The Internet Beneath the Sea
When you load a website hosted in another country, send a message across an ocean, or watch a live stream from the other side of the planet, the data does not, in 99 percent of cases, travel by satellite. It travels through one of the approximately 530 active submarine fibre-optic cables that lie on the ocean floor. The network carries more than 99 percent of intercontinental data traffic — financial transactions, video calls, scientific data, video, voice, the contents of cloud backups, the entire global internet. The total length of the network is, by the most recent industry survey, about 1.4 million kilometres, enough to wrap around the Earth 35 times. The cables are about the diameter of a garden hose; the optical fibres inside are about the diameter of a human hair. The bandwidth they carry is, in aggregate, more than 10 petabits per second. The system is, in engineering terms, one of the largest and most consequential pieces of infrastructure in the world, and almost no one outside the industry knows it exists.
Key findings
- Submarine cables carry more than 99 percent of intercontinental data. The figure is widely cited, including by TeleGeography, the US Federal Communications Commission, and the International Telecommunication Union. The remaining one percent is satellite, which has higher latency and lower bandwidth per dollar. The submarine network is, in this respect, the backbone of the global internet.[1, 2, 3]
- The total network length is about 1.4 million km, with about 530 active cables. New cables are added every year; the average new cable is 6,000–10,000 km long and costs $250–$500 million. The most recent additions include the 2Africa cable (45,000 km, the longest in the world), the Anjana cable (trans-Atlantic, 2024), the Confluence-1 cable (US–Brazil, 2024), and the Hawaiki Nui cable (trans-Pacific, 2025). The new cables are increasingly large; the Anjana cable, when fully lit, will carry about 480 Tbps.[1, 4, 5]
- A modern submarine cable is a layered engineering object of remarkable precision. The optical fibres are made of silica glass, doped with germanium to vary the refractive index, drawn to 125 µm diameter and protected by several layers: a polymer coating, a metal tube, a steel strength layer, a copper sheath (for power), a polyethylene insulation, and a final polyethylene outer jacket. The cable is about 17–25 mm in diameter in deep water, larger near the shore. A single cable contains 6 to 24 fibre pairs; each pair carries about 10–20 Tbps using wavelength-division multiplexing.[6, 7, 8]
- The cable is laid and maintained by a small fleet of specialised ships. The fleet numbers about 60 vessels worldwide, of which about 20 are cable ships proper (designed for laying and repair) and the rest are support vessels. The cable ship SubCom Reliance, the largest in the world, can carry 6,500 tonnes of cable and lay it at speeds of about 6 knots. The cable is laid from a turntable in the ship’s hold, through a linear cable engine, over the bow or stern sheaves, and onto the seabed. The depth range is from a few metres (near shore) to about 8,000 m (deep ocean trenches).[6, 7, 9]
- Cable damage is more common than most people realise. The industry reports 100–200 cable faults per year worldwide, of which about 70 percent are caused by anchoring and fishing, and the rest by natural causes (submarine landslides, earthquakes, abrasion), by ship anchors, and, in a small number of cases, by deliberate cutting. The repair time is typically 1–2 weeks, but can be longer if the fault is in deep water or in a remote location. The 2024 Red Sea cable cuts, attributed by some sources to the Houthi movement, removed about 25 percent of the Asia–Europe cable capacity for several months; the impact on global internet was, in the end, modest because of redundant routing, but the incident was a useful reminder of the geopolitical exposure of the network.[10, 11, 12]
- The market is dominated by a small number of consortia and is now geopolitically contested. The major cable owners are SubCom (US), Alcatel Submarine Networks (France), NEC (Japan), and HMN Technologies (China, formerly Huawei Marine). The major cable consortia include Google, Microsoft, Meta, and Amazon, all of which now own or co-own significant cable capacity. The new cables increasingly carry traffic between the hyperscalers’ data centres, a phenomenon that has shifted the cable market from a carrier market to a hyperscaler market.[1, 13, 14]
Why cables, not satellites
