In February 2023, two submarine cables serving Taiwan's Matsu Islands were cut six days apart.

Around 14,000 residents lost normal internet access for roughly 50 days.

The islands weren't completely disconnected.

A microwave backup remained.

But bandwidth fell dramatically.

Remote classes became difficult.

Card payments failed.

The network technically continued operating, but the experience of using it moved backwards by decades.

Now compare that with what happened near Jeddah in September 2025.

Three submarine cable systems were cut.

India, Pakistan and the UAE were affected.

Microsoft acknowledged increased Azure latency.

Yet service was substantially restored by the following day.

The cables weren't repaired overnight.

Traffic had simply rerouted around them.

Those two events describe the same physical substrate.

Understanding why they produced such different outcomes is the purpose of Article 6.

We've now examined six physical layers.

For each, we've asked the same three questions:

How concentrated is it?

How critical is it?

How long does substitution take?

Connectivity gives us the best answer we've seen.

Seconds.

When a submarine cable fails, routing protocols move traffic to other available paths automatically.

No factory needs to be built.

No ship needs to be rerouted.

No government intervention is required.

No one even needs to make a decision.

By our original instrument, connectivity should therefore look exceptionally resilient.

And it is.

But the number is misleading.

Because the alternative may begin working in seconds while the original takes weeks to restore.

The system hasn't returned to normal.

It has simply begun consuming its own redundancy.

The article therefore adds a fourth question to the Series II instrument:

How long must the system operate on its alternative before the original capacity is restored?

And how much redundancy remains during that period?

The median restoration time used in our analysis is around 40 days.

That figure carries an important sourcing caveat: the article notes that it comes from a single research citation and that we have not found an independent series against which to verify it. TIC_S2A06_Website_Full

But the analytical distinction does not depend on whether the true median is exactly 40 days.

The difference between automatic rerouting measured in seconds and physical restoration measured in weeks is enormous.

Imagine three independent routes.

One fails.

Traffic moves to two.

The service still works.

But the next failure is now more consequential.

If a second route fails before the first is restored, the network runs on one.

Nothing in the original substitution metric captures this deterioration.

The network can look healthy from the outside while becoming progressively more fragile underneath.

There are around 574 active and planned commercial subsea cable systems spanning approximately 1.42 million kilometres.

They carry about 99% of intercontinental internet traffic.

Satellite systems are important, especially for edge locations and emergency backup, but they remain a small fraction of the bulk capacity provided by fibre.

Cable failures are also routine.

Roughly 150–200 occur every year.

Most are not acts of sabotage.

Between roughly two thirds and four fifths are associated with fishing gear and ship anchors.

This matters because the policy response changes depending on the diagnosis.

If we interpret ordinary failure rates primarily as hostile activity, we fund surveillance.

If the real constraint is restoration, we need repair capacity.

Roughly 60 specialised vessels worldwide perform cable work of some kind.

Fewer than 20 are dedicated repair ships.

Those numbers are not perfectly comparable across sources — another caveat explicitly recorded in the article — but the order of magnitude is clear.

A maintenance fleet measured in the low tens supports the physical layer beneath almost all intercontinental digital traffic.

The fleet is also ageing.

About half of the broader cable fleet and nearly two thirds of maintenance vessels are projected to reach the end of their service lives by 2040.

Around a quarter of installed cable kilometres may require retirement by 2030.

The demand side is therefore growing while a meaningful share of restoration capacity approaches replacement.

Investment in subsea infrastructure has risen substantially.

Our article cites roughly $4–5 billion per year, around twice the level of a decade ago.

But most of that capital goes into new systems.

The reason is straightforward.

A new cable creates additional capacity that can be sold.

A repair ship is valuable precisely when something has gone wrong.

Between faults, much of its economic value consists of being available.

That is a poor commercial profile.

The pattern is familiar from the other substrates we've examined:

Markets are often better at financing growth than resilience.

This is why our three-year forecast assigns a 70% probability to at least one new government-funded or government-mandated cable repair vessel programme being announced in Europe, Japan or the United States by the end of 2029.

Not because governments necessarily operate infrastructure better.

Because standby capacity with high social value and weak private economics is exactly the kind of capability that eventually attracts public procurement.

This is where the analysis becomes more interesting.

The engineering of a repair isn't necessarily what consumes most of the time.

Once the vessel is on station with the right equipment and cable, locating the fault, recovering the cable, splicing it and testing the connection can take days.

The delays often happen elsewhere.

The vessel may be repairing another fault.

It may be thousands of kilometres away.

Weather may prevent work.

Security conditions may make a location inaccessible.

And government permission to enter territorial waters can take a month or more.

That gives us one of the clearest examples in Series II of a cheap institutional intervention outperforming an expensive physical one.

Governments could negotiate expedited repair access before a fault occurs.

A pre-agreed regional clearance mechanism costs almost nothing compared with commissioning another specialised vessel.

Yet it could reduce restoration time materially.

This is not an engineering solution.

It's an institutional one.

And that's precisely why conventional infrastructure analysis can miss it.

There is another weakness hidden by the aggregate numbers.

Cable systems converge.

First, geographically.

