A signal can cross 13,000 kilometres without an electrical amplifier acting on the data because a chain of erbium-doped fibre amplifiers, or EDFAs, boosts the light directly. With submarine repeaters commonly spaced about 60 to 80 kilometres apart, a route that long may pass through roughly 160 to 220 optical amplifiers before the light reaches the far shore. The repeaters still need electrical power, but they do not convert each data channel into voltage, reconstruct its bits, and launch it again.
The glass itself is extraordinarily small. A standard single-mode fibre has glass cladding 125 micrometres across, while the light occupies a mode field roughly 10 micrometres wide, according to Corning’s SMF-28 Ultra specifications. The coated fibre is thicker, but the glass strand inside the cable is still far thinner than a dry spaghetti noodle.
That distinction resolves the apparent contradiction in the question. The undersea housings contain electrically powered pump lasers, yet the communications signal remains optical from one landing station to the other. What the ocean route avoids is electrical signal regeneration, not electricity itself.
Why the signal fades in the first place
Even very pure silica absorbs and scatters a small fraction of the light passing through it. Corning specifies maximum attenuation of 0.18 decibels per kilometre at 1550 nanometres for SMF-28 Ultra, a wavelength used because silica loss is especially low there. At that rate, an unamplified signal falls to about half its original optical power after roughly 17 kilometres.
Across 13,000 kilometres, 0.18 dB per kilometre would amount to about 2,340 dB of cumulative loss if nothing restored the power. That is not a merely weak signal waiting for a sensitive detector. It is a signal extinguished long before the opposite coast.
Early optical submarine systems dealt with this loss using regenerative repeaters. Each repeater detected the incoming light, processed the signal electrically, and drove another transmitter for the next span. That architecture worked, but the submerged equipment was tied much more closely to the transmission format and rate designed into the system.
What two laboratories demonstrated in 1987
In January 1987, Robert Mears, L. Reekie, I. M. Jauncey and David Payne of the University of Southampton reported an erbium-doped fibre amplifier operating at 1.536 micrometres. Their Optical Fiber Communication Conference paper described a single-pass gain of 26 dB, or roughly a 400-fold increase in optical power.
Later that year, Emmanuel Desurvire, J. R. Simpson and P. C. Becker at AT&T Bell Laboratories published a separate high-gain erbium-doped travelling-wave amplifier. Their room-temperature device produced 22 dB of gain at 1.53 micrometres and also documented the noise added by amplified spontaneous emission. The two records should not be collapsed into one paper or one laboratory.
The amplifier works through stimulated emission, the same physical process used in a laser. A pump laser, commonly operating near 980 nanometres, raises erbium ions in the doped fibre into excited states. A passing signal photon near 1550 nanometres can then stimulate an excited ion to emit another photon matched to the signal.
Over a short coil of erbium-doped fibre, those events restore the optical power lost in the preceding span. The amplifier does not inspect the message or decide whether a pulse represents a zero or a one. It simply transfers energy from the pump light into the band carrying the data.
Erbium is unusually useful because its amplification band overlaps the low-loss window of silica fibre. With gain-flattening components, one amplifier can lift many wavelength channels at once instead of serving a single channel. That compatibility made EDFAs a natural partner for wavelength-division multiplexing, in which different data streams travel through the same fibre on different colours of infrared light.
What actually sits on the sea floor
A modern submarine repeater is a sealed pressure-resistant housing containing optical amplifiers, pump laser diodes and passive optical components. A detailed NEC technical description shows EDFAs pumped by redundant 980-nanometre laser diodes and designed to operate at depths of up to 8,000 metres. Housing materials and internal layouts vary by manufacturer, so describing every repeater as titanium would be too specific.
Repeaters on long systems are generally installed every 60 to 80 kilometres, according to TeleGeography’s overview of submarine cable construction. The precise span depends on fibre loss, available electrical power, amplifier design and the performance target for the route. A 13,000-kilometre system therefore needs a long cascade, not one extraordinary amplifier.