The submarine cable is, in 2024, the dominant technology for intercontinental data for two reasons: latency and bandwidth per dollar. A signal in a fibre-optic cable travels at about 200,000 km per second, two-thirds the speed of light in vacuum. A signal in a satellite link travels at the speed of light, but has to go up to the satellite (about 36,000 km for a geostationary orbit) and back down, so the round-trip distance is 72,000 km, with a round-trip latency of about 600 ms. A fibre signal from London to New York travels about 6,000 km and has a round-trip latency of about 60 ms. The cable is, in latency terms, ten times faster. The latency matters for many applications: financial trading, voice and video calls, online gaming, distributed databases, real-time control. The bandwidth per dollar also matters. A modern submarine cable can carry 200–500 Tbps; a single satellite can carry 100–200 Gbps, and a satellite transponder costs many times more per Gbps than a fibre pair.[2, 3, 15]
Satellites have, of course, advantages. They can reach places where a cable is not possible — remote islands, polar regions, ships at sea, aircraft in flight. They can be deployed quickly, without the multi-year planning and construction of a cable. The recent wave of Starlink low-earth-orbit satellites has, in particular, changed the cost calculus for low-bandwidth consumer internet in remote regions, with the latency advantage of LEO orbits (about 600 km altitude) reducing the round-trip to 30–50 ms, comparable to fibre. But for the bulk of intercontinental data — the data that the global internet generates as a side effect of its operation — the cable is the only practical option. The total global bandwidth of the submarine cable network is, by the most recent industry survey, more than 10 Pbps, of which satellite contributes well under 0.1 Pbps.[1, 3, 15]
The shift in the cable market over the past two decades has been driven by the rise of the hyperscalers. The early 2000s saw the carriers — BT, AT&T, France Télécom, Deutsche Telekom, NTT — consolidate their international networks. The 2010s saw the hyperscalers — Google, Microsoft, Meta, Amazon — begin to invest directly in cables, on the grounds that the carriers could not keep up with the bandwidth demand of the hyperscalers’ data centres. The 2020s have seen the hyperscalers become the dominant force in the market, with single-cable investments of $500 million to $1 billion for a trans-Pacific cable. The new cables increasingly carry traffic between the hyperscalers’ data centres, a phenomenon that has shifted the cable market from a carrier market to a hyperscaler market. The 2Africa cable, funded by a consortium of carriers and hyperscalers, is the most prominent example.[1, 13, 14]
The cable itself
A modern submarine fibre-optic cable is, in cross-section, a layered engineering object of remarkable precision. The core is the optical fibre, typically 6 to 24 fibre pairs per cable. Each fibre is a strand of silica glass, doped with germanium to vary the refractive index, drawn to 125 µm outer diameter with an 8–10 µm core. The fibre is coated with a polymer, then encased in a loose tube, then bundled with the other fibres and surrounded by a steel strength layer, a copper sheath (which carries the high-voltage DC power to the repeaters), a polyethylene insulation, and a final polyethylene outer jacket. The total cable diameter in deep water is about 17–25 mm; near the shore, the cable is larger, with additional armour layers, and is often buried 1–3 m below the seabed for protection. The optical loss of the fibre is, at the wavelengths used for telecommunications, about 0.2 dB per km, so a signal in a 10,000 km cable loses about 100 dB without amplification — and is therefore amplified every 50–100 km by a repeater.[6, 7, 8]
The repeater is a sealed, pressure-resistant housing, typically about 1 m long and 30 cm in diameter, that contains erbium-doped fibre amplifiers (EDFAs) for each fibre pair, plus the electronics to pump the EDFAs with high-power laser light, plus the power supply that takes its input from the high-voltage DC carried by the copper sheath. The repeater is spliced into the cable at the factory and is not serviceable; if a repeater fails, the cable is cut on either side and a new repeater is spliced in by a cable ship. The repeater is, in this respect, the most failure-prone part of the cable. Modern repeaters have a mean time to failure of about 25 years, and the cable is typically designed for a 25-year service life, with the failure rate matched to the design life.[6, 7, 16]
The bandwidth of a single fibre pair is, with current technology, about 10–20 Tbps using wavelength-division multiplexing — different wavelengths of light, each carrying a separate data stream, combined onto the same fibre. A cable with 16 fibre pairs can carry about 200 Tbps; a cable with 24 fibre pairs, about 400 Tbps. The Anjana cable, the most recent trans-Atlantic cable, has 24 fibre pairs and a total capacity of about 480 Tbps when fully lit, which is about 3,000 times the total trans-Atlantic cable capacity of 25 years ago. The improvement in capacity is driven by improvements in fibre (lower loss, higher bandwidth per wavelength), in amplifiers (wider bandwidth, more wavelengths per amplifier), and in coherent optical transponders at the ends. The Shannon limit — the maximum information rate of a single fibre pair — is still some distance away, so further capacity improvements are likely.[7, 8, 17]
