Undersea cables: the repair fleet is the single point of failure

The global network looks robust on a map. TeleGeography's 2026 edition shows 694 cable systems and 1,893 landing stations, with more than 1.5 million kilometres of cable in service (TeleGeography 2026b). Redundancy is real: when cables were cut in the Red Sea in September 2025, Microsoft reported Azure services restored within a day by rerouting, and the internet did not go dark (Fortune 2025). Counting cables, however, measures the wrong quantity.

A cable is a long, buried or lying asset that will be damaged, on average, roughly every two days somewhere in the world. Resilience is therefore determined by how quickly faults are found and mended, which depends on ships, spare cable, jointing crews, permission to work, and safe water. These are scarce, slow to expand, and shared. A network with many cables and few repair ships is more fragile than it appears, because redundancy is consumed whenever a fault goes unrepaired.

The frame is a system with three layers: the cables themselves, the repair fleet that maintains them, and the access, legal and physical, that lets the fleet operate. The first is growing fast. The second is ageing. The third is increasingly contested by states. A failure in any layer degrades the system, and the weakest layer is not the one most discussed.

Most faults are accidents. The International Cable Protection Committee reports roughly 200 repairs a year, and about two-thirds of faults are caused by fishing vessels and ships dragging anchors, mainly in shallow water (TeleGeography 2026a). Earthquakes, landslides, and component failures account for the rest. Sabotage is a small share by number but a large one by attention, because attribution is difficult: a dragged anchor looks the same whether it was an accident or a choice.

A repair is a logistics chain. A fault must be located, a ship mobilised, a section of cable retrieved from the seabed, a replacement spliced in, and the system tested and relaid. A repair is commonly cited at one to three million dollars, using specialised ships that cost tens of thousands of dollars a day. Restoration of service is a separate and often costlier effort, since operators reroute traffic while they wait. Duration depends on location, ship availability, and permission: a Taiwan break detected in late January was repaired by early March, with the repair vessel arriving several weeks after the fault (TeleGeography 2026a).

The ease of repair varies enormously with geography. Where ships have access, a fault takes one to two weeks. Where they need assurances of safe passage, as in conflict waters, it can take months. Taiwan's Matsu islands lost internet access for more than 50 days in 2023 while they waited for a repair vessel, after cables were cut 12 times, mostly through anchor drag (Global Taiwan Institute 2025). The same cable is thus a few weeks of inconvenience in one place and a prolonged outage in another.

Estimates put the number of ships able to lay and maintain cables at roughly 60 (Recorded Future 2025). A study by TeleGeography and Infra-Analytics, commissioned by SubOptic and reported in 2025, finds that about two-thirds of cable maintenance ships will have reached the end of their service lives by 2040, that sustaining current service levels needs roughly 3 billion dollars of investment in about 15 replacement ships and five additional ones, and that deployed cable kilometres will grow by about 48 percent by 2040 (TeleGeography 2026a). Other reports cite 47 percent of vessels reaching end of life, so the fleet-age statistic should be treated as approximate.

New ships are being ordered but slowly. Global Marine has ordered a hybrid-powered maintenance vessel from Colombo Dockyard for delivery in the fourth quarter of 2029, describing it as the first step in a phased renewal, and OMS has ordered two cable vessels from Ulstein for delivery from 2028 (SubTel Forum 2026; Data Center Dynamics 2026). A ship takes years to design and build, so orders placed now arrive after the point at which demand for maintenance has grown.

The mismatch is structural. Cable kilometres are expanding by about half over 15 years while the maintenance fleet is shrinking toward retirement, and repair is a commons: no single owner of a cable has a commercial reason to fund spare capacity that benefits the network as a whole. Consortia hold maintenance agreements by region, but those agreements assume ships that are available and waters that are open.

Risk concentrates where routes squeeze. More than 90 percent of data capacity between Europe and Asia passes through cables under the Red Sea, which enter at Bab al-Mandab, a strait under 32 kilometres wide. Three cables were damaged there in 2024, most likely by the anchor of a drifting cargo ship that had been disabled by a Houthi missile, and repairs took months because crews needed assurances that they would not be fired upon. Further cuts near Jeddah in September 2025 hit SEA-ME-WE 4, IMEWE, and FALCON (Fortune 2025). In September 2026 the Houthis seized Yemen's side of the strait, and a Saudi-backed claim to have retaken key positions was denied (Rest of World 2026).

Other corridors show the same pattern. In March 2026 Iranian drones struck Amazon data centres in the United Arab Emirates and Bahrain (Rest of World 2026). Insikt Group counted four incidents involving eight cable damages in the Baltic Sea and five incidents around Taiwan in 2024 and 2025 (Recorded Future 2025).

Attribution is the political hazard. When a fault is ambiguous between accident and sabotage, governments cannot respond proportionately, insurers cannot price risk, and repair crews cannot assume that the water is safe. The cost of that ambiguity falls on repair, which is exactly the layer that cannot expand fast.

