What holds a system together once the functions have separated

Article 8 of 8  ·  Series I of III  ·  Published 30 September 2026  ·  Analysis → Forecast → Recommendations

How this series measures things. Every article applies the same three questions to its subject. Concentration: how many genuinely independent alternatives exist, once shared upstream origins are traced rather than counted. Criticality: what stops if this fails, and how quickly. Substitution time: how long until an alternative actually functions. This article closes Series I by applying the instrument to the series itself, and opens the question Series II is built to answer.

1. The Signal

There are roughly two hundred faults on the world's subsea cable systems every year. That number has been broadly steady since 2013, even as route mileage rose from about 1.5 million to 2.7 million kilometres — which means the fault rate per kilometre has actually improved, through better surveys, deeper burial and better armouring.

Around 86 percent of those faults come from fishing and anchoring. The International Cable Protection Committee recorded 206 repairs in 2023 across 136 jurisdictions, and roughly 170 in 2025. Some 44 percent of repairs occur in territorial waters and 54 percent in exclusive economic zones, because cables are most exposed near shore.

And one figure in that dataset runs the other way. Repair response time has more than doubled over ten years. The median global repair for a subsea cable fault now runs about forty days. The longest single repair on record took 947 days.

Hold those two trends together, because they are the whole of this article in one dataset. The failure rate is stable and improving per unit of infrastructure. The recovery time is getting worse.

The consequences are visible in specific places. When the BCS East-West Interlink was severed in November 2024, it was carrying roughly a third of Lithuania's internet capacity. At least eleven cables in the Baltic have been damaged or cut since October 2023. In February 2026 the European Commission announced a 347 million euro subsea infrastructure package including a 20 million euro rapid repair pilot — which is a state response to a privately owned layer, arriving after the pattern was established.

Series I has spent seven articles on the institutional layer coming apart: obligations deferred, categories replacing cases, capacity concentrating unevenly, measurement decoupling, correction slowing, and functions leaving the state bundle for licensed private provision.

This is what sits underneath all of it. Not another institution. A physical layer with a stable failure rate, a worsening recovery time, and single points that carry a third of a country's connectivity.

2. The Mechanism

Article 7 established that the state was a bundle held together by three costs, two of which have collapsed. The obvious next question is what the separated functions rest on now, and the intuitive answer is wrong.

The intuition is that separated functions become independent. Different providers, different jurisdictions, different failure modes — the diversification this series keeps recommending, achieved at the level of a whole society.

What actually happens

Separated functions do not become independent. They converge downward onto a small number of shared substrates.

A payment provider, an identity provider, a cloud service and a logistics platform are institutionally distinct, differently owned, differently regulated and differently supervised. Underneath, they run on the same fibre, the same power, the same handful of compute regions, and increasingly the same settlement infrastructure.

The coordination that political authority used to supply is now supplied by infrastructure. That is the structural finding of this article, and it has an uncomfortable corollary: infrastructure has owners rather than obligations, and it coordinates without ever having agreed to.

Why this is the series instrument, applied at the largest scale

Every article in Series I has made one correction, in a different domain each time. Three suppliers behind one port. Nominally independent forecasts sharing a parent cause. Several services under one supervisor. Two banks clearing through one rail.

The pattern is always the same: alternatives that are genuinely separate at the level where they are counted, and identical at the level where they fail.

Institutional separation is the largest instance of it. A society that has unbundled its functions across many providers has diversified at the layer people examine and concentrated at the layer nobody does, because the substrate is boring, privately held, and reported only when it breaks.

Applying the instrument to the substrate

The substitution time is the number that matters and the one that has moved. A stable failure rate with a lengthening recovery is exactly the profile Article 1 described: the system is not breaking more often, and each break costs more, because the capacity to recover was optimised away while the capacity to fail stayed constant.

Repair depends on a small global fleet of specialised vessels, crews qualified to splice fibre at sea, and access to a location that may sit in disputed or restricted waters. None of those scales quickly, none of them was ever redundant, and all of them are commercially owned in a market with thin margins.

3. Subtheme One — Where the Seams Actually Run

If the substrate is where the concentration sits, then the joints in a modern system are not where the institutional map places them. Five joints carry most of it, and each was touched in an earlier article without being named as a joint.

Why the joints are invisible

Three reasons, and none of them is anyone's fault.

—   They are boring. A cable repair vessel is not a story until it is the only story, and by then the outage is already happening.

—   They are privately held. There is no obligation to report concentration in a commercial layer, and no supervisor whose mandate covers the layer as a whole rather than the firms sitting on it.

