WHEN CAPACITY CONCENTRATES

What changes when the historical limit on the centre stops binding

Article 4 of 8  ·  Series I of III  ·  Published 2 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. Article 3 examined what an institution does when it cannot acquire capacity. This article examines what happens where it can.

1. The Signal

Two things are happening inside the same institution at the same time, and they point in opposite directions.

In July, the European Union deferred the high-risk obligations of its Artificial Intelligence Act by sixteen months, because the authorities to enforce them had not been designated and the standards to assess conformity were unfinished. In March, it narrowed sustainability reporting to firms above a thousand employees, removing most of the population it previously covered.

Both are the response described in Articles 2 and 3: an institution that cannot match the variety it faces subtracts the variety it must handle.

Meanwhile, the same institution is building infrastructure that does the opposite.

Under Regulation (EU) 2024/1183, every one of the twenty-seven member states must make a certified European Digital Identity Wallet available by December 2026. From mid-2027, regulated sectors — banks, telecoms, public services — must accept it. A Commission implementing regulation on enrolment was published in April 2026. Four large-scale pilots ran from 2023 across twenty-six member states plus Norway, Iceland and Ukraine, involving more than three hundred and fifty public authorities and private companies, with around forty-six million euros of European funding.

A wallet certified by one member state must be accepted by service providers in all the others. Use by citizens is voluntary. Provision by states is not.

Set the two side by side. In one domain the institution is reducing the number of distinctions it recognises. In another it is building the capacity to recognise a person, a credential and an entitlement anywhere in a market of four hundred and forty million people, in a way it previously could not.

These are not contradictory policies. They are the same institution responding to the same shortage in the only two ways available, and which response it uses is determined by something specific.

That determinant is the subject of this article.

2. The Mechanism

Article 3 set out the governing constraint: a control system must possess at least as much variety as the system it regulates, and an institution that cannot acquire variety must subtract it.

That formulation contains an assumption worth examining, because for most of modern history it held so reliably that nobody stated it.

The historical limit was arithmetic, not ideology

The reason centralised administration has repeatedly failed at scale was rarely a shortage of authority or of will. It was a shortage of processing capacity.

A centre attempting to administer a complex society has to gather information about conditions it cannot observe directly, transmit it without excessive loss, integrate it into a decision, and transmit the decision back — all faster than the conditions change. Every one of those steps had a hard cost in people and time. Past a certain scale the cost exceeded what any administration could pay, and the system either decentralised, simplified, or produced decisions describing a world that no longer existed.

Decentralisation was not primarily a moral preference. It was a workaround for an information constraint. Local decision-making won because local decision-makers could observe local conditions at a cost the centre could not match, and that advantage was arithmetic rather than philosophical.

This is why the argument for distributed authority has always had two separable strands. One holds that dispersed power is desirable in itself. The other holds that it is necessary because the centre cannot process enough. The two strands agreed for so long that they were rarely distinguished.

What changes when the arithmetic changes

Machine processing attacks each step of the loop. Gathering becomes cheaper where activity is already digital. Transmission becomes effectively free. Integration becomes tractable where the data is structured. The decision can in principle be produced at machine speed.

If the constraint that produced decentralisation weakens, the second strand of the argument weakens with it. The first strand — that dispersed authority is desirable — is unaffected, and it now has to carry the whole load on its own.

That is a change in the structure of an argument rather than in anyone's politics, and it applies to every actor, not to a category of them.

A national government, a supranational body, a bank, an insurer, a platform and a large employer all face the same constraint and gain the same relief. Nothing in the mechanism is specific to a form of government. A capability available to a democratic administration is equally available to an authoritarian one, and equally available to a private firm with no political character at all.

The determinant: bought or trained

This gives the answer to the question in section 1. Whether an institution subtracts variety or acquires it depends on which kind of capacity the task requires.

The same institution can therefore be centralising and simplifying at once without inconsistency. It acquires variety wherever the required capacity is purchasable and subtracts it wherever the required capacity is a professional practice. That is not a strategy anyone announced. It is what falls out of two different acquisition timescales meeting the same deadline pressure.

3. Subtheme One — Capacity Concentrates Where the Substrate Concentrates

If variety can now be bought, the question becomes who can buy it. The answer is not determined by ambition, and it produces a divergence that has nothing to do with political system.

The four inputs

Acquiring administrative variety at scale requires four things, and each is unevenly distributed.

—   Structured data about the governed population. Not data in general, but records that are machine-readable, current, reconciled across agencies and legally usable for the purpose. Most administrations hold enormous quantities of data that fail at least one of those tests.

