Undersea Cable Timeline 1858–2026: How Ocean Fibre Became Critical Internet Infrastructure

A broken cable is serious. A broken network is something different. When a submarine cable is damaged, the systems and services relying on it don’t simply vanish — the internet’s routing layer, built on the Border Gateway Protocol (BGP), looks for another path. If a region has multiple cables and multiple landing stations, that path usually exists, and rerouted traffic keeps flowing with somewhat higher latency and less spare capacity. If a region depends on only one or two cables, the alternate paths may not exist, or may be far slower (a satellite backup, for instance), and the outage can be severe and long.

This is the central, non-sensational insight of subsea cable resilience: the network survives cable failures not because cables cannot break, but because well-connected networks are designed with alternative paths. Resilience depends on how many usable routes exist when one goes down — not on the toughness of any single cable.

Why cables win on bulk traffic: a single modern fibre-optic cable pair can carry tens of terabits per second; even large satellite constellations carry a small fraction of that per satellite, and physics keeps it that way — fibre has far lower attenuation over long distances than a radio signal has to travel to and from orbit. Fibre capacity is also upgradeable: operators can boost a cable’s usable capacity for years after it is laid by upgrading the equipment at each end, without touching the cable itself. And the economics are lopsided: the cost per transmitted bit on a modern subsea cable is a small fraction of the cost per bit over satellite links.

Why satellites still matter, a lot: ships, aircraft, disaster response, military communications, and homes and businesses in places terrestrial and submarine infrastructure has not reached. Satellite internet is often the only option, or the fastest way to restore connectivity, when ground infrastructure is destroyed — as it was, briefly, for parts of Tonga in 2022 (see case studies below).

SvalSat, the satellite ground station on Svalbard, is the world’s largest satellite ground station and part of NASA’s Near Space Network. It downloads huge volumes of satellite data — but that data still has to travel onward, and it does so over two 1,400 km subsea fibre-optic cables linking Svalbard to mainland Norway, lying as deep as 2,700 metres. That is exactly why those two cables became a security concern in 2026: threaten the cables, and you threaten the usefulness of the satellite ground station behind them, even without touching a single satellite.

Submarine cables are not owned the way a national road network or a single company’s data centre is owned. Ownership is typically a consortium: telecommunications carriers, technology companies, infrastructure investors and, in some cases, state-linked entities, sharing the cost and capacity of a single system. A cable can carry traffic for dozens of different networks simultaneously, each of which owns or leases a “fibre pair” or a slice of capacity, not the whole physical cable.

Announced capacity is not the same as lit (activated) capacity, and an announced cable is not an operational cable until it has been laid, tested and brought into service — treat any single figure above as a snapshot, not a permanent spec.

Construction is not identical across every system, but the general pattern holds: in deep water, where the main risks are natural (not human activity), cables can be startlingly thin — not much thicker than a garden hose. Near shore, where anchors, trawling and coastal construction are real risks, the same cable gets extra layers of steel armour and is often buried beneath the seabed for physical protection. There is no single universal burial depth — it varies by seabed conditions and local regulation.

A cable landing station is not a data centre — it is the physical building where an undersea cable’s signal is converted and handed off to terrestrial fibre networks. Because building a new landing station is expensive and geographically constrained (by coastline, permitting and existing infrastructure), operators often route several different cables into the same handful of stations in a country. That is efficient, but it also means a single location can matter more to a country’s connectivity than the number of cables alone would suggest — true resilience needs diverse landing points, not just diverse cables.

The Red Sea chokepoint: at the Bab-el-Mandeb Strait between Yemen and Djibouti, multiple major cables connecting Europe and Asia pass within a few kilometres of each other. That geographic concentration is efficient to build but risky to depend on — a single incident in that narrow corridor can degrade several unrelated cable systems at once, as it did in the February 2024 and September 2025 Red Sea cable cuts. It is not accurate to say the entire internet passes through Egypt or the Red Sea — but a disproportionate share of Europe–Asia capacity does, which is exactly the concentration risk this corridor illustrates.

Mumbai remains India’s primary international cable gateway, hosting the largest concentration of landing systems, with Chennai as the second-largest hub. That concentration has made both cities natural targets for new investment: Meta’s 2Africa lands at Bharti Airtel’s Mumbai cable station; Reliance Jio has been advancing systems including IAX and India-Europe-Xpress (IEX); and Google’s broader subsea push, alongside the America-India Connect initiative, centres partly on a new gateway at Visakhapatnam. The India-to-Singapore corridor (illustrated above) is one of India’s most important international routes, linking onward to Southeast Asia, and from there to the trans-Pacific and Europe–Asia corridors.

India is also actively diversifying beyond Mumbai and Chennai: the planned I-2SEA system will land at both Machilipatnam (Andhra Pradesh) and South Chennai (Tamil Nadu), and a landing at Digha (West Bengal) adds an east-coast option outside the traditional two-hub pattern. Spreading landing points across more of the coastline reduces the geographic concentration risk described above — a single storm, cable fault, or landing-station incident is then less likely to affect a large share of the country’s international capacity at once.

For most of submarine cable history, the pattern was consistent: telecom carriers formed a consortium, built a cable, and sold capacity to whoever needed international bandwidth — including, eventually, large technology companies. That changed over the 2010s and accelerated through the 2020s. Google, Meta, Microsoft and Amazon began co-financing, and increasingly leading, entire cable systems rather than just buying capacity on someone else’s.

The reasons are straightforward: growing cloud and data-centre demand needs guaranteed capacity between specific regions; lower, more predictable latency matters for services and, increasingly, for AI workloads; owning capacity outright gives more control over routing and expansion timing than leasing does; and cable ownership supports the broader push toward content distribution close to users. By 2026, hyperscalers together held roughly 71% of global lit submarine capacity and were involved in more than two-thirds of newly planned deployments — a genuine structural shift, not a marginal trend.

