Undersea Cables

The first successful transatlantic cable came ashore at Heart's Content, Newfoundland, on 27 July 1866. An earlier one had been landed nearby in 1858 and lasted about three weeks, until an engineer trying to speed up the signal ran too much voltage through it and it stopped working. The 1866 cable was laid by the Great Eastern, the largest ship in the world at the time, which steamed west from Valentia Island in Ireland with the whole Atlantic length coiled in her hold, letting it out over the stern as she went. The crossing took fifteen days. Five more cables followed over the next thirty years, and the station at Heart's Content stayed in service until 1965. Five of those old cables are still there, running out of the shore and into the water. The station and its sister at Valentia are now in line for UNESCO World Heritage status, nominated together as a single site on either side of the Atlantic.

Worldwide, what crosses the ocean floor now is about 500 cable systems and something close to 1.8 million kilometers of cable, carrying nearly all the world's internet traffic. That is enough to wrap the earth more than forty times at the equator. What I hadn't appreciated is that each of those systems runs as one continuous piece. The cable from Halifax to England is a single cable, spliced during manufacture and then laid end to end across the whole Atlantic.

A deep-water cable is a little under two centimeters across, about the thickness of a garden hose. At the centre are as many as two dozen pairs of fibre-optic threads, one fibre in each direction, each about as thick as a hair and sitting in a protective jelly, so that a cut in one place doesn't flood the whole length. Around those is a steel core for strength, then steel wires, then a layer of copper, then a plastic sheath. The copper doesn't carry data. Light fades as it travels, so every 60 to 100 kilometers there is a steel housing on the line called a repeater that boosts the signal back up to strength, and the copper carries the electricity that runs them, fed from the landing station at either end. A single shore station is pushing current through thousands of kilometers of copper to power equipment in the middle of the ocean, and that power budget is the reason a cable carries twenty-four pairs instead of two hundred. A cable is built to sit on the bottom and work for 25 years. The thick armoured cable in photographs is the shore end, wrapped until it's the thickness of a wrist because that's where anchors and trawls are. Out in deep water it doesn't need the armour.

The North Atlantic route has hardly changed since 1866. Cables still cross between Newfoundland or Nova Scotia and Ireland or England, because that's the shortest span between Europe and North America, and the same logic sets the rest of the map. Systems cross the Pacific between Japan and the west coast of North America, most of them landing in the United States, though Google's Topaz cable now runs from Japan into Vancouver by way of Port Alberni. The route from Europe to Asia is stranger. Those cables come ashore on Egypt's Mediterranean coast, cross the country on land beside the highways for a couple of hundred kilometers, then go back into the water in the Red Sea and out through the strait at its southern end.

What has changed since 1866 is what runs through them. The Victorian cables carried telegraph signals, a few words a minute, and the first transatlantic telephone cable didn't go into service until 1956. Fibre arrived in the 1980s, and glass turned out to be so much better at this than copper that the whole industry rebuilt itself around it inside twenty years.

In the wake of that rebuilding, through the late 1990s a group of new companies borrowed enormous amounts of money and laid cable on the assumption that demand would keep growing. A cable costs everything up front and earns nothing until traffic arrives, so the borrowing has to be repaid out of revenue that doesn't exist yet. A Vancouver company called 360networks built a ring under the North Atlantic linking Halifax, Boston, Dublin and Liverpool. It went public in April 2000, weeks after the Nasdaq peaked, in the largest technology share offering in Canadian history to that point. The underwriters misjudged the demand and had to drop the price at the offering. The company opened an $80 million landing station in Halifax later that year. The cable was finished on schedule and switched on, but then the traffic didn't come. In June 2001, owing about $2.7 billion, the company filed for court protection from its creditors. The cable is still there and still working, under different owners. It sold out of bankruptcy court in 2003, with its four landing stations, for $18 million, about two cents on the dollar against what it cost to build. The buyer renamed it Hibernia Atlantic, spent a decade selling capacity to banks and broadcasters, and sold the business in 2016 for roughly $590 million.

Four companies now build most of the world's long-distance cable: Alcatel Submarine Networks in France, which the French government took an 80 percent stake in, SubCom in the United States, NEC in Japan, and HMN Technologies in China. Between them they account for roughly 83 percent of everything laid in the past five years. The ships that lay it usually belong to somebody else again, marine contractors who work for whoever hires them.

Laying a cable takes years, and almost none of that is the laying. A route starts as a study of the seabed and the shipping lanes and the fisheries, then a survey ship runs the corridor and takes sediment cores, and then come the permits. Loading the cable into the ship's tanks, coiled in the exact order it will come out, can take two months on its own. Microsoft's MAREA cable between Virginia and Spain took five months of loading and laying. Out in deep water the ship simply lets the cable out over the stern and it settles on the bottom. Closer to shore it tows a plough that cuts a furrow less than a meter wide and tucks the cable underneath as it goes. Burial stops somewhere around 1,000 to 1,500 meters of water, below which there is nothing much down there to catch a cable. It is a very different way of working the seabed from deep-sea mining, where the whole point is to disturb the bottom instead of crossing it.

