Technology

What undersea internet cables do, and why India depends on a few of them

Almost everything India sends abroad online travels through glass fibres inside a few dozen cables on the seabed, most of them landing on one beach in Mumbai. Here is how those cables are built, lit, cut and repaired, and why a ship's anchor in the Red Sea can slow your video call to London.

Illustrative cover: What undersea internet cables do, and why India depends on a few of them
Illustration: Pointales

When you load a page hosted in Europe or the United States, the request leaves India through a glass fibre thinner than a hair, inside a cable about as wide as a garden hose, lying on the floor of the Arabian Sea. Submarine cables carry the overwhelming majority of intercontinental internet traffic: TeleGeography puts it at over 99% and the UN’s World Ocean Assessment at 95%. Satellites are a rounding error on that scale. This guide explains how the cables work, where India’s come ashore, why they break and how they get fixed, and what a reader can observe at home.

Why cables, not satellites

Capacity is the first reason. Meta’s 2Africa cable supports up to 21 terabits per second per fibre pair, with as many as 16 fibre pairs in one cable, the company wrote in November 2025. No satellite constellation comes close, and a cable’s capacity can be raised later by changing the equipment at each end. TeleGeography’s FAQ cites US regulator data putting satellites at just 0.37% of all US international capacity.

Latency is the second. Light in glass fibre travels at roughly 200,000 km per second, about two-thirds of its speed in a vacuum, because the fibre’s refractive index is about 1.47 (Dissecting Latency in the Internet’s Fiber Infrastructure, 2018): a round trip of about 10 milliseconds per 1,000 km of cable. A geostationary satellite adds half a second from altitude alone. Low-orbit satellites fix that but share each beam among everyone under it, which is why our guide to how satellite internet works frames them as the option for places fibre cannot reach, and a backup when cables are cut.

What a cable actually is

For most of its length a submarine cable is 17–20 mm across, “the width of a typical garden hose”, according to the UN’s World Ocean Assessment (2016). Nearer shore, where ships and trawlers are, extra steel armour takes it to 28–50 mm, and in water shallower than about 1,000–1,500 m it is usually buried by a plough towed behind the cable ship.

Layer, from the centre outWhat it does
Glass fibres, in pairsCarry the light. Each pair is one send-and-receive channel; new cables have up to 16 pairs (Meta), and 16–24 is now typical (TeleGeography)
Copper conductorFeeds high-voltage direct current from the landing stations to the repeaters along the route
Polyethylene insulationKeeps the current in and the sea out
Steel wires and outer sheathStrength for laying and recovery; heavier armour near shore

Light fades as it travels, so every few tens of kilometres the cable passes through a repeater, a sealed cylinder of optical amplifiers powered through the copper conductor from shore at thousands of volts. TeleGeography’s September 2025 piece on the economics of submarine cables prices cable at US$6,000–20,000 per kilometre, repeaters at about US$200,000 each “with dozens required for transoceanic routes”, and a 7,000 km transatlantic system at around US$250 million. Cables are designed for a minimum life of 25 years, TeleGeography notes, but are often retired earlier because newer systems carry far more per rupee.

How the data is “lit”

A fibre pair’s capacity is not fixed when the cable is laid. Each fibre carries many colours (wavelengths) of laser light at once, a technique called dense wavelength-division multiplexing (DWDM), and the terminal equipment on shore decides how many wavelengths run and at what speed. Upgrading a cable usually means replacing those terminals, not the cable, which is why TeleGeography distinguishes potential capacity from lit capacity, what is actually switched on. The newest cables add spatial-division multiplexing (SDM): more fibre pairs, each run at lower power, which the limited electricity from shore can support. Meta says this is how 2Africa doubles the pair count of older systems.

The appetite behind these designs is the cloud. TeleGeography estimates that content and cloud companies (Google, Meta, Amazon, Microsoft) now consume about 75% of international bandwidth, much of it moving data between the data centres that train and serve large language models, the same buildings our explainer on GPUs and why AI needs so many describes.

Landing stations and who owns the cables

The cable comes ashore through a buried duct to a cable landing station (CLS), a secure building holding the power-feed equipment, the terminals that turn light back into data, and the links to the country’s terrestrial fibre. In India a CLS can be set up only by a holder of an international long-distance (ILD) licence with the Department of Telecommunications’ approval, and TRAI regulates access to the station’s “essential facilities” so that other operators can buy capacity on the cables landing there (TRAI recommendations, 19 June 2023).