The legal framework adds friction. International law obliges states to criminalise wilful or negligent cable damage, but a vessel on the high seas falls under the jurisdiction of its flag state, so a coastal state that finds a ship over a broken cable often cannot detain it, investigate it, or recover costs without the flag state's cooperation. Operators can detect a fault within minutes through optical monitoring, yet the next steps, identifying the vessel, proving intent, mobilising a ship, and obtaining access, run on the slower clock of diplomacy and shipping. Until that sequence is shortened, a fault is a short event for the network and a long one for the people who depend on it.

Ownership has changed the economics. Google, Meta, Microsoft, and Amazon use about three-quarters of the world's international bandwidth, and TeleGeography expects investment in new subsea cable projects of roughly 13 billion dollars between 2025 and 2027, much of it led by hyperscalers (Rest of World 2026; Light Reading 2025). Meta's Project Waterworth is planned to span about 50,000 kilometres, linking the United States, Brazil, South Africa, India, Malaysia, and Australia, and is described as a firm owning the route that matters to it (Light Reading 2025).

Private owners are buying resilience directly. Since early September 2026, cloud firms have been telling operators how many fibre strands they need and by when. Google bought two fibre strands along Turkish state pipelines at a price reported at two to three times the roughly 6 million euros expected for newer Turkey-crossing lines, and Google and Meta have begun routing some live traffic across Iraq. Microsoft announced more than 400 million dollars of subsea and land links across the Middle East by 2030, and Iraq signed a contract with Qatar's Ooredoo on 10 September for an underground route toward Europe (Rest of World 2026).

The result is private resilience for the largest users and public fragility for everyone else. A hyperscaler with owned fibre can reroute. A national carrier, a regional operator, or a small economy relying on shared capacity cannot, and its repair queue is set by consortium agreements and the fleet that serves everyone. Resilience is being privatised at the same time that the shared repair commons erodes.

Cables carry the overwhelming majority of intercontinental traffic, and the alternatives are partial. A conference study of low-earth-orbit networks as national failover estimates that, even under optimistic assumptions, they could support only a fraction of the capacity of a single cable, for example about 32 percent of one cable serving South Africa (Assessing LEO 2025). Satellites are valuable for emergencies, when they can restore basic connectivity within days. They are not a substitute for bulk capacity.

Land routes add diversity at a cost. Overland capacity through Iraq and Turkey commands a premium, and an analyst quoted by Rest of World expects hyperscalers to keep land backups dormant until a cut, since subsea capacity is much cheaper (Rest of World 2026). Bypass subsea routes help partially: a planned Red Sea cable between Saudi Arabia and Egypt would add 140 terabits per second, but it crosses the same narrow region (Retail News Asia n.d.).

New fibre raises capacity but also concentrates risk. Cables are generally limited to about 24 fibre pairs because amplifiers must be powered from shore (Hengtong 2026), and multi-core designs aim to multiply capacity in the same cable; a deployed seven-core link reportedly reached 410.5 terabits per second over 140 kilometres in a field validation (Communications Engineering 2026). I found no deployed hollow-core subsea cable. More capacity per cable is efficient, but it means that each cut removes more bandwidth and each repair carries higher stakes.

The strongest objection is that the system has proved resilient. Hundreds of faults a year are repaired in weeks, redundant routes absorbed the Red Sea cuts within a day, hyperscalers are spending billions on diversity, new ships are on order, and ownership by well-capitalised firms may be more resilient than the consortium model it replaced. Most outages end quickly, and the number of cables is growing. This view is credible for the median fault. It is weaker for the tail: a chokepoint that closes for months, a conflict that denies access to repair ships, or a cluster of simultaneous faults that exhausts the fleet. The tail matters because the network's value is concentrated: the cables under the Red Sea carry most of the capacity between two continents, so one prolonged closure there outweighs many routine faults elsewhere.

Three paths follow. In managed renewal, governments and operators fund replacement ships, agree regional repair arrangements, and negotiate safe-passage protocols, and outages stay short. In privatised fragmentation, hyperscalers build owned routes and land links while the shared network and the repair commons decay, leaving smaller economies exposed. In denial by conflict, one or more chokepoints close for extended periods, and the world discovers that redundancy depends on a few narrow places. The signposts are observable: who controls and permits repairs at Bab al-Mandab, the number of new maintenance ships ordered beyond the two reported programmes, the volume carried on Iraq and Turkey land routes, war-risk effects in the Gulf, and whether the 3 billion dollar fleet gap attracts funding.

The record supports a selective conclusion. Connectivity stays resilient where a party can mend faults quickly, route around chokepoints, and secure access in contested waters: large cloud firms with owned fibre, and states with navies, ships, and diplomatic leverage. It degrades for those that depend on shared capacity in places where repair is slow, permission is political, or ships are far away. Resilience is increasingly a property of who owns routes, not of how many routes exist.

For governments, the priority is treating repair capacity as critical infrastructure: funding ships, building regional repair agreements, pre-negotiating safe passage, and stockpiling cable and jointing materials. For operators, the discipline is to map exposure by chokepoint and repair zone, not by cable count. For investors, the scarce asset is repair capacity and route diversity, not additional fibre capacity in already crowded corridors.

The internet's physical layer depends on a small fleet of ageing ships and the permission to use them. The cables get the attention, and the ships decide how long an outage lasts.