—   They are counted at the wrong level. A regulator counts licensed providers. A procurement team counts suppliers. Neither has a reason to ask what all of them run on, and no one is assigned the question.

The measurement that does not exist

Every joint above has good data about its own components and almost none about cross-layer dependency. Cable faults are counted. Refining concentration is reported. Payment volumes are published. Compute capacity is estimated.

What nobody publishes is how many services fail together when one joint fails. That is the number a society would need to know its own concentration, and it does not exist in any jurisdiction we have examined — not because it is hard to compute, but because computing it requires a body with a mandate spanning layers that are separately regulated and separately owned.

Which is a familiar shape from Article 5: the measurement that would reveal a problem is the one nobody is funded to produce, and its absence reads as the absence of a problem.

4. Subtheme Two — What Actually Binds a Separated System

If political authority no longer supplies the coordination, something else does. Three candidates are usually offered, and they are not equally load-bearing.

Standards and interoperability

The candidate most discussed and the weakest of the three. A common specification does hold things together — the identity programme in Article 4 is a real example, and the convergence of stablecoin regimes in Article 7 is another.

But a standard binds only those who adopt it, and adoption is voluntary until someone can compel it. The AI Act sandbox provision in Article 6 was a common design with a common deadline, and it produced one working instance out of twenty-seven. Standards coordinate willing and capable parties, which is valuable and is not the same as holding a system together.

Ownership of the substrate

The candidate least discussed and the most load-bearing. Whoever owns the fibre, the rail, the compute region or the refining capacity determines who can operate above it, on what terms, with what continuity.

This is a form of authority and it is unlike the political kind in three specific ways.

—   It carries no obligation to those it governs. A cable operator owes duties to its customers and none to the population whose connectivity depends on the cable.

—   It has no route of appeal. A commercial decision about where to lay capacity, or whether to maintain a route, is not contestable by the people it affects.

—   It is invisible until it fails, and its failures are attributed to whatever was running on top of it.

This is authority without accountability, and it arrived without anyone deciding to grant it. Not through a seizure of power but through the ordinary process of separated functions converging on shared plumbing, one commercial decision at a time.

Residual coercion

The third candidate is what Article 7 identified as the function that did not unbundle: the capacity to compel within a territory.

It still binds, and it binds asymmetrically. A state can compel a cable landing station on its soil, a licensed issuer within its jurisdiction, a data centre it can physically reach. It cannot compel a repair vessel in international waters, a refining facility on another continent, or a supervisor in another jurisdiction to protect its residents.

The European response to the Baltic cable damage illustrates the boundary precisely: surveillance, prosecution, a funding package, a repair pilot. Every one of those is a territorial instrument applied to a problem that is only partly territorial, and the prosecutions have contested jurisdiction as a live issue.

The honest summary

A separated system is held together mostly by infrastructure ownership, which has no accountability; partly by residual state coercion, which stops at a border; and least by standards, which is what everyone talks about because it is the only one that can be negotiated. That is not a comfortable finding and we have not found a way to make it less so.

5. What Most Analysis Gets Wrong

That institutional diversity implies resilience

It implies resilience at the institutional layer only. Many providers, many jurisdictions and many regulators sitting on one cable, one rail or one compute region is the same single point of failure with better paperwork. Diversity is measured where it is visible and concentration accumulates where it is not.

That the substrate is a technology problem

The subsea data says otherwise. The fault rate per kilometre has improved through better engineering. What has worsened is repair time, which is a function of vessel availability, crew supply, access to disputed waters and the commercial economics of a maintenance market. Engineering solved the part that engineering could solve.

That someone is monitoring this

Each joint has a supervisor for the firms operating on it. None has a supervisor for the layer as a whole, and no body publishes cross-layer dependency. The assumption that a systemic view exists somewhere is the specific error this article is written to correct, and it is an easy assumption because the alternative sounds implausible.

That the answer is to nationalise the substrate

Public ownership changes who holds the concentration and does not reduce it. A state-owned single cable is a single cable. Article 3 also gives the reason to be cautious: a state that acquires an asset acquires it into an institution with the correction latency measured in Article 6, and the substrate needs faster repair rather than slower decision-making. The useful question is redundancy and recovery time, not who holds the title.

6. Base, Stress and Extreme

Four paths, with our probability assessment and the condition that would falsify each. Probabilities sum to one hundred.

The base case is the one that produces no events and the most damage. Silent concentration is not a crisis; it is the absence of one, right up until the arithmetic is tested. Every article in this series has described some version of the same thing, which is why it opens the closing article rather than closing it.

