Cheap to Cut, Slow to Fix

To wound the West, all you need is an anchor. A year after I first made that argument, attackers have cheaper tools and the cables carry more, while the fleet that mends them has not grown.

ISlow to fix

The danger to the West's undersea cables is not that they can be cut. Somewhere in the world a cable is damaged three or four times a week, most often by fishing gear or an anchor.1 The danger is that they cannot be mended fast enough when several are cut at once. Nothing in the year since I first made this argument has changed that.

Take Shetland. On 3 October 2025, during Storm Amy, the Shefa-2 cable that links Shetland and Orkney to the Scottish mainland broke for the second time in ten weeks. The repair ship did not leave Calais until 20 October. Bad weather then held it off Orkney, and many customers were without broadband for almost four weeks.2

Shetland was not cut off. Its traffic could still go north through the Faroes, and BT kept its customers online over a smaller back-up cable. Customers of providers that had not paid for a back-up route, among them Sky, TalkTalk and Vodafone, had nothing to fall back on.3 In July the same cable had been broken off Orkney, most likely by a fishing boat. The switchboard at Balfour Hospital in Kirkwall went down, and the repair took eleven days.4 The owner has told MPs the cable has suffered more than twenty major breaks since it went live. It says a repair costs anywhere between £400,000 and £800,000.5

Vietnam shows the same problem at national scale. From November 2022 its five international cables failed one after another. The country got by on the two that still partly worked, land links through China and Cambodia, and capacity bought in a hurry. Its internet providers said the repairs took eight months.6 In August 2026 four Vietnamese cables were down again, taking about 30 per cent of its international capacity with them.7

Neither case involved an enemy. Undersea cables carry more than 99 per cent of the world's data traffic,8 and deliberate damage accounts for under one per cent of their faults.1 The repair system is built for that steady rate of accidents, and even at that rate repairs are slow to start. Half of all repairs on record between 2007 and 2024 began within eight days of the fault, but some wait months. The average time before a repair starts more than doubled between 2012 and 2024, from under 20 days to over 50. The number of repairs and of repair ships stayed flat, and the ITU's advisory body puts most of the rise down to slow permits.9 The US Navy puts the average repair in 2024 at 51 days.10

An adversary does not need a nationwide blackout. It needs a handful of cuts, timed to land while the few repair ships are busy elsewhere, and the queue does the rest. A 2012 study for APEC found that spare capacity would absorb a single cut in Australia. It put the cost of losing every cable landing there at about 3.2 billion US dollars of lost activity, against 2.2 million dollars to repair the cables.11 A single cut is routine; several at once are not.

Most of the trends in the first version of this essay, published in September 2025, have since moved the same way. There have been further cuts and faults in the Baltic, off Perth and around Vietnam and Taiwan, though the worldwide count has stayed at about 150 to 200 a year.1 The cables carry more, and investment in new cable is at its highest since 2000 and 2001, largely on forecasts of AI demand.12 The obvious fallback, satellite internet, has shown how much its availability depends on one company and one government. Attackers have gained cheaper tools that reach deeper. And governments have spent their new money on watching the seabed rather than on mending what lies on it.

IIGlass on the seabed

An undersea cable is a bundle of optical fibre pairs, sixteen in a modern Atlantic system, wrapped in layers of plastic and insulation and, near the shore, steel armour wire. A modern system is designed to carry hundreds of terabits per second. The Grace Hopper cable across the Atlantic is designed for 352, and Meta has announced one for 2029 with a design capacity of a thousand.13

A cable in section: the fibres at the centre, then insulation and, near the shore, steel armour wire.