Ownership follows two models. In a consortium, a dozen or more carriers share the cost and the fibre pairs: SEA-ME-WE 6, a 21,700 km Singapore–France system with 126 Tbps of design capacity, has 14 members including Airtel (Developing Telecoms, 18 February 2025). A private cable is built by one content company or carrier. Capacity is then sold as short leases or as “indefeasible rights of use” lasting 10–20 years, TeleGeography explains, which is how an Indian ISP that never touched a cable ship can own capacity to Marseille.

Where India’s cables land

Detail (with source and date)
Cables landing in India17 at the time of TRAI’s June 2023 recommendations; 18 as of 2024, 11 of them more than 20 years old (Takshashila, July 2026)
Landing citiesMumbai and Chennai have “the maximum concentration” (TRAI); Takshashila also lists Cochin, Tuticorin and Thiruvananthapuram, with new stations planned at Digha, Bhubaneswar, Puducherry and Visakhapatnam
The chokepoint on land“At least 13 of these cables land within a six-kilometre stretch at Versova beach, Mumbai” (Takshashila)
Recent international landings2Africa Pearls at Airtel’s Mumbai station, March 2025, 100 Tbps (DCD); SEA-ME-WE 6 at Airtel’s Mumbai and Chennai stations, due in service in 2026
Jio’s own systemsIndia-Asia-Xpress (Mumbai and Chennai to Singapore) and India-Europe-Xpress (Mumbai to Savona, Italy), announced May 2021, SubCom as builder, “more than 200 Tbps” over 16,000 km (Submarine Networks)
Government island cablesChennai–Andaman & Nicobar: 2,250 km, four fibre pairs, inaugurated 10 August 2020, 2 × 200 Gbps to Port Blair, 205 Gbps in use by December 2024 (PIB, DoT); Kochi–Lakshadweep: 11 islands, inaugurated 3 January 2024, from 1.7 Gbps of satellite capacity to 200 Gbps (PIB)

Geography does the rest. From Mumbai, almost every path to Europe runs up the Red Sea, past Yemen at Bab el-Mandeb and across Egypt into the Mediterranean; about 17 cables use the Red Sea (Rest of World, March 2026). From Chennai, the path east runs through the Malacca Strait to Singapore, the region’s exchange point. West carries Europe, Africa and most US traffic; east carries Asia and the Pacific crossings.

How cables break, and how they are fixed

Faults are routine. The International Cable Protection Committee counts about 150–200 submarine telecom cable faults a year worldwide, stable even as the network grew to roughly 1.7 million km in 2025. Around 70–80% are “accidental human activity, such as commercial fishing activity and ships’ anchors”; natural hazards and equipment failure make up most of the rest, and deliberate damage is under 1% in a typical year. Sharks caused zero faults between 2007 and 2014, per TeleGeography.

A repair follows the sequence that Stephen Drew of cable-maker TE SubCom laid out at a 2011 industry workshop:

  1. Locate the fault from shore. Instruments at the landing station estimate the distance to the break; confirmation “may take hours or days”, and the cable may have been dragged off its charted route.
  2. Mobilise a ship. Owners pay for maintenance agreements that guarantee a cable ship within 24 hours of a call. India has none of its own: TRAI noted in 2023 that vessels come from Dubai or Singapore, adding “10–12 additional days (both way) transit time”; Takshashila counted about 61 repair ships worldwide as of 2022.
  3. Get permits. In some waters none are needed; in others “a repair permit takes 1–6 weeks or more”.
  4. Cut, recover, splice, relay. The ship drags a cutting grapnel across the seabed, hauls up one end, tests and buoys it, recovers the other end, splices in new cable and lays it back down.

Weeks is normal; months happen. Cloudflare reported that three Red Sea cables cut on 24 February 2024, believed snagged by the anchor of the missile-struck cargo ship Rubymar, remained unrepaired on 13 May 2024. Takshashila’s estimate for a repair in Indian maritime zones is four to five months.

What the Red Sea has done to India’s internet

Users rarely notice a single fault because operators spread traffic across several cables, what TeleGeography calls “safety in numbers”. The trouble comes when several cables in one corridor fail together.

On the night of 5 September 2025, the SEA-ME-WE 4 and IMEWE systems lost part of their capacity in the Red Sea. Cloudflare’s probes recorded mean latency from New Delhi to London rising by about 20% and from Mumbai to Frankfurt by about 30%, starting around 23:00 UTC (Cloudflare Q3 2025 disruption summary). Nothing went dark; traffic took longer paths and pages on European servers loaded more slowly. That is a well-designed network under stress: slower, not off.