7. Forecast — One Year, to mid-2027

Repair time does not improve

Probability 0.75  ·  Confidence: Medium-High

We expect the median global repair time for a subsea cable fault to remain at or above thirty days, notwithstanding announced rapid-repair initiatives, because the binding constraints are vessel availability and crew supply and neither responds inside a year.

A twenty million euro repair pilot is a real intervention at a regional scale and it does not change a global median. The distinction between a regional improvement and a systemic one is the whole difficulty of intervening in a shared substrate: the layer is global, and every instrument available to fix it is national or regional.

Second-order effect. Regions that fund repair capacity improve, and the improvement is invisible in aggregate figures dominated by regions that do not. The global median will therefore understate progress in well-funded areas and overstate resilience in poorly served ones, in both directions at once.

What would weaken it. Median repair time falling materially, which would indicate that vessel and crew capacity is more elastic than the last decade suggests.

8. Forecast — Three Years, to 2029

Substrate designation spreads without redundancy obligations

Probability 0.60  ·  Confidence: Medium

We expect at least three G20 states to formally designate privately held digital or settlement infrastructure as critical national infrastructure, with reporting and security duties attached, and we expect most of those designations to carry no funded obligation to build redundant capacity.

The asymmetry has a straightforward cause. Reporting and security duties cost the state nothing and impose the burden on the operator. Redundancy costs money and produces an asset that earns nothing in normal conditions, which is the same structural position Article 5 identified for verification and Article 6 for adaptation.

Second-order effect. Designation without funding transfers the cost of national resilience to a commercial operator in a thin-margin market, which tends to reduce investment in exactly the capacity the designation was meant to protect.

What would weaken it. Designations arriving with funded redundancy or minimum recovery-time obligations, rather than reporting duties alone.

9. Forecast — Five Years, to 2031

A shared-dependency outage is documented as such

Probability 0.65  ·  Confidence: Medium

By the early 2030s we expect an outage in a G20 country to remove service from more than one nominally independent provider at once, and to be traced by a regulator or supervisor to a single shared dependency — a cable, a settlement rail, an identity framework or a compute region.

This forecast is about attribution rather than occurrence. Simultaneous failures of separately owned services already happen; what is missing is an official finding that names the shared layer as the cause rather than reporting several coincident incidents. The forecast resolves on whether the attribution is made.

Second-order effect. The first such finding will do more for cross-layer measurement than a decade of argument, because it converts an abstract concentration into an incident with a report attached. That is the mechanism by which every previous infrastructure regime was built, and it is not a satisfying one.

What would weaken it. Multi-service outages continuing to be reported as separate incidents, which would confirm the base case rather than this forecast.

10. Forecast — Ten Years, to 2036

Recovery time remains the unimproved variable

Probability 0.55  ·  Confidence: Medium-Low

Over a decade we expect the median global subsea repair time for 2035 to stand at or above its 2025 level, and we expect the fault rate per kilometre to have continued improving.

The divergence is the point. Engineering improves what engineering can reach, and repair capacity is a commercial and logistical constraint rather than a technical one. A forecast that both trends continue in opposite directions for another decade is a specific claim, and it is the cleanest available test of whether anything in this series is being acted on.

We hold this at lower confidence than the probability suggests. A decade is long enough for a securitisation response to change the economics of the repair market entirely, and the second scenario in section 6 describes exactly that. This forecast assumes it does not arrive at scale, which is an assumption rather than an analysis.

What would weaken it. Sustained public or consortium funding of repair capacity sufficient to change the global median, which would be the strongest single piece of evidence against the whole substrate argument.

11. Signals to Watch

—   Median and maximum repair times, which measure recovery rather than failure and are the variable that has actually moved

—   Route diversity: whether alternative paths are being built in the regions with one, rather than added where there are already several

—   Repair vessel and qualified crew capacity, which is the binding constraint and is reported almost nowhere

—   Designations of private infrastructure as critical, and specifically whether a funded redundancy obligation is attached or only reporting duties

—   Attribution in outage reports: whether a shared layer is named as the cause, or several services are reported as coincidentally failing

—   Any body acquiring a mandate that spans layers rather than firms

—   Cross-layer dependency measurement. We maintain this list and it is empty

12. Recommendations — Individuals

This article's practical content is thinner than the previous seven, and saying so is more useful than manufacturing tasks. Substrate concentration is not something an individual can address. What is available is knowing where you sit relative to it.