Most faults are caused by trawls and anchors, which work in water shallower than about 500 metres. When an anchor is involved, the cable usually breaks outright.14 Between 2010 and 2024, fishing and anchoring, including cases where the exact cause went unrecorded, accounted for 86 per cent of faults, and earthquakes and landslides for 7 per cent.15

Every step of a repair is slow. Testing from the shore station takes one to three days and finds roughly where the break is. Mobilising a ship takes three to five days, longer when permits are slow. The passage takes another two to five days, more than ten in remote waters. The ship then recovers the cable, which in deep water can take several days, and joints each end to new cable on deck. Each joint holds the ship stationary for 24 hours or more and needs a window of calm weather.16 Fewer than 1,000 fully qualified jointers are working worldwide, against more than 1.5 million kilometres of cable in service.17

Then there is the queue. Repair is organised through regional maintenance zones and private agreements, under which cable owners pay to keep a few ships on standby at a small number of depots. The ITU's 2026 report puts the ships dedicated under these arrangements at about 20 to 25 worldwide, not all of them on continuous standby. All of Africa has a single dedicated repair ship, based in Cape Town. A ship works one break at a time, and when several faults need the same ship, it is often sent first to the routes carrying the most traffic.18

In January 2022 a volcanic eruption cut Tonga's only international cable. The repair ship, SubCom's Reliance, was off Papua New Guinea, more than 4,200 kilometres away, and had to stop in Samoa for spares. It reached Tonga eighteen days after the eruption, and service came back after five and a half weeks.19

IIICheap to cut

The Baltic shows how hard it is to tell an attack from an accident. In October 2023 an anchor that Finnish investigators matched to the Hong Kong-flagged container ship Newnew Polar Bear damaged the Balticconnector gas pipeline and two telecom cables; China called it a storm accident.20 In November 2024 two data cables near Gotland were cut while the Chinese bulk carrier Yi Peng 3 was crossing them. It switched off its transponder for about seven and a half hours after passing Gotland, Chinese officials led the only inspection on board, and the ship sailed on. Both cables were back in service within about ten days.21 On Christmas Day 2024 the Cook Islands-flagged tanker Eagle S, believed to belong to Russia's shadow fleet, was suspected of dragging its anchor across the Estlink 2 power cable and four telecom cables. The power link was out for about six months.22 Sweden seized the Maltese-flagged bulk carrier Vezhen after a cable failed in January 2025, and in October closed the case as an accident caused by bad weather, faulty equipment and poor seamanship. On New Year's Eve 2025 telecom cables between Finland and Estonia were cut. Finnish investigators found that the cargo ship Fitburg had dragged its anchor for at least 130 kilometres, and its captain and bosun denied the charges, and in November 2026 a Helsinki court convicted them of aggravated sabotage and jailed them; the ruling can be appealed. The cable's operator rerouted its traffic, and customers were not affected.23

In the first version of this essay I said the suspects were Russian-flagged. Apart from one early suspect, they were not. They flew the flags of Hong Kong, China, the Cook Islands, Malta and St Vincent and the Grenadines. The Russian connection, where there was one, ran through ports of departure, shadow-fleet status and crews.24 That makes the argument stronger. A deniable campaign does not need a single ship flying the attacker's flag.

The pattern repeats elsewhere. A Chinese fishing boat and a Chinese cargo ship damaged the two cables to Taiwan's Matsu Islands in February 2023, and residents relied for fifty days on a microwave link carrying a fraction of normal traffic.25 In February 2025 the Togo-flagged Hong Tai 58 severed the Taiwan to Penghu No. 3 cable.26 In August 2026 two cables failed in quick succession inside the protection zone off Perth; the cause is not yet known, and their operator asked the federal police to investigate.27 In the Red Sea, the Rubymar, a cargo ship hit by a Houthi missile in February 2024, drifted and most likely dragged its anchor across three cables.28 Nobody has shown that the Houthis set out to cut cables, but their war damaged them anyway. A dispute over permits and over who controlled access to Yemeni waters kept all three out of service until July.29 In September 2025 two Red Sea systems, SMW4 and IMEWE, failed near Jeddah, and Kuwait said a third, FALCON GCX, had been cut; no cause was established.30

  1. Gotland, 2024about 10 days
  2. Shetland, July 202511 days
  3. Shetland, October 2025about 27 days
  4. Tonga, 2022about 38 days
  5. Matsu, 202350 days
  6. Red Sea, 2024about 140 days
  7. Estlink 2, 2024about 180 days
  8. Vietnam, 2022about 240 days

US Navy average repair, 2024: 51 days

How long each outage in this essay lasted.

