What happens when an undersea cable breaks is, for most people, nothing they would ever notice. The first sign is a routing change rather than a blackout. Traffic that took one path a millisecond ago now takes another, the packets land a few milliseconds later than they otherwise would have, and almost nobody spots it.
The short answer is that it depends on redundancy. A break on a route served by several cables is absorbed by the others within seconds and costs the operator money rather than costing users access. A break on a route served by one cable is an outage lasting weeks, because the repair needs a ship rather than a software change.
Cable breaks are treated as extraordinary events in the news and as routine maintenance by the industry that fixes them. Both views are defensible. The gap between them is where most of the confusion sits.
The cable itself is unremarkable
The thing carrying almost all intercontinental internet traffic is, in deep water, about as thick as a garden hose. Inside sit a handful of glass fibres roughly the diameter of a human hair, wrapped in steel wire, copper and polyethylene. Near shore the armouring gets heavier, because that is where the danger is.
The industry research firm TeleGeography counts more than 600 active and planned submarine cable systems as of 2026, totalling over 1.5 million kilometres in service. The copper telegraph lines that preceded them followed many of the same routes, for the same reason: geography has not changed.
Satellites carry 0.37% of US international capacity, on FCC data. The commission’s circuit capacity tables record every US-international cable in service. This is not a redundant backup network; it is the network.
Power is the other half of the system and gets less attention. Amplifiers spaced along the cable boost the optical signal every few tens of kilometres, and they are fed by a high-voltage current running through the copper layer from the landing stations at each end. A cable is not a passive strand of glass. It is a powered machine lying on the seabed.
Most breaks are accidents, and shallow
The International Cable Protection Committee, which collects fault data from operators, records roughly 200 faults a year worldwide. That figure has held steady since 2013 even as total cable mileage grew sharply — another way of saying the per-kilometre failure rate has been falling while the headlines got louder.
The causes are mundane. ICPC repair data attributes 28% of faults to fishing and 14% to anchoring outright, with a further 44% suspected to be one or the other but never conclusively identified. Geological events account for 7%, abrasion 4%, and failure of the cable’s own equipment just 3%.
Depth explains the pattern. A cable on the abyssal plain four kilometres down is essentially safe, because nothing human goes there routinely. One crossing a continental shelf shares water with trawl doors and anchor chains, and that is where it gets hit.
Geography compounds it. Southeast Asia, and the South China Sea in particular, accounts for something close to half of all faults recorded worldwide — a product of dense fishing activity, heavy shipping and shallow water, layered over jurisdictional disputes that make repair permits slow to obtain.
Finding the break, then fixing it
Locating a fault is the quick part, and it happens from shore. Operators fire a pulse of light down the fibre and time the reflection that comes back from the damage, a technique called optical time-domain reflectometry. Because the speed of light in glass is known precisely, the distance to the break can be calculated to within a few hundred metres of cable.
Knowing where it broke does not get it fixed. That needs a ship.
A repair vessel has to be available, has to sail to the site, and has to be permitted to work there. It drags a grapnel across the seabed to catch the cable, cuts it if necessary, lifts one end to the surface and buoys it off, then recovers the second end.
A fresh length is spliced in, tested, and lowered back down in a deliberate slack loop. Weather stops all of it.
Permitting is often slower than splicing. Because most faults fall inside territorial waters or an exclusive economic zone, repairs wait on a coastal state to issue clearance, and states in contested waters are not always in a hurry. The 2023 data includes one repair that took 947 days.
- A single repair typically costs between $500,000 and $1 million.
- In 2023 there were 206 repairs across 136 jurisdictions — 44% in territorial waters, 54% inside exclusive economic zones, and just 2% on the high seas.
- Average response times have roughly doubled over the past decade.
Why you usually do not notice
Well-connected places survive cable cuts because they are multiply connected. An operator holding capacity on four systems out of Europe loses one and shifts the load onto the other three, paying in latency and money. The failure stays invisible to the user by design, and that design is expensive.
Thinly connected places have no such option. Where a country or island depends on one or two cables, a single fault is not a rerouting event but an outage, and the clock is set by ship movements rather than by software.