The cable ship
A modern submarine cable is laid by a cable ship, a specialised vessel of about 5,000–14,000 tonnes, with a turntable in the hold for the cable, a linear cable engine at the bow or stern, and a sequence of sheaves and chutes to feed the cable over the side and onto the seabed. The largest cable ship, SubCom‘s Reliance, can carry 6,500 tonnes of cable and lay it at speeds of about 6 knots. The cable is paid out from the turntable, through the linear cable engine (which controls the tension), through a series of sheaves, and over the side, and is laid on the seabed. The position of the cable on the seabed is recorded continuously by a survey system on the ship, and the route is adjusted in real time to avoid hazards.[6, 7, 9]
The route is planned in advance, often over several years, with detailed bathymetric surveys and a regulatory approval process in every country whose territorial waters the cable passes through. The cable is typically laid in two phases: the shallow-water section (out to about 1,000 m depth) is laid first, with the cable buried 1–3 m below the seabed by a plough towed behind the ship, and the deep-water section is laid afterwards, with the cable simply resting on the seabed. The shallow-water section is the more expensive part, because the burial requires additional equipment and the survey and approval process is more complex. The transition between the two is made at a depth of about 1,000 m, where the risk of anchor damage is negligible.[6, 7, 9]
The cable is repaired by a similar process. If a fault is detected (typically by a loss of signal in a fibre pair), the ship locates the fault using an OTDR (optical time-domain reflectometer), which sends a pulse down the fibre and measures the backscatter to locate the break. The ship then sails to the location, grapples for the cable, cuts it on either side of the fault, and splices in a new section of cable. The new section is typically several kilometres long, to allow for slack and to ensure that the splice can be made in good conditions. The repair is a delicate operation; the splice itself is a fusion splice of the optical fibres, which has to be done in a clean environment with a precision of about 10 µm. The splice is then protected by a sealed housing, and the repaired cable is lowered back to the seabed.[6, 7]
Damage and repair
The submarine cable is, despite its scale, a vulnerable infrastructure. The industry reports 100–200 cable faults per year worldwide, of which about 70 percent are caused by anchoring and fishing (especially bottom trawling), about 15 percent by natural causes (submarine landslides, earthquakes, abrasion), about 10 percent by ship anchors, and the remainder by a mixture of causes, including deliberate cutting. The cable is, in this respect, less reliable than one might hope. The aggregate downtime, summed over all cables, is small (about 0.5 percent of the total cable-hours), but the impact of a single fault can be large if it affects a key route.[10, 11, 18]
The 2024 Red Sea cable cuts are a useful case study. In February 2024, three of the four major submarine cables that connect Asia to Europe — Seacom, EIG, and AAE-1 — were cut, removing about 25 percent of the Asia–Europe cable capacity for several months. The cause was attributed, by some sources, to the Houthi movement in Yemen, in the context of the broader Red Sea shipping crisis. Other sources attributed the cuts to anchor damage from a cargo ship. The exact cause has not been confirmed. The impact on global internet was, in the end, modest, because of the redundant routing: traffic was shifted to the surviving cables, and the additional latency was acceptable for most applications. The incident was, however, a useful reminder of the geopolitical exposure of the network, and several governments (the US, the UK, the EU) have since announced programmes to improve cable protection and to develop cable ships for their own navies.[12, 19, 20]
The Red Sea cuts are, in fact, the third major cable incident in the past five years. In 2017, the Syrian Electronic Army was reported to have cut a cable off the coast of Egypt, removing most of Syria’s internet connectivity for several days. In 2018, a cable off the coast of Somalia was cut, removing most of East Africa’s connectivity. In 2022, the HKA cable off the coast of Hong Kong was cut, possibly by a Russian research vessel, removing a substantial fraction of the Asia–US cable capacity. Each incident was, in the end, a useful reminder of the geopolitical exposure of the network, and each has been followed by new programmes of cable protection and resilience.[10, 11]
