Tech · Infrastructure

Why Undersea Cables Became Critical Infrastructure — and the Repair-Ship Bottleneck

📅 Sep 18, 2026 🏷️ Explainer · Evergreen 🕐 8 min read

By Luminesca · Updated Sep 18, 2026
Analysis compiled from public reporting with AI-assisted drafting. See our editorial policy.

Almost everything you send across an ocean travels through a strand of glass on the seabed — fibre no thicker than a garden hose, laid across kilometres of water, carrying the overwhelming majority of the world's intercontinental data. That network is not fragile in the way outages suggest: it is engineered to lose pieces and keep working, and it does so roughly two hundred times a year. What has changed since 2023 is the assumption underneath the design. Damage that was once presumed accidental is now sometimes deliberate, sometimes ambiguous, and frequently unattributable — and the industry that fixes it depends on a fleet of roughly sixty ageing ships that nobody has been keen to pay for.

What is actually down there

Industry mapping tracks roughly 600 submarine cable systems in service, spanning about 1.4 million kilometres of ocean floor. NATO describes them as carrying around 95 per cent of global data flows; estimates for intercontinental traffic run closer to 99 per cent, and the alliance estimates USD 10 trillion in transfers move across them every day.

Satellites are not a substitute. Even optimistic projections put total global satellite capacity at roughly 50 terabits per second in 2026, against some 8,750 for subsea cables — a gap of well over a hundred times. Orbital broadband suits islands, ships and disaster zones. It cannot move the bulk traffic that cloud and AI workloads push between continents, and TeleGeography's public cable map shows why: the redundancy that matters is other cables.

Ownership has shifted as well. Traditional carriers have stepped back and hyperscalers have stepped in: TeleGeography forecasts roughly 5 billion dollars a year of new cable investment through 2035, with more than a million new cable-kilometres laid and over 450,000 kilometres retired. Meta's Project Waterworth alone spans about 50,000 kilometres.

Breakage is routine — the system is built to absorb it

Cables break constantly. Statistics compiled by the International Cable Protection Committee put the global total at roughly 200 faults a year, the bulk caused by fishing gear and ship anchors. The response is layered: redundant routes, traffic rerouting, and a repair discipline that assumes every cable fails eventually.

What that discipline costs is where the story turns. Repairing a fibre cable runs from about 500,000 to a million dollars; a power cable is a different order of magnitude. Median global restoration time sits around 40 days. Faults also cluster unevenly — Southeast Asia, and the South China Sea in particular, accounts for nearly half of all annual faults, a function of intense fishing and of overlapping claims that slow repair permits.

Commercial logic makes concentration worse: building along an established corridor, where landing stations and approvals already exist, runs 15 to 30 per cent cheaper than opening a new one, so new systems crowd the same narrow passages operators already know are the most exposed.

What changed: from accidents to ambiguous incidents

Since late 2023, more than eleven cables and pipelines have been damaged in the Baltic Sea in incidents involving ships, many associated with Russia's sanctions-evading shadow fleet. November 2024 brought simultaneous damage to the BCS East-West Interlink and the C-Lion1 cable. On Christmas Day that year, the Estlink 2 power cable and four telecom cables were damaged together; Estlink 2 took more than seven months to repair, at an estimated 50 to 60 million euros.

Finland seized the cargo ship Fitburg in December 2025; in March 2026 its National Bureau of Investigation confirmed the vessel had dragged its anchor across the Elisa cable over tens of kilometres. January 2026 brought six outages in six consecutive days.

Asia has its own version. Taiwan's Matsu Islands have had their cables cut more than twenty times in five years by one industry tally; in a 2025 case a court sentenced the captain of a Chinese-crewed, Togo-flagged vessel to three years for intentionally severing a cable. In the Red Sea, when AAE-1, Seacom and EIG were cut in February 2024, roughly a quarter of Asia-to-Europe traffic was disrupted and AAE-1 stayed down until July. Four more systems failed in one night in September 2025, in a strait where fourteen to seventeen systems pass through a corridor under 23 kilometres wide.

Attribution is the uncomfortable part. By mid-2026, officials in several Baltic and North Sea governments had grown more confident that most recent breaks were accidents rather than sabotage. Investigators keep returning to one limitation: on a tracking feed, an accidental anchor drop and a deliberate drag look identical.

The bottleneck is not bandwidth — it is ships, permits and money

The world maintains roughly 60 specialised cable ships, fewer than twenty of them dedicated solely to repair. The counts are contested — the ITU's advisory body has cited 80 to 100 hulls with only 20 to 25 under maintenance agreements — and there is no authoritative public register of the fleet that keeps the internet alive.

Those ships are old. Roughly 65 per cent of maintenance vessels reach the end of their working lives within fifteen years, and by 2040 about two-thirds of the maintenance fleet will be past service life. Replacements are hard to conjure: a purpose-built ship costs 100 to 150 million dollars and takes at least three years, and almost none were ordered for over a decade.