—   Identity infrastructure. A reliable way to establish that a person is who they claim to be, that scales, works across jurisdictions and survives challenge. This is precisely what the European wallet programme is building, and it is the piece most often missing.

—   Compute, and the electricity underneath it. Article 1 established that the constraint on compute is no longer demand but the physical chain that supplies it — grid connection, transformers, cooling. Administrative capacity now inherits a supply constraint from a completely different sector.

—   Standards and legal authority to use all three together. The technical capability to link records is worthless where the linkage is unlawful, and the legal authority is worthless where the records will not reconcile.

Why this produces divergence rather than a trend

The four inputs are complements, not substitutes. An administration with excellent registries and no identity layer cannot act on them. One with an identity layer and unreconciled registries authenticates a person and then cannot determine their entitlement. One with both and no lawful basis has built something it may not use.

Complementary inputs produce threshold behaviour. Below the threshold, additional investment in any single input yields very little. Above it, the whole stack becomes usable at once.

So this does not arrive as a gradual improvement everywhere. It arrives as a small number of administrations crossing a threshold and acquiring a capability the others do not have, while the others continue subtracting variety because subtraction remains their only available move.

The European rollout illustrates the unevenness inside a single legal deadline. Every member state faces the same December 2026 obligation. As of mid-2026, a handful had public developer environments that organisations could actually test against, several had citizen-facing pilots without them, and others had announced projects with nothing yet visible externally. A common deadline does not produce common capability, because the deadline was set on the legal layer and the constraint sits on the data layer underneath it.

The uncomfortable implication

An administration's ability to govern with fine resolution is becoming a function of its infrastructure rather than of its competence in the conventional sense.

A well-regarded institution with fragmented registries and no identity layer will subtract variety, because it must. A less-regarded one with a clean population register, a working identity system and adequate compute can acquire variety, because it can. On any conventional ranking of governance quality the first outranks the second, and on the question of what it can actually do the ranking inverts.

This is the same finding as Article 1, in a different domain. Nominal capability and usable capacity are different quantities, and only the second one determines outcomes.

4. Subtheme Two — What Capacity Does Not Buy

Variety solves one problem completely and leaves an adjacent one untouched, and the two are easy to confuse because both sound like knowing more.

Distinguishing is not correcting

Requisite variety is a statement about the ability to tell states apart and respond differently to each. It says nothing about whether the responses are right.

A control system with high variety and poor feedback does not fail visibly. It acts decisively, at scale, on a model of the world it cannot check — and its confidence rises with its capability, because the capability is real and the error is in a different place.

Correction requires something variety does not supply: a channel through which the world contradicts the model, that the centre cannot suppress or reinterpret, and that arrives fast enough to matter. Those are three separate properties, and no amount of processing capacity produces any of them.

Why acquired capacity tends to weaken feedback

There is a mechanism here rather than a mere possibility, and it operates without anyone intending it.

The channels that previously carried correction were largely the ones the centre could not process. Local officials with discretion. Intermediaries who saw the case and the rule at the same time and noticed when they diverged. Appeals that forced a decision to be defended in front of someone who had not made it. Each of those was expensive, and each was a place where reality entered the system in a form the centre had not specified.

Acquiring capacity removes the reason those intermediaries existed, and removing them removes the correction they carried as a by-product. The efficiency gain is measured. The lost feedback is not, because nobody was measuring what the intermediary noticed.

Article 2 identified the irreducible step: a decision carrying consequence must be attributable, contestable and legitimate, and none of the three scales with compute. That step is now visible as the limiting factor rather than as an obstacle. It is where the correction enters.

The binding constraint moves

Where capacity can be bought, the question stops being whether the centre can process enough and becomes whether it can find out that it is wrong.

That is a harder question and a far less discussed one, partly because it has no procurement solution. An administration can buy an identity system. It cannot buy a channel through which its own errors reach it undistorted, because any channel it builds and controls is one it can also filter, and a channel it does not control is one it cannot rely on.

This is the subject of Article 5, and it is the reason this article stops here rather than forecasting outcomes from capability alone. Capability tells you what an institution can do. It does not tell you whether what it does will work.

5. What Most Analysis Gets Wrong

That this is a question about regime type

The mechanism is indifferent to it. Processing capacity is a capability, and capabilities do not carry political character. The same identity infrastructure serves fraud prevention, benefit delivery, tax administration and surveillance, and which of those it serves is determined by law and oversight rather than by the technology. An analysis that sorts countries into categories and predicts from the category will mispredict, because the determining variable is infrastructure and the categories do not track it.