Who owns the cables has changed more than how they're built. For most of the twentieth century a cable belonged to a group of national telephone companies who shared the cost of building it and took a share of the capacity in return. That still happens, but Google now has interests in something like 36 cable systems and Meta in around 20, and the large technology companies together account for roughly 71 percent of the international capacity actually in use. They aren't buying capacity from the phone companies any more. They're laying their own.

Cables break constantly. Between 150 and 200 faults happen worldwide every year, and that number has stayed roughly flat even as the network has grown. Almost none of it is in the deep ocean. Fishing gear and anchors cause between 70 and 80 percent of faults, and only about 2 percent happen out on the high seas. The dangerous water is the shallow water where people are working, which is the same water the shipping and port industry operates in. Canada produced the best illustration of this I've come across. A cable in the St. Lawrence kept snagging a Québec crab fisherman's anchor, and in June 2006, working from an old map he'd seen in a museum and believing the cable was abandoned, he hauled it up and cut it twice with an electric circular saw. The repair came to $980,433.54. Weeks later he read a newspaper story saying a Telus cable had been deliberately cut and the authorities were looking for whoever had done it, and he came forward. The case made it all the way to the Supreme Court of Canada, which capped what he owed at $500,000 under the Marine Liability Act, the statutory limit for a boat that size. That was well below the repair bill, but his insurance did not cover him. He paid it himself.

Telling an accident from an attack is harder than it sounds, and the question comes up more now. The routes are not secret. Cables are marked on nautical charts precisely so that fishing boats and ships know to keep their gear clear, which means anyone who wants to find one can. Roughly ten cables have been damaged in the Baltic since Russia invaded Ukraine in 2022. Around Taiwan, cables to the Matsu Islands have been cut more than twenty times in five years, and in 2023 the islands went more than fifty days without internet. NATO started patrolling the Baltic in January 2025.

Nobody can tell you how many of those breaks are deliberate. The industry body that keeps the fault statistics does not investigate incidents and has never published such a figure, and accidental damage still accounts for the overwhelming majority. The trouble is that an anchor dragging across a cable leaves the same mark whether the master was careless, asleep, or following instructions, and sorting out which one it was falls to police and intelligence services rather than to anyone who works on cables. The results vary. Sweden detained a ship in early 2025, established that it had cut a cable, and concluded the crew had done it by accident. American intelligence reporting on three of the Baltic incidents pointed to negligence rather than malice. Taiwan put a captain in prison for three years, because his ship dropped anchor inside a marked cable zone and then steered a zigzag across it.

Natural causes account for about a tenth of cable faults, and the largest one on record happened in Canada. In November 1929 an earthquake on the Grand Banks set off an underwater landslide that broke twelve transatlantic cables at once. They broke one after another, farther and farther downslope as the landslide moved, so the times the circuits went dead were a record of how fast it travelled. Working backward from those times was how the speed of an underwater sediment flow was first measured. The first cable came back into service after 35 days. The last repair wasn't finished until August of the following year. Eight cable ships spent about 855 days between them putting the twelve back together.

Repairing a break is closer to fishing than to engineering. The landing station can tell roughly how far along the cable the break is, and the survey records turn that distance into a position on the seabed. A ship goes out to that spot and drags a grapnel across the route until it hooks a cable two centimeters wide, sometimes kilometers below the hull. The crew cuts it, raises one end, tests it to find out which way the fault lies, buoys that end, then retrieves the other. The damaged length comes out and a fresh section goes in, because hauling both ends to the surface uses up slack that has to be replaced. Each fibre is spliced by hand, the joint is rebuilt to take the pressure, and the cable is returned to the seabed. The splice itself takes a day. Getting the ship there takes about a month. Three weeks of that is the ship still in port, waiting on paperwork, weather, or its turn.

There are only about thirty maintenance ships in the world, and their average age is 29 years. A ship like that costs too much to keep on standby for one cable, and it sits idle most of the time, so the companies that own cables pay into a shared agreement instead. Members split the cost of keeping a vessel waiting with spare cable already loaded, paying every year whether anything breaks or not, and whoever breaks first gets priority. The Atlantic one was set up in 1965, the same year the station at Heart's Content closed, and keeps three ships, at Portland in England, at Brest, and at Curaçao. The North American agreement has 31 member companies and one ship, Cable Innovator, which covers the northeast Pacific out of Port Angeles and Victoria. That's the ship that would sail if a cable broke off British Columbia.

Canada has roughly nineteen cable systems in service, on all three coasts. Cables leave Halifax for Cork and for Dublin, and one of them is the cable 360networks built, still working. Greenland Connect comes ashore in Newfoundland on its way to Nuuk and Iceland. Connected Coast finished in 2024, roughly 3,400 kilometers around the British Columbia coast, and getting there meant permits for 137 landing sites and agreement with First Nations and property owners along the whole route. In Nunavik, the Kativik Regional Government has laid more than 1,800 kilometers of fibre through Hudson Bay and Hudson Strait to connect eight communities. Nunavut still runs on satellite, which is one more line in the northern infrastructure gap.

I went into this expecting the story to be about scale, and it turns out to be about repair. The network carries almost everything and it is cut a couple of hundred times a year, mostly by anchors, and it keeps working because a few dozen elderly ships sit waiting under agreements written in the 1960s and somebody hauls the cable up and splices it by hand.