Thirteen of India's Eighteen Internet Cables Come Ashore in Six Kilometres. The Rest of the World Is a Graph.

Let me ruin your mental model of the internet in one paragraph. It is not the cloud. It is 574 cables lying on 1.42 million kilometres of seabed, carrying 99% of intercontinental traffic, coming ashore at roughly 1,900 landing stations worldwide. Satellites, the thing everyone pictures when you say "the internet," carry a share that is small and shrinking. Everything else you love about AI infrastructure, the GPUs, the fabs, the gigawatt data centres, is downstream of wires laid by ships, and the number of ships that can repair those wires is genuinely, embarrassingly thin.

I went looking for the India version of this because I live inside it, and the India version is close to criminal. At least 13 of India's 18 intercontinental cables come ashore within six kilometres of one another at Versova beach in northwest Mumbai. Together they carry as much as 95% of India's international bandwidth to Europe, Africa and West Asia. Eleven of the eighteen are more than 20 years old against a nominal 25-year design life. And while this pile of fibre sits in a dense Mumbai neighbourhood, Iranian state-linked media has published detailed maps of the Gulf cable routes and called the regional concentration a "fatal weakness" for the digital economy, while demanding formal sovereign control over the fibre running through the Strait of Hormuz.

This is the same story I have told three times in the AI Pipeline series, except this one has a coordinate instead of a company.

What the internet physically is#

TeleGeography's headline numbers are the least visible, most load-bearing infrastructure statistics in technology: 574 cables, about 1.42 million kilometres of seabed, around 1,900 landing stations. Hyperscaler investment is now dominant, with Meta, Amazon, Google and Microsoft all putting money directly into cable capacity. Meta's Project Waterworth, announced in February 2025, is being described as the world's longest subsea cable.

Why does nobody budget for this? Because a cable is undifferentiated physical infrastructure with a long design life, a high failure rate (anchors, trawlers, abrasion, seismic events), and a repair supply chain that barely exists. A cable is cheap relative to everything it carries. The only scarce things are the ship and the landing permit.

The Versova problem#

Here is how the BBC put it in September 2026: "One of India's biggest digital vulnerabilities sits on a six-kilometre stretch of Mumbai's Arabian Sea coast." Around Versova, at least 13 of 18 intercontinental cables come ashore within six kilometres of one another. Anwesha Sen of the Takshashila Institution calls it "a critical vulnerability due to severe geographic concentration." If several of those cables were cut simultaneously, either accidentally or deliberately, a bulk of India's westward bandwidth to West Asia, Africa and Europe would be impacted.

The mitigation has to come first, because the doomsaying version of this story is wrong. Cable architecture allows rerouting. India's cables land at Mumbai, Chennai, Kochi, Tuticorin and Trivandrum, so a single cut does not cause a blackout; traffic diverts to the southern cities. I sat with TeleGeography's map and the landing-station coordinates and tried to answer the question nobody has published: what is the actual westward capacity if every Versova cable goes dark at once, and can Chennai, Kochi, Tuticorin and Trivandrum absorb it? The honest answer from public data is that rerouting is real, and the risk that matters is a correlated multi-cable event, not a single cut. That distinction is the difference between a solvable problem and an unmanageable one, and most coverage of this topic refuses to make it.

The ratio that completes a pattern#

Now the number that made me want to throw something. Broadband India Forum's comparison: India consumes roughly 20% of the world's data. India hosts 21 cable landing stations out of roughly 1,900 globally. That is about 1%.

India generates about 20% of the world's data, hosts about 3% of the world's data centres, and now about 1% of the world's cable landing stations. Three different measurements of the same asymmetry, in three different subsystems of the same pipeline. I documented the first two in Pipeline Part 4. The third one arrived in a report and finished the pattern for me. The country that consumes a fifth of the world's data owns a hundredth of the doors it comes through.

The fleet is old, and the ships are few#

The Takshashila paper again: 11 of India's 18 international cables are more than 20 years old, approaching the nominal 25-year operating life. Amajit Gupta, group CEO of Lightstorm, put it plainly: a significant share of India's landing cables are approaching economic end-of-life, and "that concentration is exactly the risk the industry needs to design away from."

But follow Gupta on the risk he separates from sabotage, because he is right about the base rates: "there are cable faults - a far more common risk than deliberate attacks." Anchors and trawlers kill more cables than saboteurs do. Any post that defaults to "attack surface" framing is wrong about how cables actually die, and I am not going to be one of them.

The live case that proves how ambiguous this gets: in August 2026, two cables, Indigo West and Indigo Central, suffered faults "straight after each other and in close proximity" inside the Submarine Cable Protection Zone off Perth. Subco CEO Bevan Slattery pointed to "very suspicious activity from a vessel near the location." The Australian Federal Police is assessing a report. There is no public evidence yet that the cables were deliberately damaged. And that is the entire point: nobody can tell the difference quickly, and the distinction does not matter to the twelve million people whose latency just tripled.