Power reaches that cascade through the cable’s conductive structure as a constant direct current supplied from landing stations. An NEC paper on submarine power-feeding equipment describes feeding from both ends and notes that long-distance systems can require total feed voltages around 15,000 volts. The high voltage is there to run the repeaters, including their pump lasers.
The data signal nevertheless stays in the optical domain along an ordinary repeatered trunk. It arrives at each EDFA as light and leaves as stronger light, without an optical-electrical-optical conversion in between. That is the defensible meaning of an all-optical amplified link.
The sophisticated decoding electronics can remain at the terminal stations, but the original draft went too far by saying that all electronics stay dry on land. Pump laser diodes are active semiconductor devices on the seabed, and some submerged equipment can include control functions. The important boundary is the signal path, not the physical location of every powered component.
How the bandwidth scales
An EDFA can amplify a band rather than one predetermined bit stream. Engineers divide that band into many wavelength channels, modulate each channel at a high rate, and combine them in the same fibre. The amplifier boosts the collection together, which is why the wet plant can remain useful as terminal technology improves.
Google’s Dunant cable shows how far this approach has scaled. When it became ready for service in February 2021, Google reported a design capacity of 250 terabits per second across 12 fibre pairs between Virginia Beach and Saint-Hilaire-de-Riez. Averaged across the design, that is more than 20 terabits per second per fibre pair.
Dunant also illustrates why newer systems add spatial paths instead of forcing ever more power through one core. Its space-division multiplexing design used 12 fibre pairs and shared pump lasers among multiple pairs, allowing more parallel fibres within the cable’s power budget. More glass strands can be a better bargain than driving each strand deeper into the nonlinear regime.
This transparency has practical consequences after a cable is laid. Operators can replace terminal transponders with equipment using denser modulation, stronger error correction and better digital processing, provided the new signal still fits the wet plant’s optical bandwidth and noise budget. The submerged fibre does not become infinitely upgradeable, but it is less closely tied to one bit rate than an old regenerative chain.
What still limits the journey
Amplification restores power, not purity. Every EDFA adds amplified spontaneous-emission noise, and that noise accumulates as the signal passes through amplifier after amplifier. Eventually the receiver must distinguish a structured signal from a luminous background that the amplifiers themselves have built.
The fibre adds other distortions too. Chromatic dispersion causes different spectral components to arrive at slightly different times, while polarization effects alter how the signal evolves. At high optical powers, the Kerr effect makes the glass nonlinear, so the channels begin to disturb one another.
Coherent receivers and digital signal processing moved much of the compensating intelligence to shore. A 2008 Bell Labs account of coherent detection with DSP describes how a receiver can recover amplitude and phase and compensate for linear distortions introduced during propagation. It cannot remove every impairment, but it can recover information that a simpler detector would lose.
Nonlinearity remains a physical constraint rather than a software bug. Work by René-Jean Essiambre and colleagues on the capacity limits of fibre-optic networks showed how the optical Kerr effect places bounds on information transport. Modern design therefore balances launch power, channel count, amplifier noise, fibre count and electrical power instead of pursuing one unlimited number.
Why the simple version misses the best part
The familiar explanation says that the internet crosses oceans because light travels through glass. That is only the beginning. Without a chain of powered optical amplifiers, even exceptionally transparent glass would lose the signal thousands of kilometres before land appeared again.
An EDFA is not a repeater in the old regenerative sense. It does not decode the bits, identify the traffic or care whether the light represents a video call, a bank transfer or a page request. Its indifference is precisely what lets many wavelengths and changing terminal formats share the same submerged route.
For another view of how materials reshape a signal, see this guide to Wi-Fi dead spots inside a home. At a still larger scale, the account of Voyager 1’s 22-hour radio delay follows a signal whose problem is not glass loss but the spreading of radio energy across interstellar distance.
On the sea floor, the result is a sequence of dark pressure housings separated by tens of kilometres of cable. Inside each housing, electricity drives a pump laser and erbium transfers that energy into passing infrared light. The signal reaches the next housing without ever becoming an electrical message, then repeats the exchange across the ocean until shore electronics finally turn the light back into data.