The cloud is glass on a seabed.

Everything has an address

Satellites feel like the obvious answer. They are visible, modern, and the word "cloud" points upward. But the overwhelming majority of intercontinental data — the figure usually quoted is around 99 percent — does not go up at all. It goes down, into the sea.

When you open a website hosted in Frankfurt from a café in Phu Quoc, your request does not dissolve into an abstraction. It travels a specific, traceable, physical path: a tower, a terrestrial fibre run, a building on a coastline, and then several thousand kilometres of cable lying on a seabed. Every hop has a street address and an owner.

There is no cloud. There is only someone else's computer, and a very long piece of glass that reaches it.

The uncomfortable version

This is not pedantry. Almost every strange behaviour of international connectivity — latency that will not improve, outages that hit whole regions at once, prices that vary by geography — becomes obvious the moment you stop thinking of the network as weather and start thinking of it as plumbing with a map.

What a submarine cable actually is

For most of its journey across an ocean, a modern cable is about as thick as a garden hose. That is the first surprise. The second is what is inside: the glass fibres that carry every bit of it are each roughly the diameter of a human hair, and a whole system may contain only a couple of dozen of them.

Everything else in the cable exists to protect those hairs, or to feed them power.

Light loses strength over distance, so every 50 to 100 kilometres the signal passes through a repeater — a sealed cylinder containing optical amplifiers. There is no power station in the middle of the Pacific. Every one of those amplifiers is fed direct current sent thousands of kilometres from the shore, at voltages high enough to be lethal.

Why a cable cannot simply hold more fibre

The intuitive way to increase capacity is to add fibres. Glass is cheap; a hair-thin strand costs almost nothing. Why do systems typically carry between eight and twenty-four fibre pairs rather than hundreds?

Because of the repeaters. Each fibre pair needs amplification, every amplifier needs power, and all of that power must arrive down a single copper conductor from land. Push past roughly two dozen pairs and you cannot deliver enough electricity along the length of the cable to run them. The ceiling on the internet's busiest routes is not optical. It is electrical.

Landing stations, and the places everything narrows

A cable is useless until it comes ashore. It does so at a landing station — typically an unremarkable low building near a beach, heavily secured, where the submarine cable terminates and joins the terrestrial network. There are close to nineteen hundred of them worldwide.

These buildings are the point where the physical internet becomes national again. A landing station sits in a country, under that country's licensing regime, subject to its law enforcement, its outages and its politics. The same sovereignty that governs mobile spectrum governs the coastline.

And the routes are far less distributed than the map suggests. Geography funnels cables through a handful of narrow passages:

• The Red Sea & Egypt
  A very large share of Europe–Asia traffic passes through this single corridor, with terrestrial crossings between the Mediterranean and the Red Sea. Damage here is felt across two continents at once.
• The Luzon Strait
  Between Taiwan and the Philippines — a dense bundle of trans-Pacific and intra-Asian systems, in a region prone to typhoons and undersea landslides.
• The Strait of Malacca & Singapore
  The maritime funnel of Southeast Asia, and consequently one of its most important digital ones. Heavy shipping traffic makes it a persistent fault zone.

A redundancy plan that routes traffic through three different cables which all pass through the same strait is not redundancy. It is one cable wearing three names.

It is almost always an anchor

Cables break regularly — something in the order of a hundred to two hundred faults a year, globally. The popular explanation involves sharks. The real one is duller and far more common: fishing gear and ship anchors, in shallow coastal water, account for the large majority. Most damage happens within sight of land.

1. DetectionOperators see the fault instantly, and can locate it along the cable's length with considerable precision from the signal itself.
2. DispatchA specialised repair ship is assigned. The global fleet is small and ageing, and vessels are shared across regions — so this step can take days before anyone leaves port.
3. RecoveryThe ship grapples the seabed, hooks the cable and lifts one severed end to the surface. In deep water this alone can take a day.
4. SpliceThe break is cut out and a fresh section spliced in, fibre by fibre, on the deck of a moving ship.
5. ReturnThe repaired cable is lowered back and, where required, reburied.

The whole process is measured in weeks, not hours — and that is with cooperative weather and available permits. Repairs in contested or restricted waters have taken far longer, because a ship cannot begin work until it is authorised to be there.

Who pays for the ocean floor now

For most of the cable era, systems were funded by consortia of telecom carriers — a dozen national operators splitting the cost and the capacity. That model built the internet's first global routes.

It is no longer the whole picture. Over the last decade the largest content and cloud companies became major cable investors in their own right, funding systems outright or taking substantial stakes, because their internal traffic between data centres now justifies owning the road rather than renting it.

The consequence worth noting: the physical internet is increasingly shaped by where the world's computing is concentrated rather than where its people are. That is not automatically bad, but it is a different logic from the one that laid the first cables.

Latency is geography, and geography does not negotiate

Light travels slower in glass than in vacuum — roughly two-thirds the speed. This gives every route a hard, permanent floor beneath which no amount of engineering can pass.

Mumbai to London and back is a journey of many thousands of kilometres each way, and the round trip cannot be completed in less than roughly a tenth of a second. No amount of money removes that. You can add capacity, you can add cables, you can compress and cache and pre-position — but you cannot shorten the planet.

This is why the location of a server is not an implementation detail, and why a traveller whose traffic is routed back to their home country feels it. Distance is the one cost in networking that is genuinely non-negotiable.

Why any of this matters above the water line

It would be easy to read all of this as trivia. It is not, and the reason is this: the physical layer is the most global part of the internet, and it stops being global at the shore. Glass crosses oceans indifferently. The moment it lands, it enters a jurisdiction — a licence, a regulator, a national operator, a set of commercial agreements.

That pattern repeats at every layer above it. The cable is international; the landing station is national. The radio spectrum is physics; the licence to use it is sovereign. The internet is borderless; the connectivity that reaches a traveller is sold country by country, by companies that were licensed there.

Which is the honest reason a tourist landing in a new country still struggles to get online in a world with a million and a half kilometres of cable already in the sea. The capacity has been there for years. What is missing has never been the glass. It is everything built on top of it.