Taiwan: 50 days on microwaves#

If you want one sentence that answers "does this matter," here it is. In February 2023, two PRC-registered vessels severed two undersea cables, cutting internet access for the 13,000 residents of Taiwan's Matsu Islands, who relied on microwave radio transmissions for 50 days until a cable ship completed repairs. Thirteen thousand people lost internet for fifty days because two cables were cut and one ship was available.

The precedent kept repeating. In January 2025, Chunghwa Telecom reported damage to an international cable and identified a suspicious vessel, the China-linked cargo ship Shunxin 39, near the affected area; Taiwan's Coast Guard said the vessel may have been involved in deliberately cutting the cable. These details sit in the legislative record of S.2222, the Critical Undersea Infrastructure Resilience Initiative Act.

The Red Sea version has been running since November 2023. At least 17 submarine cables transit the Bab el-Mandeb corridor, carrying the majority of traffic between Europe, Asia and Africa. Houthi attacks caused multiple cuts indirectly from 2024, and in March 2024, four of 15 oceanic cables in the Red Sea were cut, forcing rerouting of as much as 25% of traffic between Asia, Europe and the Middle East. The detail I cannot get over: the repairs required permits from the Yemeni maritime authority. A geopolitical dependency inside a repair. Somebody had to ask a faction's permission to fix a cable, and the entire internet between three continents was queuing behind that permission.

The Gulf turns the map into a threat#

Per Submarine Networks' World Telecommunication Day 2026 special issue: after threats to map and potentially cut undersea cables, the IRGC-affiliated Tasnim news agency published a detailed map of undersea cable routes in the Gulf, describing the high concentration in that narrow passage as a "fatal weakness" for the region's digital economy. Iranian state-linked media escalated to demanding formal sovereign control over the fibre-optic infrastructure running through the Strait of Hormuz. For the record, and the record matters: officially, no cable has yet been directly severed by military action in the Gulf. Publishing the map is the escalation. The map is the threat.

Governments have finally noticed#

The response cluster is real. The US and allies issued a joint statement on cable security. NATO held its first meeting of its Critical Undersea Infrastructure Network. Australia launched a Cable Connectivity and Resilience Centre. The EU issued a recommendation on submarine cable resilience. And Japan, which found its own landing stations dangerously concentrated in two regions, Chiba and Mie Prefectures, approved a draft framework in June 2026 that does the thing every other government is still talking about: it spends money to geographically distribute.

Japan's I-AM Cable, from NTT Data, about US$1 billion, 320 Tbps design capacity over 260 kilometres, lands in Chiba, Mie and Fukuoka, in service in Q3 2027, explicitly to "disperse international submarine cable landing sites from the Tokyo and Osaka metropolitan areas." The E2A trans-Pacific cable adds landings at Tomakomai in Hokkaido and Itoshima in Fukuoka. And NEC committed over ¥100 billion, about US$636 million, for fiscal 2026 through 2030, including roughly ¥50 billion to secure up to five dedicated cable-laying and maintenance vessels, because a cut you cannot reach is a cut that stays cut.

Japan's framing is the sentence I want carved above every telecom ministry door: the government is moving undersea cables "from a largely private-sector-led domain to a nationally governed strategic" asset class. That is the same move France and Japan made with lithium and rare-earth processing. It is a declaration that a coordinate is now a sovereignty question.

Where this could be wrong#

Concede the largest objection up front: cables are redundant by design, and rerouting works. Except redundancy was not sufficient for Matsu for 50 days, and my claim is about correlated failure and concentration, not single cuts. Concede the next one too: sabotage framing is alarmist, and Gupta is right that faults beat attacks. Perth has no public evidence of deliberate damage, the Shunxin 39 attribution is an allegation by Taiwan's Coast Guard, and the post you are reading is not built on sabotage. Concede the third: part of this is about China, from Bangladesh blocking US-managed cable ships in a dispute over Chinese backbones to the PRC-registered vessels at Matsu. But my claim is about landing-point geography, not nationality, and Japan's own stations are dangerously concentrated in two prefectures. Geography is the part that has a solution. And concede the data issue: TeleGeography's route and age data is substantially paywalled, so the headline numbers here are TeleGeography-derived and sourced through the Takshashila paper and the BBC's characterisation, not a cable-by-cable dataset I hold.

The one fixable chokepoint#

Every layer in my pipeline series ended at a chokepoint that is structural. Lithography ends at ASML in Veldhoven. Fabrication ends at TSMC and its 70% share. Realistically, nothing fixes either. The subsea layer ends at a coordinate, Versova beach, six kilometres, thirteen cables, and unlike the others it is mitigable, deliberately, with money. Japan's I-AM Cable is about a billion dollars to add a landing point. That is the price of the fix.

Which means this is the only chokepoint in this entire body of work with a known mitigation that is not being taken at sufficient scale, for geographic and political reasons rather than technical ones. India generates a fifth of the world's data and owns a hundredth of the doors. The fix is a billion dollars per landing point. The reason it has not happened is not physics. It is a choice.