TeleGeography and Infra-Analytics modelled the consequence: 287 repairs a year by 2040, 49 per cent of them in East Asia, and a need for fifteen replacement ships plus five more in Asia — roughly 3 billion dollars just to hold today's service level. Maintenance is funded through decades-old cooperative zone agreements written for cheap shipping and accidental faults, and permits are the real schedule risk: approval to enter territorial waters can take a month or more, which is much of why AAE-1 stayed dark into July 2024.

What is being built instead

Governments have started treating the seabed as territory. On 5 February 2026 the European Commission adopted a Cable Security Toolbox alongside 347 million euros under the Connecting Europe Facility, including thirteen Cable Projects of European Interest staged to 2040. Sixty million euros is earmarked for repair equipment, twenty million for SMART cable sensing and twenty million for modular repair kits pre-positioned at ports. Proposals are open only to public bodies with an emergency-response mandate.

NATO's contribution is presence: Baltic Sentry, launched in January 2025, patrols with frigates, maritime patrol aircraft and naval drones, and the alliance says response time to suspicious incidents fell from seventeen hours to about one. Detection is improving in parallel — distributed acoustic sensing turns an existing cable into a sensor array — while the FCC and the United Kingdom have both tightened cable rules.

Note what almost none of this does: add a hull. Detection improves attribution, patrols improve deterrence, regulation changes who may own landing stations. Only the EU's twenty-million-euro repair-module call aims squarely at the queue.

What it means

The strategic asymmetry matters most. An adversary does not need to cut many cables; it needs to cut enough, in enough places, to saturate a repair fleet that already absorbs about two hundred accidental faults a year. Geography does most of the targeting, because redundancy has been consolidated into a handful of narrow corridors by commercial pressure rather than engineering necessity.

That reframes resilience. Diversity of routes and landing stations is the strongest lever, since the best backup for a broken cable is a different cable landing somewhere else. Deeper burial in anchor-prone shallows helps, as does better marine traffic management and cooperation with the fishing and shipping industries that cause most faults. Detection makes incidents attributable, which changes the incentive to cause them.

The unglamorous lever is the one nobody wants to sign: paying for maintenance capacity before it is needed. New cables have a business model — they sell capacity — which is why five billion dollars a year flows into construction. Repairs are a cost centre. Around half the industry respondents in TeleGeography's survey said they would pay more for faster repairs. What is missing is a mechanism, and a decade of ships.

Frequently asked questions

How much of the internet actually runs through undersea cables?

Almost all of it. NATO's own figures put around 95% of global data flows through undersea cables, and industry estimates for intercontinental traffic run closer to 99%. TeleGeography's cable map tracks roughly 600 systems in service spanning about 1.4 million kilometres of seabed. It is not only browsing either: NATO estimates the cables carry about USD 10 trillion in financial transfers every day, which is why the alliance now treats them as critical national infrastructure rather than telecom plumbing.

Can satellite constellations replace undersea cables?

No, not at this scale. Even optimistic projections put total global satellite capacity at roughly 50 terabits per second in 2026, against about 8,750 terabits per second for subsea cables — a gap of well over a hundred times. Satellites are genuinely useful for islands, ships, aircraft and remote regions, and as backup when a cable fails, but they cannot carry the sustained bulk traffic that cloud and AI workloads move between continents. The two technologies solve different problems.

Who owns and repairs submarine cables?

Ownership has shifted decisively toward hyperscalers. Google and Meta are now among the largest investors in new systems, and Meta's Project Waterworth is a single company's plan spanning roughly 50,000 kilometres. Repair is a separate, much older industry: a global fleet of about 60 specialist vessels, most of them converted second-hand ships, operating under cooperative zone agreements in which cable owners pay standing fees to keep a ship on call. That model was designed for accidental faults, not sabotage campaigns or hundred-million-dollar newbuilds.

How long does it take to repair a broken cable?

Longer than most people assume, and the delay is rarely technical. Industry measurements put median global restoration at roughly 40 days. When the AAE-1 system was severed in the Red Sea in February 2024, it was not fully restored until July — nearly five months — largely because permits and access in a fragmented conflict zone stalled the work. Approval to enter territorial waters can take a month or more on its own, and vessels may spend two to three weeks simply transiting to the site. Splicing the fibre is the quick part.

Are the recent cable cuts sabotage or accidents?

Often nobody can say, and that ambiguity is the real problem. More than eleven cables and pipelines have been damaged in the Baltic Sea since late 2023 in incidents involving ships, many of them linked to sanctioned shadow-fleet vessels dragging anchors for tens of kilometres. But by mid-2026 officials in several Baltic and North Sea governments had grown more confident that most recent breaks were accidental rather than directed. Investigators keep hitting the same wall: on a tracking feed, an accidental anchor drop and a deliberate drag look identical.