair. In deep water, the surrounding cable is thin and unarmoured, because at four kilometers down there is nothing to protect it from. Near shore it thickens considerably — steel wire, extra sheathing, and burial under the seabed — because that is where the danger is.
The landing stations are the part that most rewards a visit. They are ordinary low buildings near ordinary beaches, usually with a small unexplained fence and no signage. There is no moment on the journey where the internet looks like the internet.
What Actually Breaks Them
Somewhere between 150 and 200 cable faults happen every year. The International Cable Protection Committee attributes 70–80% of them to accidental human activity — overwhelmingly commercial fishing gear and ships’ anchors. Dragged anchors alone account for roughly 30% of incidents, around sixty faults a year.
That number has stayed roughly flat for over a decade while total cable mileage grew enormously, which means the fault rate per kilometer has fallen substantially. Better route surveys, deeper burial, better armoring, better charts in the hands of fishing fleets.
The sharks did not do it. In 2014, a Google engineer mentioned in passing that the company wrapped some of its trans-Pacific cables in a Kevlar-like material, an old video of a shark mouthing a cable resurfaced, and a durable myth was born. The reality is that fish bites were a genuine nuisance for early deep-water fiber systems in the 1980s and are now statistically invisible: ICPC data cited by TeleGeography records zero cable faults from fish bites between 2007 and 2014. Bites also leave evidence — teeth in the sheathing — so they are unlikely to be hiding inside the “cause unknown” category. The threat to the internet is a trawler, not a predator.
Why You Never Notice
Because the industry buys redundancy on purpose. Networks spread capacity across multiple cables, so a single break reroutes rather than disconnects. Most of those 200 annual faults never become news because nobody experiences them.
The exception is chokepoints, where redundancy stops being real. The Red Sea and the Strait of Bab el-Mandeb are the only practical wet route between Europe and Asia, and more than a dozen systems thread through it. In February 2024, the anchor of the sinking cargo ship Rubymar severed three cables in a single pass. In September 2025, further cuts near Jeddah slowed connectivity across South Asia and the Gulf. Diversity on a map is not diversity if every line converges on the same eighteen-mile gap.
The Baltic has had its own run of incidents, and here the reporting deserves skepticism. Attribution is genuinely hard, the base rate of accidental anchor damage is high, and Finland’s own security service has publicly attributed several cases to badly maintained ships and undertrained crews rather than orchestrated sabotage. Deliberate interference is a real category. It is not yet the usual explanation, and treating every fault as an attack makes the actual attacks harder to spot.
The Bit That Should Worry You
Repair depends on a global fleet of around sixty specialized vessels, and fewer than twenty are dedicated purely to repair work. Many are decades old. Fixing a break means locating the fault, grappling the cable off the seabed, splicing in a fresh section, testing it and laying it back — a job that requires a ship to hold position for days, sometimes in politically awkward or actively dangerous water.
Repairs also need permits. The longest single repair recorded in a recent ICPC dataset took 947 days, most of it spent not on engineering but on paperwork. When five of Vietnam’s cables were damaged in early 2023, the country lost the majority of its international capacity and full repairs were not completed until late that November — because the ships were busy elsewhere.
Why It Still Matters
The useful correction is not “the internet is fragile.” It is that the internet is an estate: a physical inventory of cable, buildings and ships, with an aging maintenance fleet, a queue, and a small number of geographic pinch points that no amount of network engineering can route around.
Everything built on top of it — cross-border payments, cloud regions, the entire architecture of remote work — is priced as though connectivity were an ambient property of the world. It is a service, delivered by about sixty boats.