Immediate — 30 days

Find out what your connectivity actually depends on. In many places the answer is several independent paths and the question is uninteresting. In others it is one cable, one landing station, or one route through a neighbouring country, and that is worth knowing before it matters rather than discovering it during an outage.

Build — 12 months

Establish which of your functions fail together. Take the dependency map from Article 7 and ask a narrower question of it: if connectivity goes for a week, what stops? For most people the honest answer includes payments, identity verification, work, records and communication simultaneously — not because they chose a single provider but because everything they chose runs on the same layer.

Then hold one offline capability for each function you cannot afford to lose for a week. A means of payment that does not require connectivity. A copy of critical documents that does not require authentication. A way to be reached that does not depend on one platform. None of this is preparation for catastrophe; it is a forty-day median repair time, applied to your own life.

Position — 3 years

Where a decision involves location — living, working, holding assets — treat substrate redundancy as a factor alongside the ones normally considered. Regions with one path to the outside world are exposed in a way that does not appear in any conventional assessment of them, and the exposure is knowable in advance from public cable maps.

Avoid. Treating this as a reason for anxiety. The failure rate is stable and improving per unit of infrastructure. The problem is recovery time, which affects the length of an outage rather than its likelihood, and length is exactly what a modest amount of preparation addresses.

Why this works. You cannot influence substrate concentration and you can know your own exposure to it, which converts an unbounded worry into a specific and mostly manageable one.

13. Recommendations — Business

This is where the article's content is operational rather than observational, because a business can trace its own substrate and almost none has.

Immediate — 60 days

Take your vendor list and collapse it to substrates. Which cloud regions, which settlement rails, which identity providers, which physical routes. The list of vendors is long and the list of substrates is usually short, and the second is your actual dependency map.

Then run the multi-service question. For each substrate, list every function of the business that stops if it fails. Most organisations discover one or two substrates behind four or more functions, which is a correlated failure invisible on any supplier assessment.

Build — 12 months

Set your continuity planning against a forty-day recovery rather than a forty-hour one for anything that depends on physical infrastructure. That is the published median for subsea repair and it is a better planning assumption than any service level agreement, because an agreement allocates liability and does not restore a cable.

Where you contract for redundancy, verify that it is redundant at the substrate level. Two providers, two contracts and two invoices routinely resolve to one physical path, and the provider is frequently unable to tell you whether they do.

Position — 3 years

Expect substrate designation to reach you as a compliance obligation if you operate at any scale on infrastructure a state considers critical, and expect the obligation to arrive as reporting and security duties rather than as funding.

And treat recovery time as a procurement criterion in its own right. Price, capability and uptime are all standard. Time to restore after a physical failure is rarely asked, frequently unknown to the vendor, and it is the variable that determines what an outage costs you.

Avoid. Accepting a redundancy claim that has not been traced to physical paths. This is the single most common instance of the failure this entire series describes, and it is the one where verification is cheapest — the information exists, and asking for it costs an email.

Why this works. Substrate concentration is measurable at the level of one organisation even though it is unmeasured at the level of a society. You cannot fix the layer; you can know precisely how exposed you are to it, which is more than any regulator currently knows about it.

14. Recommendations — Capital

The structural observation is that recovery capacity is a real constraint on a widely held set of assets, and it is not priced because it is not reported.

Immediate — this quarter

For holdings whose operations depend on connectivity, settlement or compute, establish which substrates they sit on and whether positions that look diversified converge underneath. Sector diversification does not diversify a cable route or a settlement rail, and this is the same correlated exposure the series has described in supply chains, in institutions and in its own forecast record.

Build — 12 months

Separate two exposures. Failure probability is one question and is broadly stable. Recovery duration is another and has doubled over a decade. For a leveraged asset, the second determines whether an outage is an inconvenience or a covenant event, and only the first is normally modelled.

Then look at the maintenance market itself. Repair capacity is commercially owned, thin-margin, and structurally under-supplied relative to the mileage it serves. That is a supply and demand observation rather than a recommendation, and its consequences run in more than one direction — including for the operators who depend on the capacity and cannot compel it.

Position — 3 years

Watch designation and what accompanies it. A designation with reporting duties transfers cost to an operator and tends to reduce its investment. A designation with funded obligations changes the economics of the layer. These look similar in an announcement and are opposite in effect, and the difference is visible in the instrument.

Avoid. Reading a national resilience package as a change in the global constraint. The layer is worldwide and every available instrument is national or regional, which means well-funded regions improve and aggregate exposure moves very little.