The 2026 conflict around Iran raised the stakes. Rest of World reported on 4 March 2026 that both the Strait of Hormuz and the Red Sea were “effectively closed to commercial traffic”, though the cables were not themselves targeted. Twelve days later, Capacity Media reported that Alcatel Submarine Networks had declared force majeure and stopped work in the Gulf, leaving its cable ship Île de Batz laid up near Dammam and the 2Africa Pearls extension, due in service later in 2026, unfinished. Cable ships cannot repair, or finish laying, what they cannot safely reach.

Redundancy for India therefore means more cables on the west route, so one anchor cannot remove a large share of capacity; more east-route capacity via Singapore, so Europe-bound traffic can go the long way round if it must; and more landing points outside Versova. Takshashila’s July 2026 paper proposes that no single location should carry more than 40% of India’s intercontinental bandwidth by 2030.

Regulation and security, in plain terms

TRAI’s recommendations of 19 June 2023, summarised by PIB the next day, are the current policy map: “Essential Services” status for landing stations and cable maintenance, notification as critical information infrastructure, two tiers of landing station (a main CLS that handles all clearances, and lighter “CLS points of presence” inside data centres), a committee to find a viable model for Indian-flagged repair vessels, cable depots on both coasts, and fewer repeat security clearances for repair crews. TRAI’s own status table marks the package “Partially Accepted”. Takshashila, citing September 2025 rules for captive telecom services, argues the critical-infrastructure designation and an Indian-flagged repair ship in the Arabian Sea by 2027 remain open.

The security questions are physical. A cable can be tapped only with access to the fibre or the station, which is why stations are licensed, guarded and fitted with lawful-interception equipment, and why TRAI wants the party seeking a landing permit to prove it owns the asset in Indian waters. The more practical risk, as every incident above shows, is a dragging anchor in a narrow strait.

What you can observe at home

Median round-trip times between Microsoft’s own Azure data centres for the 30 days to 30 July 2026 (Azure latency table) show the geography. A home connection adds its own tens of milliseconds, but the pattern holds.

From Central India (Pune) toMedian round tripStraight-line distance, roughly
UAE North (Dubai)32 ms1,900 km
Southeast Asia (Singapore)53 ms3,900 km
East Asia (Hong Kong)87 ms4,700 km
West Europe (Netherlands)139 ms6,900 km
UK South (London)141 ms7,200 km
East US (Virginia)201 ms13,000 km

Singapore is closer in time than its distance suggests because the cable path is nearly direct; Europe is further because the path detours through the Red Sea and Egypt and several landing stations. A page served from Virginia costs a fifth of a second before any computing happens, which is why global services place servers, and increasingly AI inference, inside India. The distances and light-speed arithmetic are Pointales’ own; the measured figures are Microsoft’s.

Outages are regional by construction: a Red Sea fault slows traffic to Europe and the US east coast while Singapore-hosted services are untouched, and a Malacca-side fault does the reverse. Content delivery networks hide most of it by serving cached copies from Mumbai or Chennai, which is why a cable cut feels like “video calls are laggy” rather than “the internet is down”.

Common misunderstandings

  • “Satellites will replace cables.” They complement them: satellites win on reach, cables on capacity and latency, by orders of magnitude.
  • “Cables are cut by sharks or spies.” Anchors and fishing gear cause most faults; deliberate damage is under 1%.
  • “India has dozens of landing cities.” It has a handful, and one beach in Mumbai carries most of the west-bound traffic.
  • “More cables mean faster internet.” New cables add capacity and resilience; latency is set by the route and the speed of light in glass.

The point: India’s international internet rides on roughly 18 submarine cables, most of them coming ashore within a few kilometres of each other in Mumbai, and almost all of the west-bound ones threading the Red Sea. The cables are mature engineering: hair-thin fibres, copper to power the repeaters, steel where ships are. What is fragile is the geography and the repair chain, and the 2024–26 Red Sea and Gulf disruptions showed both. The fixes are known: more landing points, more east-route capacity, and a repair ship that does not have to sail from Dubai.

Sources

Chander Prakash

Chander Prakash

Chander Prakash is the founder and editor of Pointales. He reviews every story before it is published and sets the publication's editorial standards, with a focus on clear, well-sourced explanations of business and technology.