That centralisation is the automatic consequence

Capacity to process is not the same as authority to decide, and the two can move independently. An administration can acquire the ability to see a case in detail and simultaneously delegate the decision, and several deliberately do. The wallet architecture is an example: it was designed so that credentials sit with the holder rather than in a central repository, which is a capability increase without a corresponding concentration of data. Whether that design survives contact with operational pressure is a separate question, and a fair one.

That the private sector is a bystander

The same acquisition is available to banks, insurers, platforms and large employers, and several are further along than most administrations. A firm that can verify identity, assess risk and price individually at scale has acquired administrative variety over its own population of customers. It is subject to different constraints and to weaker contestability requirements, because a commercial decision generally owes fewer reasons than a public one.

That deferral in one domain indicates retreat in all

This is the error the opening section is designed to prevent. An institution deferring in one area and building in another is not inconsistent and is not confused. It is doing the only two things available to it, in the domains where each is possible. Reading the deferral as a general direction of travel, and missing the acquisition happening beside it, produces a forecast that is wrong in both directions at once.

6. Base, Stress and Extreme

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

The second path deserves more weight than it usually receives. The European programme is deliberately built as a common specification with cross-border acceptance, which is a mechanism for lowering the threshold rather than for raising the leaders. If it works, it is the strongest available counter to the divergence in the base case.

7. Forecast — One Year, to mid-2027

The identity deadline is met on paper and unevenly in practice

Probability 0.75  ·  Confidence: Medium-High

We expect most member states to have something they can call a compliant wallet by the December 2026 deadline, and a substantial minority to have nothing that organisations can meaningfully integrate with. The distinction between a citizen-facing pilot and a working relying-party ecosystem is the one that matters operationally, and it is not the one the deadline measures.

The gap is not negligence. The obligation sits on the legal layer and the constraint sits underneath it, in population registers, agency data reconciliation and procurement capacity that a directive cannot legislate into existence.

Second-order effect. Firms building on the wallet face a fragmented rollout while a single legal framework tells them the market is unified. The cost of that mismatch falls on the private integrators rather than on the administrations, and it will be read as a technology problem rather than as an administrative capacity problem.

What would weaken it. Broad availability of testable relying-party environments across most member states by mid-2027, which would indicate that the underlying data constraint was smaller than we assess.

8. Forecast — Three Years, to 2029

The acceptance obligation is deferred or narrowed

Probability 0.60  ·  Confidence: Medium

From mid-2027 regulated sectors are required to accept the wallet. We expect that obligation to be deferred, phased or narrowed in scope before it applies uniformly — for exactly the reason the AI Act obligations were deferred, and in exactly the same form.

The pattern from Article 3 predicts this specifically: where an obligation requires capacity that is not yet in place, the deadline moves and the scope narrows, and the change is presented as proportionality or phased implementation rather than as a capacity shortfall.

Second-order effect. Every deferral transfers cost from the party that was behind to the party that was ready. Organisations that built early hold capability they cannot yet use, and the deferral rewards those that waited. Repeated across programmes, this teaches a lesson about preparation that is expensive to unlearn.

What would weaken it. The acceptance obligation applying on schedule and uniformly across regulated sectors, which would be the first major European digital deadline in recent years to do so.

9. Forecast — Five Years, to 2031

Acquired capacity becomes the sole route in specific services

Probability 0.50  ·  Confidence: Medium

By the early 2030s we expect several administrations to have made digital identification the only practical route to at least one significant public service, not by prohibiting alternatives but by allowing them to atrophy — longer waits, fewer offices, reduced staffing on the manual path.

This is how a voluntary system becomes effectively mandatory without any decision to make it so. The alternative remains legally available and becomes practically unusable, which is form three from Article 3 operating in reverse: the obligation does not stop binding, the alternative stops functioning.

Second-order effect. The population that cannot use the digital route — for reasons of capability, documentation, or status — becomes concentrated in the atrophying channel, which is also the channel with declining resources. Exclusion arrives as a resourcing decision rather than as a policy, and therefore without a debate.

What would weaken it. Explicit legal guarantees of a maintained non-digital route with resourcing attached, rather than a statement of principle that a service must remain accessible.

10. Forecast — Ten Years, to 2036

The divergence is between infrastructure, not between systems of government

Probability 0.55 for a pronounced version  ·  Confidence: Medium-Low

Over a decade we expect administrative capability to correlate more strongly with data and identity infrastructure than with regime type, national income or conventional governance ranking. Two administrations with similar political systems and different registries will diverge; two with different political systems and similar infrastructure will converge on what they can do, whatever they choose to do with it.