Tonga is the clearest illustration. When the Hunga Tonga eruption tore through the seabed in January 2022, it destroyed the single 827-kilometre cable linking the country to Fiji and the wider internet. There was no alternative path to fail over to, because there was no alternative cable.
The repair ship Reliance replaced a 92-kilometre section, and the country came back online on 22 February — about five weeks after the eruption. For that period an entire nation ran on satellite capacity, which is exactly the fallback the 0.37% figure describes.
Sabotage is easy to allege and hard to prove
Since the Baltic incidents of recent years, deliberate damage has become the default public explanation for any fault in politically tense water. Sometimes that suspicion is warranted. Usually the evidence never gets far enough to say.
The difficulty is specific rather than general. Establishing a physical cause for cable damage is hard enough; proving intent is much harder. A dragged anchor leaves the same marks on the seabed whoever was holding the wheel, and whatever they meant by it.
The Washington Post has reported that some Baltic incidents were assessed as accidents caused by inexperienced crews on poorly maintained vessels. That is not quite exculpatory. A badly run shadow fleet damages things without needing to be ordered to.
Four faults a week is the baseline. Any given one is far more likely to be a trawler.
The law was written for telegraph cables
The legal regime is older than the technology it governs. The Convention for the Protection of Submarine Telegraph Cables dates from 1884 and still binds its parties, and as NOAA’s summary of the framework notes, its core provisions were carried into modern law largely intact.
Article 113 of the UN Convention on the Law of the Sea requires states to make it a criminal offence for their own ships to break a submarine cable outside territorial waters. Read quickly, that sounds like protection. Read carefully, it is an obligation on the flag state — the country whose flag the offending vessel flies — and on nobody else.
That is the whole problem. A ship registered in a jurisdiction with no interest in prosecuting will not be prosecuted, and comparatively few states have passed the domestic legislation that would let them act even if they wanted to. Coastal states have clear authority over artificial islands in their waters and markedly less over a cable crossing them.
The record bears this out. Finnish prosecutors pursued the tanker Eagle S over cables cut in late 2024 and the case was dismissed by a Helsinki court on jurisdictional grounds; a Swedish investigation into the bulk carrier Vezhen closed without charges. Investigators keep reaching the same wall, and it is a legal one rather than a forensic one.
This matters because the response menu is thin. Attribution too weak for a courtroom still has to carry a political decision, and the instrument governments reach for is economic rather than military — the same problem, and the same limits, that shape how international sanctions work.
Who owns the cables, and who fixes them
Ownership has shifted underneath all of this. Cables were historically built by consortia of telecoms carriers sharing the cost of a route none could justify alone. That model still exists, but it is no longer where the money is.
TeleGeography’s tracking puts content and cloud companies at roughly three-quarters of all international bandwidth actually used, and close to 90% of trans-Atlantic capacity. Google holds a stake in something like 34 systems and Meta in about 20, with Microsoft and Amazon accounting for a further handful between them. Meta’s 2Africa, a 45,000-kilometre system looping the continent, is among the longest ever laid.
They are not laying it out of altruism. They lay it because their own traffic is what fills it, and video fills more of it than anything else — the same appetite that keeps shifting shows between streaming services is what makes another trans-Atlantic system worth the capital.
The ships have not kept pace. Cable maintenance vessels are specialised, few and old; TeleGeography estimates that roughly two-thirds will reach the end of their service life by 2040, and puts the cost of renewing the fleet — 15 replacement ships plus five additions — at around $3 billion.
Nobody obviously owns that bill. Maintenance is bought through shared agreements between consortia, which handles routine faults well and gives no single company a reason to fund spare capacity for a bad year. The companies commissioning most of the new cable have so far not been the ones buying ships.
What to watch when the next cut makes the news
Three things separate a routine fault from a real problem, and none of them is whether the word sabotage appears in the coverage. How many other cables serve the affected route. Whether the fault sits where a coastal state has reason to slow-walk a permit. How long a ship takes to arrive, which is a question about the fleet rather than about the incident itself.
If the answers are several, no and days, the story is maintenance. If they are one, yes and nobody knows, it is worth following.
For a related mechanism, see our explainer on how export controls work.