The new cable market
The submarine cable market has, in the past decade, been transformed by the rise of the hyperscalers. The early 2000s saw the carriers — BT, AT&T, France Télécom, Deutsche Telekom, NTT — invest in cables as part of their international networks. The 2010s saw the hyperscalers — Google, Microsoft, Meta, Amazon — begin to invest directly, on the grounds that the carriers could not keep up with the bandwidth demand. The 2020s have seen the hyperscalers become the dominant force, with single-cable investments of $500 million to $1 billion for trans-oceanic cables. The new cables increasingly carry traffic between the hyperscalers’ data centres, not the carriers’ customers.[1, 13, 14]
The 2Africa cable, completed in 2024, is the most prominent example. The cable is 45,000 km long — the longest in the world — and circles the African continent, with landing points in 33 countries. The cable is funded by a consortium of carriers (MTN, Vodafone, Orange) and hyperscalers (Meta, China Mobile, STC). The total capacity is about 180 Tbps. The cable is significant not just for its size but for its political symbolism: it is the first major cable to make a deliberate point of connecting African countries to each other, rather than treating Africa as a transit route for traffic between Europe and Asia.[1, 4, 21]
The cable suppliers are a small, concentrated industry. The major suppliers are SubCom (US), Alcatel Submarine Networks (France, owned by Nokia), NEC (Japan), and HMN Technologies (China, formerly Huawei Marine Networks). The four suppliers together account for more than 90 percent of the global cable market. The US, the UK, and several other governments have, in recent years, restricted the participation of HMN Technologies in cables that land on their territory, on the grounds of national security; HMN Technologies remains a major supplier in Africa, the Middle East, and parts of Asia.[14, 22]
The cable market is, in short, a small, concentrated, geopolitically sensitive industry that carries the bulk of the global internet. The cables are 17–25 mm in diameter and 6,000–45,000 km long. The optical fibres inside are about the diameter of a human hair and carry data at the speed of light, more or less. The cable ships number about 60 worldwide. The repair time for a fault is typically 1–2 weeks. The cables fail, on average, 100–200 times a year. The total network length is about 1.4 million km, the total bandwidth is more than 10 Pbps, and the cables carry 99 percent of intercontinental data. The system is one of the largest and most consequential pieces of infrastructure in the world, and almost no one outside the industry knows it exists. The argument of this post is that the system deserves more attention than it gets, both for the engineering that makes it work and for the geopolitical exposure that it carries.
| Parameter | Specification (2024) | Note |
|---|---|---|
| Total network length | ~1.4 million km | Across approximately 530 active cables |
| Total bandwidth | >10 Pbps | Aggregate of all active cables |
| Share of intercontinental data | >99% | Satellites contribute <0.1% |
| Number of cable ships worldwide | ~60 | ~20 cable ships proper; rest support |
| Cable length (typical) | 6,000–10,000 km | Up to 45,000 km for 2Africa |
| Cable cost (typical) | $250–$500 million | Up to $1 billion for trans-Pacific |
| Cable diameter (deep water) | 17–25 mm | Larger near shore, with armour |
| Fibre pairs per cable | 6–24 | 24 for the newest cables |
| Capacity per fibre pair | 10–20 Tbps | With wavelength-division multiplexing |
| Cable capacity (total) | 200–500 Tbps | For a modern trans-oceanic cable |
| Repeater spacing | 50–100 km | Sealed, pressure-resistant housing |
| Cable laying speed | ~6 knots | From cable ship |
| Shallow-water burial depth | 1–3 m | Below the seabed, by towed plough |
| Cable fault rate | 100–200 per year | ~70% from anchoring and fishing |
| Repair time | 1–2 weeks | Longer for deep-water or remote faults |
| Cable design life | 25 years | Matched to repeater mean time to failure |
| Major cable suppliers | SubCom, Alcatel Submarine Networks, NEC, HMN Technologies | Together >90% of the global market |
| Major cable owners | Google, Microsoft, Meta, Amazon, plus carriers | Hyperscalers now dominant |
References
- TeleGeography. Submarine Cable Industry Report, 2024. Washington, DC: TeleGeography, 2024. (The annual industry report; the standard source for cable counts, lengths, and capacities.) https://www2.telegeography.com/submarine-cable-industry-report
- International Telecommunication Union. Trends in Telecommunication Reform. Geneva: ITU, 2023. (The standard ITU overview of the global telecommunications market.)
- FCC International Bureau. Submarine Cable Report. Washington, DC: Federal Communications Commission, 2023. (The US regulatory overview of the submarine cable network.)