What has changed since 2025 is the attacker's toolkit. Ukraine's sea drones, together with missile strikes, pushed Russia's Black Sea Fleet out of Sevastopol, and by May 2026 its warships rarely sailed out of Novorossiysk into open water. In December 2025 Ukraine said it had hit a Russian submarine in Novorossiysk with an underwater drone. Russia said no ship was damaged, but British intelligence judged the boat had likely been left unable to sail.31 In February 2025 engineers at a Chinese state research centre published a design for a compact cable cutter rated to 4,000 metres, for crewed and uncrewed submersibles; one commentator questioned its relevance, since cables are not armoured at such depths.32 At its September 2025 military parade China showed two classes of large uncrewed submarine. In September 2026 satellite images showed a 204 metre ship under construction in Shanghai. Analysts assess it as a mothership for those submarines and say it could be used to cut cables.33

Russia has gone further. In January 2025 the UK defence secretary said that Yantar, a ship run by Russia's Main Directorate of Deep-Sea Research, had been mapping Britain's undersea infrastructure, and in November 2025 he said it had directed lasers at RAF pilots tracking it.34 Reuters reported in September 2026 that British, Norwegian and American forces had interrupted a Russian rehearsal of cable-cutting methods near Svalbard that spring; no cable was damaged.35 An underwater vehicle no longer needs a ship on station to stay on task. Commercial seabed docking stations already let one stay on the bottom for more than six months, recharging and taking new instructions, where an offshore field supplies the power.36

The defender has gained drones to watch cables, but none that can mend one. A repair ship has to hold still over the fault for a day or more per joint, which in a war zone makes it an easy target. A UK parliamentary committee heard that repair ships would probably not sail into a conflict zone without an escort.37

The uncrewed craft are cheap. Ukraine's Sea Baby sea drone was priced at about 218,000 dollars in a 2024 fundraiser. A small tethered underwater vehicle lists from about 4,900 dollars; military-grade autonomous ones have no published price.38 The hull is the expensive part. This summer an old Aframax tanker sold for scrap went for under eight million dollars, while a fifteen-year-old one at market was valued at about 54 million.39 My own estimate is that a campaign of several cuts would cost about ten million dollars with a near-scrap hull. That is a small sum for a state that already runs a shadow fleet: Western owners sold at least 230 old tankers into Russia's for more than six billion dollars.40 Against such a campaign there are some 60 to 70 ships able to repair a cable, of which only 20 to 25 are dedicated to repair or on long-term charter for it, and the ITU says fewer still could sail at once.41 Spare routes absorb a single cut, as the Baltic cases showed. Several cuts on one corridor, timed to land while those ships are committed elsewhere, would leave faults waiting with no ship free to take them.

IVMore at stake, fewer ships

The cables are now part of the AI build-out. Content and cloud companies used 75 per cent of international capacity in 2025, and TeleGeography puts them at 80 per cent or more of demand on the Atlantic, Pacific and intra-Asian routes.42 It counts more than 16 billion dollars of new cable due between 2026 and 2029, and says the industry has not seen investment on this scale since 2000 and 2001.12 Meta's Waterworth programme runs to more than 50,000 kilometres; Amazon Web Services is building its own Atlantic cable, Fastnet, due in 2028; Microsoft has joined a new cable from India to Malaysia and Singapore.43 EXA Infrastructure says its own Atlantic capacity, which it expected to last until 2030, will run out in early 2027. It also says the cloud companies are keeping more capacity on their own cables for AI.44 TeleGeography cautions that AI traffic cannot yet be measured separately, and that its effect on international capacity has so far been limited.45 The cables are being laid in the expectation that it will grow. Either way, the more of the economy runs as always-on services in distant data centres, the more an outage costs. When the Red Sea cables failed in September 2025, Microsoft warned Azure customers to expect higher latency on traffic through the Middle East, adding that undersea fibre cuts can take time to repair.30