It is tempting to credit the current cable-building boom entirely to AI. That overstates it. Cloud computing, video streaming, social media, financial networks, enterprise software and everyday consumer traffic were already driving sustained demand for new subsea capacity well before generative AI became a mainstream workload. What AI adds is a further, real increment on top of that existing demand: training data and datasets sometimes move between regions, inference requests can cross borders depending on where a service’s infrastructure sits, and hyperscalers building out AI-focused data-centre clusters (in India, the US, the Middle East and elsewhere) do factor future AI capacity into their subsea investment planning.

What is not accurate is the idea that every individual AI prompt crosses an ocean, or that AI alone explains the pace of cable construction in 2025–2026. It is one layer in a much larger, pre-existing demand curve.

Ordinary risks — anchors, fishing gear, seabed earthquakes and landslides, equipment ageing and construction activity — explain the overwhelming majority of the roughly 150–200 cable faults recorded worldwide each year. Global internet resilience is built around this reality: engineers expect faults, and networks are designed to route around most of them without users noticing.

Security risks — deliberate interference, sabotage, espionage-linked activity and wartime targeting — are a separate, much smaller category by volume, but a much higher-consequence one when they occur, because they can be deliberately timed, targeted at chokepoints, or paired with other actions. The Baltic and Svalbard incidents covered later in this article sit in this second category, and this article is careful to label suspicion, investigation and confirmation separately — because conflating “damaged” with “sabotaged” is exactly the kind of claim that erodes trust in reporting on this topic.

The repair-ship problem: estimates of the global specialised cable-repair fleet range from roughly 60 to about 80 vessels, depending on how “dedicated” is counted, and a large share of that fleet is decades old — industry estimates suggest around two-thirds could reach end of service life by 2040. Repair capacity is not evenly spread across the world’s oceans, and getting a ship to a remote fault site, securing permits to work in another country’s waters, and waiting out bad weather often takes far longer than the actual splicing work once the ship is on site. This is a genuine, widely discussed resilience gap — distinct from, and arguably more urgent than, the sabotage concerns that get more headlines.

In the open ocean, cables typically rest directly on the seabed, which can be several thousand metres down — the two Svalbard cables, for example, run as deep as 2,700 metres. Closer to shore, in shallower water where fishing, anchoring and coastal construction are common, cables are frequently buried beneath the seabed for physical protection, typically to a depth of roughly one to a few metres, though the exact depth varies by seabed type and national regulation. There is no universal burial depth that applies everywhere.

NATO launched Baltic Sentry in January 2025, in direct response to the accumulating pattern of Baltic infrastructure incidents. The mission combines frigates, maritime patrol aircraft and a small fleet of naval drones, integrating individual member states’ surveillance systems with alliance-wide resources to improve detection of suspicious activity near undersea infrastructure. It is coordinated through Allied Joint Force Command Brunssum, with Allied Maritime Command playing a central role. As of September 2026, Baltic Sentry remains active, and its assets and information-sharing arrangements were the template extended to the joint UK-Norway-US response near Svalbard.

Svalbard’s importance here is specific, not sweeping: it hosts critical satellite-ground-station infrastructure and sits in a strategically sensitive Arctic location where Russian and NATO interests increasingly overlap. It is not accurate to describe Svalbard as somehow “running the global internet” — its two cables are locally vital, not globally central. What makes the episode significant is what it reveals about intent and capability: a rehearsal of a cable-disabling technique, tracked and disrupted before completion, in an Arctic corridor of growing strategic competition between Russia and NATO members.

A country cannot be reliably “switched off” by one cable cut if it has multiple independent cables, landing points and cross-border terrestrial links — traffic simply reroutes. The real vulnerability sits with places that structurally cannot have that redundancy: small island states, remote territories, and any location served by only one or two systems. This is why the Tonga case study below matters more for understanding real-world risk than any single Baltic or Svalbard incident.

The ITU’s International Advisory Body on Submarine Cable Resilience, formed in the wake of the Baltic incidents, published its recommendations in 2026 after a February summit in Porto, Portugal, and later follow-up meetings. Its core themes: encourage route diversity so fewer regions depend on a single corridor; expand and modernise the global repair-ship fleet; speed up permitting for repair vessels working in foreign or contested waters; improve international information-sharing about suspicious maritime activity near cables; and build spare-parts and landing-station resilience so a single incident cannot cascade into a wider outage. None of these are quick fixes — they are multi-year infrastructure and diplomatic commitments, and their eventual impact will depend on follow-through by individual governments and cable operators.

International payments, cross-border financial trading, banking systems and enterprise cloud services do route over submarine cables when they cross oceans — that dependency is real. It does not follow that cutting one cable stops global banking; financial networks, like the rest of the internet, are built with redundant paths and, for the most latency-sensitive systems, dedicated diverse routes precisely because the consequences of an outage are so high.

Expect continued growth in cable count and route diversity, driven by hyperscaler investment and rising cloud/AI-adjacent data-centre demand. Expect fibre capacity per cable to keep increasing through better transmission electronics, without necessarily needing more physical cables for the same traffic growth. Expect hyperscaler ownership to keep rising as a share of total lit capacity. Expect more government and NATO-level monitoring of undersea infrastructure following the Baltic and Svalbard episodes, and continued ITU/ICPC-led work on repair-fleet capacity and route diversity. Arctic cable proposals exist and are being discussed as ice conditions and strategic interest change, but there is no reliable basis yet for predicting the Arctic will become a dominant route — that remains a proposal, not a guaranteed outcome.