The same narrow corridors repeatedly appear:

the Red Sea and Bab el-Mandeb;

the Luzon Strait;

the Strait of Malacca;

the approaches to Egypt.

In February 2024, one dragging anchor damaged three cable systems in the Red Sea.

Roughly a quarter of internet traffic between Asia, Europe and the Middle East was disrupted.

Three nominally separate assets shared one physical risk.

Second, cables converge at landing stations.

Suitable coastlines are limited, and terrestrial backhaul already exists in particular places.

Third, ownership is concentrating.

The article cites industry estimates suggesting hyperscale technology companies now control around 71% of global subsea capacity and approximately 84% of trans-Atlantic bandwidth demand.

Those figures are industry compilations rather than audited disclosures, another limitation explicitly recorded in the article. TIC_S2A06_Website_Full

But the structural implication is important.

A business can buy connectivity from two providers and believe it has diversified.

Yet both providers may ultimately depend on infrastructure owned by the same firms or passing through the same corridor.

We saw the same pattern in Article 5.

Diversity at the invoice. Convergence two tiers underneath.

At the core, between major economies, the network is densely meshed.

Several cable failures can produce little more than latency.

At the edge, a territory may depend on one or two physical systems.

There, redundancy may be better described as a spare.

This is why the Matsu and Red Sea incidents aren't contradictory.

The core absorbed multiple failures.

Matsu lost two routes and spent around seven weeks on a thin microwave backup.

Both statements about global connectivity are true.

But neither average tells you what will happen at a particular location.

This is why connectivity resilience should increasingly be measured at the place, not simply at the national or global level.

For most people living in well-connected parts of Europe and North America, very little.

The network is highly resilient.

But if you live on an island, in a remote territory or somewhere dependent on very few international connections, learn how many physical cable routes actually serve your location.

If connectivity supports your livelihood, prepare for the realistic failure mode.

Not necessarily no internet.

Bad internet for weeks.

Maintain offline copies of essential documents.

Have a payment method that doesn't depend entirely on continuous connectivity.

Know how you'll communicate if bandwidth becomes extremely limited.

Satellite connectivity can improve this situation substantially.

But it should not yet be treated as a full substitute for bulk fibre capacity.

Ask your providers which physical cable systems actually carry your traffic.

Then ask:

Which corridors do they share?

Which landing stations?

Which underlying infrastructure owners?

Provider diversity is not necessarily physical diversity.

Then change the continuity test.

Most organisations simulate a total connectivity failure.

But networks are specifically designed to prevent that outcome.

The more realistic scenario is prolonged degradation:

higher latency;

lower bandwidth;

partial synchronisation;

timeouts;

retry storms.

Systems that behave predictably when a link is either healthy or dead can behave unpredictably when it spends weeks somewhere in between.

That is the state worth testing.

Separate cable construction from cable repair.

They are different economic systems.

New cable capacity has customers, contracts and visible growth.

Repair capacity is standby infrastructure with an ageing fleet and weak commercial incentives for expansion.

That is where public intervention becomes increasingly plausible.

Also track actual vessel orders rather than government strategies.

A strategy isn't a ship.

And specialised shipbuilding lead times mean programmes announced today may not materially expand the fleet for years.

Our five-year forecast therefore assigns a 70% probability that the global dedicated repair fleet remains below 30 vessels at the end of 2031.

For 2027:

65% — at least four publicly reported submarine cable damage incidents occur across the Baltic Sea and around Taiwan combined.

By the end of 2029:

70% — at least one new government-funded or government-mandated cable repair vessel programme is announced in Europe, Japan or the United States.

By the end of 2031:

70% — the global number of dedicated cable repair vessels remains below 30.

Across 2027–2036:

55% — at least one sovereign state experiences national-scale internet degradation lasting more than seven consecutive days because of submarine cable damage.

Each has a fixed resolution date and an observable criterion. They are forecasts designed to test the mechanism, not claims of certainty. TIC_S2A06_Website_Full

Series II now contains six physical layers:

Electricity.
Frontier computing.
Maritime chokepoints.
Water.
Refined materials.
Connectivity.

Something has become clear.

Concentration has held up as an analytical variable.

Criticality has held up.

Substitution time has now failed three times — and each failure has been different.

For water, the alternative existed but belonged to another user.

Substitution was political.

For refined materials, the alternative could be built but its economics could be influenced by the incumbent.

Substitution was financial.

Connectivity produces a third problem.

The substitution time is accurate.

It's simply the wrong measurement.

Traffic reroutes in seconds.

The system doesn't fail when the alternative engages.

The risk accumulates during the weeks before the original is restored.

That means the framework needs a companion variable:

How long must the system operate on its alternative?

And what redundancy remains while it does?

That distinction reaches beyond submarine cables.

A resilient system isn't simply one that absorbs the first failure.

It is one that can restore its margin before the next failure arrives.

That is the sixth row.

And it changes the instrument.

Next in Series II:

The final physical layer — the one sitting underneath all six.

THRIVE IN CHAOS
Signal Over Noise

Analysis → Forecast → Recommendations
Signal → Meaning → Action → Stability

Probability-based analytical forecasts, not certainties. THRIVE IN CHAOS is a Decision Intelligence system, not investment, legal or tax advice.