The Gulf shows where that leads. More than 90 per cent of data capacity between Europe and Asia passes through cables under the Red Sea, while Gulf states build AI campuses such as the planned 5 gigawatt UAE-US site.46 In March and April 2026 strikes hit Amazon Web Services data centres in the UAE and Bahrain, and in September AWS said it could not restore access to some customers' data.47 In March, Alcatel Submarine Networks (ASN) declared force majeure in the Gulf, leaving its cable installation ship in port at Dammam and stalling Meta's 2Africa extension.48 In May, Iranian officials and state-linked media suggested fees on the cables under the Strait of Hormuz and exclusive Iranian rights to repair them, at a time when only one repair ship was reported inside the Gulf.49

The boom and the last bust

We have been here before, and the last time explains why the repair fleet is so thin. In the late 1990s new companies borrowed to lay cable on speculation. Global Crossing, which had reportedly spent 13.7 billion dollars on its network, filed for bankruptcy in January 2002.50 Tyco wrote its global network down by 2.5 billion dollars in 2002, citing massive overcapacity, and in 2004 India's VSNL agreed to buy all 60,000 kilometres of it for 130 million dollars.51

The crash shrank the cable-ship fleet. Global Marine, then the world's largest cable ship operator, reported 23 ships in 2001 and 15 at the end of 2003. Its installation revenue fell by 81 per cent in a year, and its owner, which had paid about 908 million dollars for it in 1999 including assumed debt, sold it in 2004, after its bankruptcy, for one million dollars.52 Sixteen new cable ships were delivered between 1999 and 2003, most of them ordered at the peak. Then none was delivered between 2004 and 2010, many older ships were scrapped, and only five arrived between 2011 and 2020.53 Owners have been wary ever since, and no new cable entered service across the Atlantic for twelve years after 2003.54 Global Marine's incoming chief executive put the problem plainly in 2022: a cable ship is a 30-year investment, and the demand forecast is out of date by the time it is built. He asked how he was expected to convince a banker to lend the money.55 In recent renewals of the Yokohama maintenance zone's agreement, the ITU says, the contract term was cut from fifteen years to five, and lenders have since grown reluctant to finance new repair ships.56

Now the money is back. Meta expects capital spending of 130 to 145 billion dollars this year, and Amazon spent 173 billion in the twelve months to June.57 The Bank of England has warned that US share valuations are close to their most stretched since the dot-com bubble, and Google's chief executive has said no company would be immune if the AI bubble burst.58 Chip makers show how to build through a boom with the last bust in mind. Micron lost 2.3 billion dollars in a single quarter of the 2022 to 2023 memory slump.59 It is building new capacity now against 26 multi-year take-or-pay agreements carrying 32 billion dollars of customer commitments, mostly cash deposits. TSMC's chief executive said in January 2026 that he was very nervous about whether AI demand was real, and that he had talked to the cloud providers and checked their finances. TSMC has since raised its 2026 capital budget to between 60 and 64 billion dollars.60 Repair ships had underwriting like that once, in the Yokohama zone's fifteen-year contracts, and no longer do. The companies laying cable are spending the way the industry did in 2000. The companies that own repair ships are cautious because of what followed in 2002.

So the market will lay cable on the boom and leave repair to what is left over. By 2040 about two thirds of the maintenance fleet will reach the end of its working life. The industry's own estimate is that holding service at today's level needs about three billion dollars, to replace 15 ships and add five more.61 Only a handful of new repair ships are on order, none due before 2028.62 No boom will pay for the spare capacity a crisis needs. Spare capacity is idle most of the time, and a crewed cable ship is expensive to keep idle. Since the last bust, owners have been unwilling to build one without a contract long enough to pay for it. Uncrewed units could be far cheaper to leave waiting, which is the case section VI makes.