Why this works. Recovery time is published, has moved decisively in one direction for a decade, appears in no valuation model, and determines the cost of the failures that do occur. Unmeasured variables that determine outcomes are where the information advantage sits, and this one requires reading industry repair statistics rather than market commentary.

15. What Would Change Our Mind

Each forecast carries its own weakening condition. Three developments would undermine this article's argument as a whole.

—   Repair and recovery capacity expands enough to reverse the decade-long trend in restoration times. That is the load-bearing empirical claim and the cleanest thing to check.

—   Institutional separation demonstrably produces substrate diversification rather than convergence — route diversity, settlement diversity and compute diversity rising as functions unbundle. Our claim is that the opposite happens and it is directly testable.

—   A body acquires a cross-layer mandate and publishes dependency measurement. This would falsify our claim that the measurement gap is structural rather than incidental, and we would treat it as the most welcome result in the whole series.

The closing tally for Series I. Thirty-two forecasts across eight articles. Eleven resolve against European institutions, which was twelve of sixteen at the midpoint and is now roughly a third — the concentration recorded in Article 4 was corrected by finding better-documented material elsewhere rather than by adding weaker examples, which was the stated remedy. Ten of the thirty-two share a parent cause with at least one other, and every instance is recorded in the article that produced it. None of this makes the record good; it makes it legible, which is a precondition rather than an achievement. The record becomes worth anything only when the first forecasts resolve, and the earliest of those falls due in January 2027.

Founder's Lens

[ EDITORIAL GATE — WRITTEN BY HAND BEFORE PUBLICATION. Never generated. Replace this marker with the founder's text, or record a suspension. ]

16. Bottom Line — and What Series II Measures

Series I began with a claim that the world is becoming less forgiving, and traced it through seven mechanisms.

One correction runs through all eight. Alternatives that are genuinely separate where they are counted, and identical where they fail. Three suppliers behind one port. Two banks on one rail. Twenty-seven states on one specification. Four forecasts against one institution. Institutional diversity on one cable.

And underneath the institutional layer sits a physical one with a stable failure rate, a recovery time that has doubled in a decade, single points carrying a third of a country's connectivity, and no body with a mandate to measure any of it across layers.

Which is where Series I stops and Series II begins.

Series I asked what happens to institutions when they cannot keep pace. Series II asks the narrower and more measurable question: where exactly are the joints, and what does the arithmetic look like at each one. Energy and the grid. Compute and the chips underneath it. Water. Refined materials. Connectivity. Logistics corridors. The infrastructure seams that this article has only named.

Each of those articles will apply the same three questions and contribute one row to a single comparative table. Concentration, criticality, substitution time, measured substrate by substrate. By the end of Series II the reader will hold a matrix rather than eight essays, and the matrix is the point.

The instrument has been established here. Series II is where it gets used on the physical world.

And the closing observation of Series I is the one that carries across. The world is not becoming less forgiving because things break more often. Things break at roughly the same rate they always did. It is becoming less forgiving because the time to recover from each break is getting longer, in institutions and in infrastructure alike, while the number of things that depend on each break staying short keeps rising.

Chaos is the rising cost of the next choice. Most of that cost is time.

Forecast record

Four forecasts, one per horizon, each with a threshold, a named verifier and a resolution date, recorded before the outcome is known.

Correlation, recorded rather than assumed away. The first and fourth share both a verifier and a parent cause in repair market capacity, and are effectively one observation measured twice at different horizons. The second and third are independent of them and of each other. So this set contains three independent observations rather than four, and we would rather state that than let a count of four stand.

Directional statements elsewhere in this article carry no threshold and are deliberately excluded from the record.

Sources

A note on what this article does not claim. Several Baltic incidents are the subject of live prosecutions in which jurisdiction and attribution are contested, including by governments with the strongest interest in establishing them. We use the incidents to establish that damage occurred, that repair took time, and that redundancy varied by region. We make no claim about intent in any individual case, and the argument does not require one: an accidental anchor and a deliberate one produce the same forty-day median.

In this series

—   Previous: Article 7, State Unbundling — what happens when functions that arrived as one package come apart.

—   This completes Series I. Series II applies the same instrument to the physical substrate, one joint per article, building a single comparative table.

—   The method behind the Chaos Index and this series: /methodology

THRIVE IN CHAOS

Decision Intelligence for an Uncertain World

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

Signal Over Noise  ·  thriveinchaos.ai

AI intelligence system with human editorial oversight.

Forecasts are probability-based analytical assessments, not certainties. This material supports independent judgment and does not constitute financial, legal or investment advice.