Why fiber-optic beat copper for long distances

Copper carries electrons; fiber carries photons. The reasons one wins over kilometers come down to physics — how fast the signal fades, how much room there is around the carrier to put signal in, and the fact that light doesn't care about your neighbor's microwave.

The picture version

Five pictures for a reader who has only ever plugged in a cable. The prose below fills in the seams the pictures skip.

Why it exists

At some point you’ve tried to run an Ethernet cable somewhere it didn’t want to go — across a house, out to a garage, down to a basement — and found out the hard way that the spec gives up at 100 metres. Meanwhile, the video call that cable carries crosses an ocean and back with no trouble at all. The cable to your desk can’t manage a football pitch; the link to another continent doesn’t blink. The difference isn’t budget. It’s that one of them is copper and the other, for almost its entire length, is glass.

For about a century, a wired long-distance signal meant a copper wire. Telegraph, telephone, then coaxial cable for TV and the early internet backbones — all electrons moving down metal. (Microwave relay towers and, later, satellites carried plenty of long-haul traffic too; the point is that the cable was always copper.) The whole industry was tooled for it.

Then, over a few decades, the long-haul backbone of the internet quietly switched. Today, essentially every transoceanic link, every cross-continental trunk, and most of the fiber pulled into apartment buildings is glass, not copper. Copper still wins in the last few meters — your laptop’s USB cable, the patch cable behind your desk, the power line — but the moment a signal needs to go more than a few hundred meters at any serious bandwidth, it goes through a laser and a strand of glass.

The interesting question isn’t “is fiber faster?” It’s why physics forces this. Copper didn’t lose because the cable industry got lazy. It lost on three specific limits where glass’s numbers are better by orders of magnitude — not because glass has no limits, but because its limits sit somewhere far more convenient.

Why it matters now

Every API call you make, every model weight downloaded from a hub, every video frame streamed to a phone — almost all of it crosses fiber somewhere along the path. The reason datacenters cluster where they do, and the reason cloud regions feel “close” or “far,” is shaped in part by where the fiber already runs and what it cost to put there — alongside power, land, and cooling. When a hyperscaler announces new capacity between two regions, the thing being added is optical: more wavelengths lit on an existing pair, a newly lit pair, or a new cable. Nobody’s answer to that problem is copper.

It also explains a recurring engineer question: “why can’t I just run a really long Ethernet cable?” The answer is the same physics that killed copper trunks.

The short answer

fiber-optic = glass waveguide + laser + photodetector

Picture to keep: a flashlight shone down a very long, very clear pipe, versus a shout passed down a line of people. The light arrives dimmer but intact; the shout gets garbled by every person who repeats it and by every other conversation in the room.

A fiber link encodes bits as pulses of light, sends them down a hair-thin strand of very pure glass that traps the light by total internal reflection, and recovers them with a photodiode at the far end. Copper encodes bits as voltage changes on a metal conductor, and the receiver has to recover them from a smeared, noisy waveform — amplifying, equalizing, recovering the clock, and thresholding. The medium is most of the story — glass loses far less signal per kilometer, carries far more bandwidth, and ignores almost all the electrical noise that copper picks up.

How it works

Start from your too-long Ethernet cable and try to fix it. Each fix runs into the next wall, and the walls are what force glass.

Naive attempt: just run more copper. Longer cable, thicker conductors, a bigger amplifier at the far end. This is exactly what the industry did for a century, and it works — up to a point. Three separate physical effects set that point, and none of them yield to spending more money on copper.

1. Attenuation — how much signal you lose per kilometer.

Every medium absorbs and scatters some of the signal as it propagates. Copper’s loss comes from resistive heating, from the insulation absorbing energy, and, at high frequencies, from the skin effect: the higher the frequency, the worse it gets. That’s why old coaxial cable runs needed amplifiers every kilometer or so, and why high-speed Ethernet over copper caps out around 100 m before the spec gives up.

Glass, in the wavelength windows used for telecom (around 1310 nm and 1550 nm), is astonishingly transparent. Modern single-mode fiber loses on the order of 0.2 dB per kilometer at 1550 nm. That means a signal can travel tens of kilometers before it needs help, and even then the help is often an EDFA: a stretch of doped fiber pumped by a laser that amplifies the light as light, no electronic round-trip.