Satellites will not cover it

The fallback is smaller than it looks. SpaceX's own filing in May 2026 put Starlink's whole constellation at over 700 terabits per second of downlink, summed across the planet.63 That is the same order as a single new transatlantic cable: EXA's Meridian, due in 2029, is designed for over 500, and Meta's Petal, due the same year, for a thousand.64 Satellites carry a very small share of international traffic, and their traffic still has to come down at one of more than 400 ground stations and join networks on land.65

It is also conditional. In 2022 Elon Musk refused Ukraine's request to extend Starlink to Sevastopol for an attack on the Russian fleet. In February 2025 Reuters reported that American negotiators had raised cutting Ukraine's access during talks on minerals, which Washington denied. In March Musk wrote that Ukraine's front line would collapse if he turned Starlink off, while adding that he never would.66 In February 2026 SpaceX and Kyiv brought in a whitelist of registered terminals, and Kyiv said those used by Russian forces had been switched off. In October 2026 Ukrainian units found military terminals restricted overnight; Musk said such limits were for the US military to decide, a US official said Trump had not ordered them, and the service was restored.67 Iran jammed Starlink in January 2026, and a software fault took the whole network down for two and a half hours in July 2025.68 SpaceX itself warns investors that governments may be reluctant to rely on Starlink because its availability could be restricted for geopolitical reasons.69 Europe's own constellation, IRIS², will not launch its first satellites before 2029.70

Whose ships

The firms that make, lay and mend cables are national champions. Four suppliers build about 98 per cent of the world's cable systems.71 The French state bought 80 per cent of one of them, ASN, in 2024.72 Orange Marine belongs to Orange, whose largest shareholder is the French state, and a witness told a UK parliamentary committee he would bet the French government would influence its repair priorities.73 China's repair ships are plainly state-owned. S.B. Submarine Systems, a leading Asian repair firm on the rota of the Yokohama zone, has been wholly owned since 2023 by China Comservice, which the state-owned China Telecom group controls, and China Telecom's own cable subsidiary commissioned a new laying and repair ship in July 2026.74 China's cable builder is private on paper. HMN Technologies, formerly Huawei Marine, was sold to the Hengtong group in 2020. Hengtong is controlled by Cui Genliang and his son; Cui senior is a former PLA Air Force communications specialist, sat in the National People's Congress for ten years and is secretary of the company's own Communist Party committee. The United States put HMN on its Entity List in 2021.75 SubCom, the only American supplier, is controlled by funds managed by the private equity firm Cerberus. Cerberus's co-founder, Stephen Feinberg, sat on the President's Intelligence Advisory Board in Donald Trump's first term and became US Deputy Secretary of Defense in 2025, after pledging to divest his interests in the firm. Cerberus's venture arm was launched in 2023 by a man who had spent seventeen years running In-Q-Tel, the investment firm the CIA set up. I have found nothing that gives In-Q-Tel any part in SubCom itself. Reuters reported in 2023 that SubCom is the US military's exclusive undersea cable contractor and runs the two ships of the US Cable Security Fleet.76 Global Marine, Britain's main repair fleet, was sold in 2025 to a Singaporean infrastructure fund, and the same committee heard that Britain has no UK-owned repair capability.77 When Washington wanted the SeaMeWe-6 cable built by SubCom rather than HMN, it used diplomatic pressure, export loans and grants until the consortium switched.78

None of these firms is a government front. But during the 2026 fighting in the Gulf, ASN, 80 per cent owned by the French state, declared force majeure and left its installation ship in port. In a crisis a cable ship goes only where its owner will send it, and for several of these firms the owner is, wholly or in part, a state.

VWatching without mending

Governments have moved, and most of the money has gone to sensors. NATO's Baltic Sentry, set up in January 2025, added at least twenty uncrewed surface vessels to its ships and aircraft.79 After it began, no suspicious cable incident was reported in the Baltic for almost a year, until the Fitburg. NATO officials credit deterrence, though the quiet cannot be shown to be the patrols' doing. When deterrence fails, the repair queue is as long as it was before.80 Britain's Atlantic Bastion pairs frigates with uncrewed surface and underwater vessels to find submarines, with an extra 1.5 billion pounds over four years. A further 330 million pounds goes to protecting undersea infrastructure, including upgrades to RFA Proteus, a survey ship bought second-hand in 2023 to watch the seabed.81 Australia is buying dozens of Ghost Shark uncrewed submarines, and the US Navy is funding Orca.82 All of these find, track and deter; none of them mends a cable.

