The Cloud Has a Seabed Problem: How Submarine Cables Break and Get Repaired

The word cloud was a marketing decision, and it has quietly distorted how almost everyone thinks about digital infrastructure. It suggests weightlessness, atmosphere, something floating above the physical world and beyond its constraints. The reality is a few hundred fibre optic cables, most of them no thicker than a garden hose across their deep-sea sections, lying on the seabed. When a video call connects London to Singapore, the data travels through glass strands on the ocean floor, passing through repeaters powered by a current fed from shore, under water that is in places more than five kilometres deep.

This matters because physical things break, and these particular physical things break roughly 200 times a year. For most of the world most of the time, the breakage is invisible, absorbed by rerouting across spare capacity. For a meaningful minority of places it is not invisible at all. The International Telecommunication Union's International Advisory Body on Submarine Cable Resilience approved its final report on 10 July 2026, concluding a two-year effort involving more than 270 organisations from over 75 countries, and its findings amount to a careful statement that the system is less robust than its users assume. Here is what sits underneath the cloud.

 

The Internet Runs Underwater

 

Submarine cables carry more than 99 percent of international data traffic. Cloud services, video calls, streaming and financial settlement all depend on them, and satellites, despite considerable recent investment, carry a small fraction of the total.

The distinction between international and total traffic is worth preserving, because the headline figure is often repeated loosely. A great deal of internet traffic never crosses an ocean at all, since content delivery networks cache popular material close to users precisely to avoid the expense and latency of a long-haul crossing. What travels on submarine cable is the traffic that must cross: data to and from a server on another continent, a transaction cleared in another jurisdiction, a corporate network spanning regions. That category happens to include most of what makes a modern economy function internationally, which is why a cable fault that affects a small share of packets can affect a very large share of consequence.

The architecture also explains why satellite constellations do not solve this. Low-earth-orbit systems have genuinely changed connectivity for remote users and for places with no cable at all, and they provide a valuable backup. They do not approach the capacity of a modern submarine cable, which carries hundreds of terabits per second on a single system. Satellite is a lifeline for a cut-off island. It is not a substitute for a trans-Pacific trunk route.

 

An Anchor Can Break the Connection

 

Around 200 cable faults are reported worldwide each year, a figure the International Cable Protection Committee notes has stayed broadly steady from 2013 through 2024 even as total cable route mileage grew by roughly half. Stability in the absolute count against a growing network is, in fairness, a sign that protection practices work.

The causes are unglamorous and overwhelmingly human. Roughly 44 percent of faults are attributed to suspected fishing or anchoring without a determination between the two, 28 percent specifically to fishing activity and 14 percent specifically to anchoring, which puts around 86 percent of all faults down to vessels interacting with the seabed. Geological events account for about 7 percent, abrasion against rock for 4 percent, and component failure for only 3 percent. The cables themselves are remarkably reliable. It is the traffic above them that is the problem.

Geography follows from this. Around 44 percent of faults occur in territorial waters and 54 percent in exclusive economic zones, with just 2 percent on the high seas. Cables are most vulnerable close to shore, in shallow water, where fishing is concentrated, where ships anchor, and where the cable has to come ashore at a landing station alongside every other cable serving that coast. The deep ocean, counterintuitively, is the safe part of the journey.

Whether a fault is noticed depends entirely on redundancy. A country connected by a dozen systems reroutes and nobody notices. A country connected by one or two does not. When cables serving Taiwan's Matsu Islands were cut in February 2023, roughly 13,000 residents lost normal internet access for about fifty days. That is the same event, physically, as a cut that a well-connected economy would absorb within seconds.

 

Finding the Fault Is Only the Start

 

Locating a break is the easy part. Operators send an optical pulse down the fibre from the landing station and measure the reflection, which gives the distance to the fault with considerable precision. Everything after that is a marine operation.

A specialist cable repair ship has to be available, has to transit to the site, and has to obtain permission to work there. Once on station in deep water, the crew drags a grapnel along the seabed to hook the cable, cuts it, recovers one end, buoys it off, then recovers the other end. A new section is spliced in aboard the ship in a jointing cabin, each fibre fusion-spliced and tested, the joint rebuilt and sealed against pressure, and the repaired cable laid back down with enough slack to reach the surface, which means every repair makes the cable slightly longer. In deep water the whole sequence can take weeks. Weather stops it. Sea state stops it.

The permission problem is the one that has grown worst and the one the ITU report singles out. Since 98 percent of faults occur inside national jurisdiction, nearly every repair requires consent from a coastal state, and the legal architecture is weaker than most people assume. The UN Convention on the Law of the Sea establishes freedom to lay and maintain cables on the high seas, but inside an exclusive economic zone a repair ship is operating with the coastal state's permission rather than by right, and inside territorial waters more so. Permitting regimes were not designed for emergencies. The consequence is measurable: industry data indicates repair response times have more than doubled over the past decade, and the longest single repair recorded in 2023 took 947 days. That is not an engineering failure. It is a paperwork failure.

Costs vary enormously by cable type. A fibre optic repair typically runs between 500,000 and one million dollars. A subsea power cable repair runs between ten million and a hundred million, and the EstLink 2 electricity interconnector damaged in the Gulf of Finland in December 2024 was estimated at 50 to 60 million euros.

 

The Repair Fleet Is the Real Bottleneck

 

Beneath the permitting problem sits a harder one. The world has roughly 60 ships capable of repairing a submarine cable, a specialised fleet operated under maintenance agreements that pool the cost across cable owners, and it is ageing.