2. Bandwidth — how many bits you can pack per second.

What limits a channel isn’t the carrier frequency itself but how much spectrum around it the medium will carry usably — and a higher carrier makes far more of it available in absolute terms. Copper trunks pushed gigahertz-class signals; optical telecom uses light around 200 terahertz, and the low-loss windows in silica are themselves terahertz-wide. That’s roughly five orders of magnitude more carrier, and the usable band grows with it — not in strict proportion, since loss, dispersion, and nonlinear effects decide where the window actually ends. Combined with WDM, a single modern fiber pair carries many terabits per second by running dozens of wavelengths in parallel down the same glass.

The exact records keep moving and depend on the system, so a specific peak number here would be stale by the time you read it. The shape of the gap — orders of magnitude, not percent — is the durable claim.

3. Noise immunity — what does the environment do to your signal.

Copper is an antenna. It picks up EMI from anything radiating nearby, and it radiates back. That’s why ethernet pairs are twisted, why shielded cables exist, and why running data cables next to power runs is a recipe for grief. It’s also why long copper runs have ground-loop problems: the two ends of the cable can sit at different ground potentials, and current flows through the shield.

Glass is a dielectric. It does not couple to electromagnetic fields in any practical way at the energies involved. Electrically, a fiber doesn’t care that it’s running next to a 13.8 kV power line, doesn’t have ground loops, and doesn’t leak the signal as radio noise. For the same reason, fiber is much harder to passively eavesdrop on — you have to physically tap the glass or bend it enough to leak light.

And then glass has its own wall. Fiber isn’t free of effects. Different wavelengths travel at slightly different speeds in glass (chromatic dispersion), and in multimode fiber different ray paths arrive at slightly different times (modal dispersion). Both smear pulses out and limit how fast you can clock the link before bits overlap. Single-mode fiber and dispersion-compensating modules exist precisely to manage this. It’s a real constraint, just one that turns out to be much more tractable than copper’s loss curve.

Why copper still wins in the last meter. Glass is brittle, hates tight bends, and requires a laser plus a photodiode at each end. Copper is cheap, flexible, terminates with a crimp tool, and carries power as well as signal — which is why PoE exists, and why the cable to your monitor is still copper. Fiber’s advantages only start paying for themselves once the run is long enough that copper’s losses dominate the bill of materials.

The flashlight-in-a-pipe picture is right about loss and wrong about one thing worth knowing: light in a fiber isn’t bouncing around freely, it’s confined — the core is doped to have a slightly higher refractive index than the cladding around it, so light travelling down the core at a shallow enough angle is totally internally reflected back in rather than escaping. That confinement is why the pipe can be bent around a building and still deliver a beam.

So: you started with fiber-optic = glass waveguide + laser + photodetector. What did this post add? — + and the medium is the reason, not the transceiver. Copper didn’t lose because we got better lasers. It lost because a metal wire is simultaneously a resistor, an antenna, and a low-pass filter, and glass is none of those things.

Famous related terms

• Single-mode fiber — single-mode = thin core + one light path — small enough core (~9 μm) that only one spatial mode propagates, killing modal dispersion. The default for long-haul.
• Multimode fiber — multimode = wider core + many light paths — cheaper transceivers, shorter reach. Common inside datacenters.
• WDM / DWDM — WDM = many wavelengths + one fiber — multiplex independent channels by color. The reason a single fiber pair under the ocean carries the traffic of a country.
• EDFA — EDFA = doped fiber + pump laser — amplifies light optically, so long-haul links don’t need to convert back to electronics every span.
• Skin effect — skin effect = high frequency + current crowding — why copper’s loss gets worse the faster you try to clock it.

Going deeper

• The 2009 Nobel Prize in Physics documentation on Charles Kao — for “who worked out that glass could be made transparent enough to be a cable, and when?”, which is the moment the whole industry became possible.
• Neal Stephenson, “Mother Earth Mother Board” (Wired, 1996) — for “what does it actually take to lay one of these cables?”, a long, funny, reported piece that makes the physics concrete by following a cable ship. Dated on the technology, unmatched on the feel of it.
• Rabbit hole: TeleGeography’s public submarine cable map — for “where is all this glass, physically?”, which turns the argument into a map you can scroll. (For the equations rather than the atmosphere, Govind Agrawal’s Fiber-Optic Communication Systems is the standard textbook.)