RFA Proteus watches the seabed around Britain. It is not built to repair a cable.

On repair, the record is thin. A UK parliamentary committee recommended in September 2025 that Britain buy a sovereign repair ship by 2030, noting that its main UK-based repair ship, CS Sovereign, was built in 1991. The government agreed that a UK-based, UK-flagged repair ship is still needed, and said in May 2026 that it would decide how to keep one by the end of the year, but its Defence Investment Plan defers "additional specialist shipping" for undersea infrastructure to the next parliament, without saying whether that includes repair ships.83 The US Navy laid up its only cable ship, USNS Zeus, built in 1984, between July and September 2025. Congress has asked for a plan to replace it, and in September 2026 the Navy hired a ship designer for feasibility studies. A 2019 law provides five million dollars a year each for two private ships to be available in a Cable Security Fleet, subject to yearly appropriation. One study reports that the money was cut in 2024.84 In the statement accompanying its 2026 defence bill, Congress also asked the Pentagon to report by 1 November 2026 on whether a cable could be repaired within 100 hours; if the report has been delivered, it has not been published.85

The European Commission wrote in February 2025 that today's repair ships would be insufficient against simultaneous attacks. Since then it has opened calls to pay for repair modules that fit to existing ships: a 20 million euro pilot in the Baltic and a 40 million euro call reported to cover the Mediterranean and Atlantic. The reserve fleet set out in its 2025 action plan does not yet exist.86 Japan's communications ministry has asked for about 100 billion yen to subsidise three to five new cable ships.87 In the first version of this essay I named Australia as the exception. It is not: a 2026 study funded by its defence department found no national cable strategy, no sovereign repair ship and no guaranteed access to one.88 At a summit in Porto in February 2026, attended by representatives of more than 70 countries, the ITU and ICPC advisory body affirmed a declaration naming faster repair as an aim. It is non-binding and commits no ships.89

The law is moving faster than the fleet. The UN Convention on the Law of the Sea obliges each state to make it an offence for its own ships and nationals to damage a cable beneath the high seas wilfully or through culpable negligence. But it gives no other state an express power to arrest a foreign ship outside its territorial sea.90 When Finland charged three officers of the Eagle S, a Helsinki court dismissed the case for lack of jurisdiction in October 2025. The Court of Appeal reversed that in August 2026, holding that Finland can try the officers of a ship that kept dragging its anchor after the authorities had made contact. The defendants have left the country. The ruling is not final: in October the officers and the shipowner asked the Supreme Court for leave to appeal.91 Taiwan jailed the Hong Tai 58's captain for three years and has since raised its penalties, and Hong Kong, as the flag state, has charged the captain of the Newnew Polar Bear.92 Britain's offence still sits in the Submarine Telegraph Act 1885, which a government white paper of October 2026 proposes to replace.93 The American one dates from 1888 and sets a maximum fine of 5,000 dollars, though general federal sentencing law probably raises it.94 Better law helps deter. It does not put a ship over the fault.

VIWhat must change

In war, days of lost communications could delay mobilisation, sow confusion or fracture alliance coordination. The aim is to make a loss of service short enough not to matter strategically. When restoration takes days, cutting cables is a nuisance. When it takes weeks, cutting them becomes policy.

Make restoration time the measure. Set targets by depth and weather, such as 48 hours for a near-shore fault in fair weather and seven days in deep water, publish them, and report against them. On the industry's own figures, 48 hours is at the edge of what even a ship on standby can do. In the Yokohama zone, where two ships are kept on standby, most faults in the core area are reached within 24 to 48 hours, against five to seven days for ships not on standby. A shallow-water repair then takes one to three days once the ship is on site.95

Pay for availability. The market will not build spare capacity, so governments should buy access to it the way they already buy sealift: a yearly stipend that keeps working commercial ships on call. The United States does this with its Maritime Security Program and set out to do it with its Cable Security Fleet. American officials told auditors that such a stipend is small next to the cost of owning ships that are not needed day to day.96 Give owners contracts long enough to finance a hull, as the fifteen-year Yokohama terms once did. Pool the ships among allies, with at least one ready to sail within twelve hours in each ocean that matters, and train the jointers to work them.

