More than 95 percent of intercontinental data, every video call, cloud backup, and AI inference request that crosses an ocean, travels through roughly 570 fiber optic cables lying on the seafloor. Most executives who rely on this infrastructure daily have never seen a map of it. Fewer still know that the ships capable of fixing it when it breaks number 62, worldwide, and that fleet is aging out faster than it is being replaced.
This article is grounded in current advisory work, not retrospective analysis. Mark Lynd is a 5x CEO/CIO/CISO with Thinkers360 Top 10 global rankings across Cybersecurity and Artificial Intelligence and was ranked #1 globally in Cybersecurity in 2023. He is currently Head of Executive Advisory and Strategy at Netsync, advising enterprise C-Suites and boards on the AI and cybersecurity questions moving fastest in 2026. The frameworks and patterns referenced here are from active engagements this quarter.
I've spent this year advising boards on AI infrastructure risk, and submarine cables come up less than power and water because they feel abstract. A data center's electric bill is a line item someone can point to. A severed cable three hundred miles offshore is not. That gap in visibility is exactly why it belongs on the agenda. The companies moving fastest on AI are also the ones most dependent on cross-border data movement, and that movement runs almost entirely through infrastructure they don't own, can't inspect, and increasingly can't assume is safe from deliberate interference.
The redundancy assumption breaks at the repair layer
The standard reassurance about submarine cables is that there are hundreds of them, so no single cut matters much. That's true for isolated accidents. Fishing trawlers and ship anchors cause the overwhelming majority of the roughly 200 cable faults recorded worldwide each year, and the network usually reroutes around a single break without most users noticing.
The reassurance stops working when faults cluster faster than the world's repair capacity can absorb them, or when the cuts happen in places where getting a repair ship on site is itself a political problem.
The global fleet of cable-laying and repair vessels is 62 ships. That fleet is aging. Nearly half of it is projected to reach the end of its operational life by 2040, while the total length of cable in the water is expected to grow by roughly 48 percent over the same stretch, according to industry projections cited by the Bulletin of the Atomic Scientists. A single specialized repair vessel costs upward of $100 million to build, and the shipyards that build them are booked years out. Redundancy in cable count is not the same thing as redundancy in the capacity to fix what breaks, and the second number is shrinking relative to demand, not growing with it.
Tonga found this out directly. When its single connection to the global network was severed by an undersea volcanic eruption, restoring service took more than a month, because the nearest capable repair ship had to be dispatched from thousands of miles away. That was a natural disaster with no adversary and no ambiguity about intent, and it still took weeks. A deliberately obstructed repair, in contested or heavily trafficked waters, would take longer, and the parties responsible for the damage have every incentive to make that repair as slow and expensive as possible.
Three theaters, one pattern
Three regions have turned this from a theoretical resilience gap into a live geopolitical one, and each shows the same shape. A vessel with no clear reason to be where it is. An anchor dragged for miles rather than dropped. A cut that officials describe as consistent with intent but almost never prosecute as such.
In the Baltic Sea, at least 11 cables and pipelines have been damaged since October 2023. The pattern includes the cargo ship Newnew Polar Bear, linked to the Balticconnector pipeline rupture that October; the bulk carrier Yi Peng 3, whose anchor dragged across two data cables in November 2024; and the tanker Eagle S, which Finnish investigators say cut a power interconnector and four communications cables on Christmas Day 2024 while leaving a 62 mile anchor trail on the seabed. NATO responded in January 2025 by launching Baltic Sentry, adding frigates, patrol aircraft, and naval drones to monitor the seabed, and the Joint Expeditionary Force stood up an AI-driven tracking system called Nordic Warden. Western intelligence officials have told reporters that some of these incidents look more like poorly maintained shadow-fleet tankers dragging anchors than coordinated sabotage. Whether the cause is negligence or intent, the effect on the network is identical, and the ambiguity itself is part of the problem. It is very hard to hold anyone accountable for damage that can always be explained as an accident.
In the Red Sea, four major cables, including SEA-ME-WE 4 and the Europe India Gateway, were cut within weeks of each other in February and March 2024, during a period of intensified Houthi attacks on shipping in the same corridor. Those four systems together carried an estimated quarter of all data traffic between Europe and Asia. Traffic rerouted, but not for free. Latency rose 20 to 40 milliseconds on the surviving direct paths, and closer to 60 milliseconds for traffic diverted around the Cape of Good Hope. For real-time AI inference or financial trading systems, that is not a rounding error.
Near Taiwan, coast guard authorities detained a Chinese-crewed vessel in January 2025 and another in February 2025, each suspected of severing undersea cables connecting the main island to outlying territories. Taiwanese officials have described a pattern of vessels with murky ownership records operating in ways inconsistent with normal maritime traffic near cable routes, consistent with gray-zone pressure that stops short of anything treaties clearly prohibit.
None of this infrastructure is optional for enterprise AI. Training runs move data across borders to reach available compute. Inference traffic for global user bases crosses oceans by default. Multinational cloud replication, the kind every enterprise architecture diagram takes for granted, depends on exactly the cables sitting at the bottom of these three seas.
What a board should actually ask
The scale of financial activity riding on this infrastructure is harder to pin down precisely than the traffic percentage. TeleGeography, the industry's most cited data source on cable capacity, has examined the widely repeated claim that $10 trillion in financial transactions cross submarine cables daily and found it plausible as a floor rather than a precise figure. Their own reconstruction, built from SWIFT message volume and average transaction values, puts the true number higher, but the honest answer is that no one publishes an audited total. Treat $10 trillion as a conservative estimate of scale, not a citation-grade statistic, and treat that uncertainty itself as a sign of how little visibility even sophisticated financial institutions have into a network they depend on completely.
The counterargument, made by cable operators and some maritime security analysts, is that this is overstated. The internet has run on submarine cables for three decades through wars, embargoes, and thousands of accidental cuts, and it has never gone down globally. Redundant routing, they argue, is exactly why isolated incidents in the Baltic and Red Sea caused rerouting and latency, not outages. That argument holds for isolated events. It says nothing about what happens when faults cluster in the same year across three separate theaters while the repair fleet needed to respond to all three simultaneously does not exist. Resilience built for random accidents is not automatically resilience against coordinated or opportunistic pressure applied in more than one place at once.
For a company building AI infrastructure or relying on multinational cloud architecture, the right questions for leadership and the board are specific ones. Which submarine cable routes does our primary AI workload actually depend on, and do we know, not assume. What is our contractual and technical fallback if a route serving a major region degrades for weeks rather than hours. Does our cloud or colocation provider disclose cable-route diversity, or do we just trust that "multi-region" means physically diverse paths. And who owns this risk internally, because right now, for most companies, the honest answer is no one.
The internet has always depended on a small number of ships that most of its users will never see. What has changed is that some of the parties cutting the cables now know that, and are testing what happens when they do.