The consequence is that a great deal of comparative analysis organised around political categories will predict poorly, and will keep being produced because the categories are what the available data describes.

We hold this at lower confidence than its probability suggests. It requires a measurement that does not exist — comparable administrative capability data across jurisdictions — and forecasts that depend on the emergence of their own measuring instrument resolve badly.

What would weaken it. Capability converging across administrations with very different infrastructure, which would indicate that the four inputs are more substitutable than we assess.

11. Signals to Watch

—   Relying-party environments, not citizen pilots. Whether organisations can actually test against an identity system is the operational measure; whether citizens can download an app is not

—   Population register reconciliation projects, which are unglamorous, rarely announced, and the actual constraint

—   Public compute allocation and grid connection for government workloads, where administrative capacity meets the physical constraint from Article 1

—   Deferral or narrowing of acceptance obligations, which is the Article 3 mechanism arriving in the identity programme

—   Resourcing of non-digital service channels, measured in staffed hours rather than in stated availability

—   Appeal volumes and outcomes where automated assessment has been introduced. A collapse in successful appeals can mean the system is accurate or that the appeal has stopped working, and the two require different evidence to distinguish

—   Whether any administration publishes the error rate of an automated determination. Almost none do, and the first that does will be more informative than any capability announcement

12. Recommendations — Individuals

The practical consequence of acquired capacity is that you become more legible to institutions, faster, and with fewer intermediaries between their model of you and their decision about you.

Immediate — 30 days

Find out what your administration already holds and whether it is correct. Where a register can be inspected, inspect it. Errors in a low-resolution system produced friction that a human resolved. Errors in a high-resolution system propagate at machine speed into decisions that were never individually made, and correcting them afterwards is markedly harder than correcting them before they are relied on.

Build — 12 months

Maintain a functioning non-digital route for anything that matters. Not as a protest, and not from distrust of the technology, but because a single route with no fallback is the dependency profile this series keeps describing. Keep documents you can produce in physical form, know which office still handles the manual path, and use it occasionally so that you discover it has closed before you need it.

Where digital identity becomes available, understand what each credential presentation actually discloses. Selective disclosure is a design property of these systems and it is only a benefit to someone who uses it deliberately.

Position — 3 years

Assume that the intermediaries who used to notice when a rule fitted your case badly will not be there. There is no substitute for them, and the realistic adaptation is to make your position legible in the terms the system uses rather than to rely on someone reading it sympathetically.

Avoid. Treating the manual route as permanently available because it remains legally available. Alternatives rarely close by decision. They close by resourcing, and the announcement comes after the closure rather than before it.

Why this works. You cannot influence what capability an administration acquires. You can influence what it holds about you, whether you have a second route, and whether your situation is expressed in a form the system can read. All three are cheap now and expensive later.

13. Recommendations — Business

Two distinct exposures arrive here at once: what acquired capacity does to your obligations, and what it does to your customer-facing processes.

Immediate — 60 days

Establish whether you fall within the acceptance obligation from mid-2027 and, if so, what your integration path actually is in every market where you operate. A single legal framework does not imply a single technical route, and the readiness gap between member states is currently wide enough that market-by-market planning is required.

Then check the other direction. Where an administration acquires the capacity to verify something directly, requirements it previously placed on you as an intermediary may disappear, and revenue attached to performing that verification may disappear with them.

Build — 12 months

Build to the specification rather than to any national implementation. Where a common standard exists, an integration written against the standard survives a change of national supplier and one written against a supplier does not. This is the ordinary substitution-time argument applied to a compliance system.

Do not stand down early capability if the acceptance obligation is deferred. The obligation is likely to arrive with a new date rather than disappear, and the pattern across recent European programmes is that firms which dismantled readiness paid twice.

Position — 3 years

Assume identity and credential verification become cheap and universal in your market, and ask what in your business currently earns money because they are expensive. Onboarding friction, manual verification, document handling and eligibility checking are all revenue lines in some businesses and cost lines in others, and the same change moves both.

Where you make individual determinations about customers, expect the contestability requirement to tighten before the capability requirement does. The ability to explain a decision after the fact is becoming the binding constraint on deploying automated assessment, and building for it later is considerably more expensive than building for it now.

Avoid. Building an integration whose value depends on a specific national implementation surviving. Several current wallet implementations will be replaced, and an integration written against a supplier rather than a specification will be replaced with them.

Why this works. Acquired administrative capacity changes what is scarce. Anything whose value came from verification being difficult is exposed, and anything that becomes possible only once verification is easy is an opportunity. Both are visible in advance, in published specifications, months before they are visible in a market.

14. Recommendations — Capital

Two exposures, and they run in opposite directions from the same cause.