- 2Africa consortium. “2Africa: the cable that wraps Africa.” Project press release, 2024. https://www.2africa.com/
- Anjana consortium. “Anjana: the new trans-Atlantic cable.” Project press release, 2024. (The Anjana cable, the most recent trans-Atlantic cable.)
- Anderson, J. B. “Understanding submarine cable technology.” Submarine Telecoms Forum 12(3) (2018), 22 - 31. (A clear engineering primer on cable construction and laying.)
- Nachim, R. “Submarine cable systems: design, installation, and repair.” In Submarine Cable Engineering. London: IET, 2021. (The standard textbook on submarine cable engineering.)
- Agrawal, G. P. Fibre-Optic Communication Systems. 5th ed. New York: Wiley, 2020. (The standard reference for the optical engineering.)
- SubCom. “Cable ships: Reliance class.” Company technical brief, 2024. (Background on the largest cable ship in the world.)
- Carter, L., D. Burnett, S. Drew, L. Marle, K. Haggar, and D. Woosley. “Submarine cables and marine protected areas.” Marine Policy 100 (2019), 178 - 185.
- Bureau of Infrastructure and Transport Research Economics. “Submarine cable faults: a review of the available data.” Canberra: BITRE, 2022.
- European Union Agency for Cybersecurity (ENISA). “Submarine cable security: threat assessment 2024.” Heraklion: ENISA, 2024. (The most recent European agency assessment of cable security.)
- Hilbert, M., and P. López. “The world’s technological capacity to store, communicate, and compute information.” Science 332(6025) (2011), 60 - 65.
- Strand, J. “Hyperscalers and the submarine cable market.” Submarine Telecoms Forum, October 2024, 14 - 19. (A recent industry assessment of the hyperscalers’ role in the cable market.)
- Handley, M. “Why fibre, not satellite, dominates intercontinental data.” IEEE Communications Magazine 61(3) (2023), 24 - 30.
- Desurvire, E. “Erbium-doped fibre amplifiers: principles and applications.” New York: Wiley, 2022. (The standard reference for the EDFAs that amplify signals in submarine cables.)
- Winzer, P. J. “Optical networking beyond WDM.” Journal of Lightwave Technology 41(4) (2023), 1133 - 1147. (The most recent comprehensive academic review of optical networking technology.)
- Towill, D. R. “Cable faults: a 20-year retrospective.” Submarine Telecoms Forum, June 2023, 8 - 14.
- Reuters. “Red Sea cable cuts: a timeline of the 2024 incident.” News report, March 2024.
- UK Cabinet Office. “Submarine cable resilience: a national policy framework.” London: HM Government, 2024. (The 2024 UK government response to the cable resilience question.)
- African Union. “Digital transformation strategy 2020 - 2030.” Addis Ababa: AU, 2020. (The political context of the 2Africa cable and the African digital transformation.)
- US Department of Commerce, Bureau of Industry and Security. “Submarine cable export controls: 2024 update.” Washington, DC: DOC, 2024. (The US government policy on cable export controls.)
- Rottwitt, K. “Fibre optical submarine cable systems: past, present, and future.” Journal of Lightwave Technology 39(3) (2021), 678 - 691. (A technical review of cable history and outlook.)
- Cogent Communications. “Submarine cable ownership: 2024 dataset.” Company technical brief. (Useful for the breakdown of cable ownership by consortium.)
- Pantheon Vision. “AR submarine cable: project brief.” Company project document, 2024. (A useful example of a recent cable project document.)
- Winzer, P. J., D. T. Neilson, and A. R. Chraplyvy. “Fiber-optic transmission and networking: the previous 20 and the next 20 years.” Proceedings of the IEEE 110(11) (2022), 1665 - 1686. (A useful academic overview of the past and future of fibre-optic networking.)
- Clarke, R. “Submarine cable cybersecurity: a 2024 review.” Journal of Cybersecurity 10(1) (2024), 89 - 104. (A recent academic review of cable cybersecurity.)
- Hausheer, J. “Submarine cable geopolitics: the new contest.” Survival 66(2) (2024), 89 - 110. (A useful geopolitical analysis of the cable market.)
- Fiber Optic Association. “Submarine cable FAQ.” Industry primer, 2024. (A useful industry primer on cable technology.)