Analysis by TeleGeography and Infra-Analytics found that around two thirds of cable maintenance vessels will reach the end of their service lives by 2040, while total cable kilometres are projected to grow by about 48 percent over the same period. Closing that gap would take roughly three billion dollars: fifteen replacement ships plus five additional vessels simply to hold current service levels as the network expands. Nobody obviously owns that bill. Cable maintenance is funded through shared agreements that have been optimised for decades to be as cheap as possible, which worked when the fleet was young and is precisely the wrong structure for financing a fleet renewal.

This is the quiet centre of the resilience problem. Adding cable routes is comparatively easy, because hyperscale cloud companies now fund much of the new construction themselves and have strong commercial reasons to do so. Maintaining the ability to fix what exists attracts no such investment, because repair capacity is a classic public good: everyone benefits from it, nobody individually benefits enough to fund it, and its absence is invisible until the week it matters.

 

Accident or Sabotage

 

The run of incidents since late 2023 has been widely read as a campaign of deliberate interference, and here the evidence requires care rather than confidence.

The record is substantial. The Balticconnector pipeline and adjacent telecom cables were damaged in October 2023 by the anchor of the Hong Kong-flagged NewNew Polar Bear. Three cables in the Red Sea were cut in February 2024 by the drifting Rubymar after it was struck by a missile. C-Lion1 and the BCS East-West Interlink were cut in the Baltic in November 2024 with the Chinese bulk carrier Yi Peng 3 nearby. EstLink 2 and four telecom cables were damaged in December 2024, with anchor drag marks on the seabed and the Cook Islands-registered tanker Eagle S detained. Taiwan recorded incidents in January and February 2025. Three major cables including SEA-ME-WE 4 and IMEWE were cut in the Red Sea in September 2025, slowing traffic across India, Pakistan and the Gulf. Finland and Estonia lost connections again in December 2025, with anchor drag confirmed by March 2026. In August 2026, the Indigo West and Indigo Central cables off Perth were damaged with a UAE-managed tanker under investigation by Australian federal police.

The outcomes are much less uniform than the pattern suggests. Swedish authorities found no conclusive evidence of wrongdoing in the Yi Peng 3 case and ruled the January 2025 Latvia to Gotland damage accidental. A Finnish court dismissed the case against the Eagle S captain in October 2025. United States and European officials concluded that several of the Baltic incidents were accidental rather than sabotage. Against that, the captain of the Hong Tai 58 was convicted in Taiwan and sentenced to three years over the Taiwan to Penghu No. 3 cable, and the NewNew Polar Bear captain faced criminal damage charges in Hong Kong. The April 2026 Matsu fault turned out to have been caused by the wreck of a stranded fishing boat shifting onto the cable.

Two things are true at once. Anchor dragging is a genuinely common accident, particularly among poorly maintained vessels in bad weather, which is why it accounts for a large share of ordinary faults every year and why investigators keep reaching inconclusive findings in good faith. And a method that is indistinguishable from an accident, in waters where attribution is legally difficult and enforcement powers over a foreign-flagged vessel in transit are thin, is an unusually attractive one for a state that wants deniability. The honest position is that the cluster of incidents is anomalous, that individual cases have mostly not been proven deliberate, and that the current legal regime is poorly equipped to establish the difference either way.

 

What Resilience Actually Requires

 

The ITU Advisory Body's recommendations fall into three areas: timely deployment and repair, better risk identification and monitoring, and greater connectivity and geographic diversity. Underneath those headings the specific asks are concrete, and they point at exactly the weaknesses above.

Streamlined permitting, so repair ships can work without waiting months for consent. Emergency procedures agreed in advance rather than negotiated during an outage. Investment in repair capacity, meaning ships and crews, not only in new routes. Route diversity, so that the countries served by one or two cables are served by more, and so that traffic is not funnelled through the same handful of chokepoints at Bab al-Mandab, the Luzon Strait and the Egyptian land crossing. Better monitoring to detect damage early and to establish what caused it. Attention to the regions where a single fault still means a national outage. And integration of climate risk, since seabed landslides, cyclones and changing sediment conditions all bear on where cables can safely lie.

None of this is technologically difficult. It is institutionally difficult, which is harder. Cables are built by private consortia, repaired by a shared commercial fleet, regulated by coastal states individually, and governed internationally by a convention drafted before the internet existed. The ITU's report does not resolve that fragmentation and does not claim to. What it does is establish, with the authority of a broad multilateral body, that the infrastructure beneath the cloud is a shared responsibility rather than someone else's problem.

The practical takeaway for anyone with an interest in the blue economy is that digital resilience is now an ocean infrastructure question. It depends on seabed surveys, cable-aware marine spatial planning, enforcement against anchoring in protection zones, a working repair fleet and crews trained to use it. More cable routes, faster access for repairs, fewer single points of failure. The cloud, for all its weightless branding, rests on the seabed, and it will be no more reliable than the ships and the rules that look after it.

 

Note: This article reflects the state of the sector as of October 2026, drawing on sources including the ITU International Advisory Body on Submarine Cable Resilience final report of July 2026, International Cable Protection Committee fault statistics, TeleGeography and Infra-Analytics research on cable maintenance capacity, and reporting on individual incidents. The figure that submarine cables carry more than 99 percent of data traffic refers to international traffic rather than all internet traffic. Investigations into several incidents described here remain open, and attributions are stated as the responsible authorities have reported them.