Immediate — this quarter

Identify holdings whose revenue depends on verification being expensive: identity checking, document verification, onboarding services, eligibility assessment, parts of compliance infrastructure. Where a public identity layer becomes universal and free at the point of use, the addressable problem shrinks even though the underlying need does not.

Build — 12 months

Look at the other side of the same trade. The wallet programme creates demand for integration, credential issuance, relying-party infrastructure and the verification layers that sit underneath the identity layer but are not provided by it. A government credential establishes who someone is; it does not perform the risk assessment that regulated firms still require.

Then treat administrative infrastructure as a jurisdictional variable in its own right, separate from tax, rule of law and political stability. Two jurisdictions with identical statutes and different registries produce different operating costs, different timelines and different predictability, and the difference is durable because registries take years to fix.

Position — 3 years

Watch where administrative capability and the physical constraint from Article 1 intersect. Government compute workloads compete for the same grid connections and the same power as commercial ones, and an administration that cannot secure them cannot acquire capacity regardless of its budget or its intent.

Avoid. Pricing identity infrastructure as a technology theme. It is a procurement cycle with legislated deadlines, which means the revenue is more predictable in timing and more capped in size than a technology narrative implies, and the deadlines move.

Why this works. This transition is unusually legible in advance. Specifications are published, deadlines are legislated, and implementing regulations are dated. The information is public and largely unread, which is the most reliable form of edge available.

15. What Would Change Our Mind

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

—   Administrative capability converges across jurisdictions with very different data and identity infrastructure. That would indicate the four inputs are substitutable and the threshold behaviour we describe does not exist.

—   Institutions acquiring capacity demonstrably improve their error detection at the same time. If acquired capability arrives together with stronger correction rather than weaker, the central caution in section 4 is wrong.

—   Regime type predicts administrative capability better than infrastructure does. If a categorisation by political system outperforms a categorisation by registry and identity infrastructure, our framing is the wrong one.

A limitation that has now become a pattern, and we would rather name it than let it accumulate quietly. Articles 3 and 4 both draw their case material from European Union instruments, and seven of the eight forecasts across the two articles resolve against European institutions. The reason is that the EU documents its own process in unusual detail, which makes the mechanism legible there. But a forecast portfolio concentrated in one jurisdiction is exactly the failure this series describes in section after section: nominal diversification across topics, single upstream origin. We are applying our own instrument to ourselves and reporting the result, which is that the concentration is real, we have not yet corrected it, and the correction is to find non-European case material of comparable documentary quality rather than to add weaker examples for balance.

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

The historical limit on centralised administration was arithmetic. A centre could not gather, transmit and integrate enough information fast enough, and decentralisation was the workaround. That constraint is weakening wherever the required capacity can be bought rather than trained.

Which is why an institution can defer one programme for lack of assessors and build another to a hard deadline in the same year. Judgement takes a decade to acquire. Infrastructure takes a procurement cycle. The response is determined by which kind of capacity the task needs, and not by any strategy about how much to centralise.

The divergence this produces runs between infrastructures rather than between systems of government. The four inputs — structured registries, identity, compute, legal authority — are complements, so capability arrives as a threshold rather than as a gradient, and an administration below the threshold keeps subtracting variety no matter how well regarded it is.

And capacity does not buy correction. Variety is the ability to tell states apart; it says nothing about whether the response is right. The channels that used to carry correction were largely the intermediaries the centre could not afford, and acquiring capacity removes the reason they existed. The efficiency gain is measured. What the intermediary noticed was never measured, so its loss will not appear in any evaluation of the change.

For anyone whose position depends on being seen accurately: check what is held about you while checking is still cheap, keep a second route open before the first becomes the only one, and express your situation in the terms the system uses rather than relying on someone to read it sympathetically.

The question is no longer whether the centre can process enough. It is whether it can still find out that it is wrong.

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 and concentration. All four resolve against European institutions and all four share a parent cause in the eIDAS 2.0 programme. This is a single-family set, not four independent observations, and it compounds the jurisdictional concentration recorded in section 15. Counting these as four separate confirmations of anything would overstate the evidence by roughly a factor of four.

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

Sources

The readiness assessment is marked as a reported observation rather than as a measurement. No official comparative readiness statement exists, the available trackers use their own criteria, and we use the point only to establish that variation exists rather than to rank any state.

In this series

—   Previous: Article 3, The Simplification Response — what an institution does when it cannot match the variety of what it faces.

—   Next: Article 5, The Measurement Trap — what happens to a system that can see everything and cannot tell when it is wrong.

—   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.