Disperse the standby. The economics that make uncrewed craft cheap to attack with should also make them cheap to hold in reserve. Idle cable ships cost tens of thousands of dollars a day, and a couple of Pacific repair ships already do only two repairs a year and otherwise stand by.97 A shore-operated or resident underwater vehicle carries no crew. My assumption, which no published price yet tests, is that such units can be cheap enough to leave idle almost all of the time.

The hardware is close. Commercial vehicles can already live on the seabed for months where an offshore field supplies the power, so far for inspection only. DeepOcean's 24 metre shore-operated Challenger carries a work-class vehicle rated to 1,500 metres, with repair work and light lifting on its task list. In a trial at Plymouth reported in October 2026, a smaller uncrewed vessel and vehicle worked from shore in one-metre waves, which the operator says opens a weather window crewed vessels cannot work in.98 In shallow water, finding the cable, cutting out the damage and gripping the ends are already routine work for remotely operated vehicles, though today they are launched from the cable ship.99

Units placed along the busiest routes could, as the vehicles mature, do those steps before the ship arrives and hold the ends. I have found no case of a resident unit doing them on a live fault, and the ship would still lift, test, joint and lay back. But every hour of work done before the ship arrives is an hour the ship does not spend over the fault, so each ship held ready could clear more faults in a campaign. I have found no uncrewed system that has jointed a fibre, and no published costing of such a fleet. The components already exist, though. When a container ship's anchor cut four cables off Shantou in 2023, in water too shallow for a cable ship, a barge from a pre-arranged list and a small team with a shallow-water vehicle did the work.100 Congress's 100-hour question asks about splicing robots by name.101

Clear the paperwork in peacetime. The ITU attributes most of the doubling in the time before a repair starts to permits. Singapore has cut its approval time to three to five working days, and Britain, South Africa and the Netherlands pre-authorise repair ships.102 Every coastal state should do the same, including for repair ships that are only passing through.

Stage the spares. Keep cable, repeaters, jointing kits and grapnels at ports near the riskiest routes, with customs cleared and loading rehearsed. A custom repeater can take more than a year to make.103

Change the law on cable damage. Give states jurisdiction over damage to cables that land on their shores, sanction the owners and insurers of ships that cause it, and hold flag states to account.

Buy the fallback in advance. Satellite back-up should be contracted before it is needed, from more than one supplier and more than one country.

Rehearse simultaneous faults. Run exercises that stack several faults at once, across regions and owners, and time every stage from the alert until traffic first passes through the mended fibre.

VIITrack gangs

In the Second World War, Allied air forces learned that railway lines were easy to cut and easy to mend. In France in 1944, German railway troops and Organisation Todt labour sometimes restored through traffic within hours of an attack on a rail centre. In northern Italy they kept repair material and Italian labour units along the main lines.104 The units proposed in section VI would do the same for cables: material and labour in place before the damage. Isolated cuts achieved little. Sustained, overlapping attacks had much more effect. In Italy the Germans were never quite able to keep up with the damage, and on the Loire the number of trains crossing fell from 400 in the first week of April 1944 to no more than 14 in the week ending 16 June. Supplies still got through by by-pass, road and night repair.105 A repair organisation that can keep up with one cut can still be overwhelmed by many.

That is the cable problem. Today's repair system is built for the ordinary rate of accidents, and even at that rate repairs are slow to start. It was not built for a campaign.

We have built an empire on glass. Keeping it intact means putting repair time at the centre of the planning: pay for ships to be ready, train the people, stage the kit, clear the permits and measure the time it takes. Make outages short and predictable, and cutting a cable becomes a headline, not a strategy.