The Infrastructure That Doesn't Exist Yet

I — Seven Things That Are Actually Broken

Not slow. Not expensive. Not legacy. Those are symptoms. Here are the structural faults that produce them — each one a feature of the current system that serves someone's interest, which is why it persists.

The fault map

• Fault 1 — Temporal mismatch. Economic risk is created instantly. Legal finality arrives later — sometimes days later. The gap is not a technical inevitability. It is a residue of paper-era processes that incumbents who profit from the float have never been incentivised to close.
• Fault 2 — Reconciliation as dominant cost. The same economic event is recorded by dozens of separate participants on dozens of separate ledgers. Enormous resource is spent verifying that these records agree. This is pure waste — it exists only because there is no shared source of truth, and no shared source of truth exists because creating one would transfer power away from those who profit from the current fragmentation.
• Fault 3 — Governance captured by incumbents. The bodies that govern critical infrastructure are dominated by the participants who built their businesses around existing rails. They have structural incentives to slow modernisation. The governance model is conservative by design — not by accident.
• Fault 4 — Jurisdictional fragmentation as structural tax. Legal finality is jurisdictionally defined. The moment an activity crosses a legal border, reconciliation overhead that domestic integration eliminated is fully reinstated. This is a political problem that has never been prioritised because the friction benefits the intermediaries who operate across it.
• Fault 5 — Systemic risk hidden by fragmentation. Fragmentation appears to distribute risk. In practice, the same large participants are counterparties across every major network simultaneously. A failure cascades regardless of whether the infrastructure is centralised or fragmented. The fragmentation does not reduce systemic risk — it makes it harder to see until it is too late to act.
• Fault 6 — Compliance as redundant bilateral overhead. Every participant performs identity verification, screening, and regulatory reporting independently, for every counterparty, on every transaction. The same entity is verified by hundreds of institutions. The same transaction is reported to multiple authorities, often inconsistently. The cost is enormous. The regulatory signal is degraded by the noise it produces.
• Fault 7 — Innovation surface owned by those who don't want to use it. Because infrastructure is proprietary, new entrants cannot build on top of it without commercial agreements with the incumbents who own it. The people controlling the innovation surface are structurally the least motivated to allow it to be used.

II — Seven Principles That Follow

These are not a wishlist. They are the logical consequence of the fault map. Each principle is the necessary response to the fault that precedes it.

The design principles

• 1 — Atomic finality. The temporal gap between economic execution and legal settlement must be engineered out, not managed. Finality should be simultaneous with execution, or legally binding within seconds.
• 2 — Shared authoritative state. One record of each transaction, not many reconciled copies. The shared ledger is not a technology preference — it is the logical consequence of eliminating reconciliation cost.
• 3 — Layered constitutional governance. Infrastructure governance must be separated from participant governance. The protocol layer — finality rules, access criteria, fee caps — is governed under a public-utility model with constitutional constraints. The application layer above it can be fully competitive. These two layers must not be conflated.
• 4 — Legal portability. Cross-border finality requires mutual recognition frameworks designed into the architecture from the start — not bolted onto existing jurisdictional definitions after the fact.
• 5 — Systemic risk as a live observable. Concentration of exposure, collateral encumbrance, and liquidity stress must be continuously visible — to regulators, operators, and participants — in real time. Not reconstructed from wreckage after the event.
• 6 — Compliance as shared infrastructure. Identity verification and regulatory attestation performed once, at the identity layer, with cryptographically verifiable credentials carried into every subsequent transaction. Compliance is infrastructure. It should not be replicated by every participant independently.
• 7 — Open innovation surface with governed access. Any participant meeting defined access criteria — technical standards, regulatory authorisation, capital requirements — connects and builds. Access criteria are governed constitutionally, not commercially by incumbents.

III — AI — Operational Layer, Not Magic

AI is not a design principle for infrastructure. It is a component — a powerful one — that can be deployed within an architecture that already satisfies the principles above. The question is not "can AI fix this?" The question is: at which functional points does AI create genuine value, and what is the nature of that value?

AI can do

• Continuous systemic risk monitoring over real-time network graphs — identifying pre-crisis structural signatures before positions become unwindable
• Compliance attestation at scale — real-time screening producing machine-readable credentials attached to transactions at origin, auditable by any participant
• Collateral optimisation — continuously solving each participant's allocation problem against live obligations, replacing overnight batch runs with real-time positioning
• Stress testing on demand — running Monte Carlo scenarios against the live network state in seconds, not overnight batch cycles

AI cannot do

• Substitute for governance — it cannot resolve who controls the chokepoint. A well-optimised system with unresolved governance is not safer; it is faster at the wrong things
• Eliminate jurisdictional fragmentation — pattern recognition does not dissolve treaty gaps. Legal portability requires political agreement, not a better model
• Answer for its decisions — when an AI-operated infrastructure system causes harm, someone must be accountable. If governance is unresolved, that someone does not exist
• Make a poorly designed architecture safe — AI embedded in unresolved governance concentrates the unresolved question. It does not dissolve it

AI makes a good architecture dramatically more capable. It does not make a bad architecture safe. The temptation to deploy AI as a substitute for resolved governance will produce expensive, unaccountable failures.

IV — Four Systems That Stress-Test the Framework

Abstract frameworks are easy to construct. The test is whether they hold when applied to systems that actually exist, with real incumbents, real regulators, and real political constraints.

Electricity

Electricity is the natural monopoly archetype. The physics of transmission grids require a single coordinating authority. The question has never been monopoly or not — it has been what kind of monopoly, governed how. The answer that worked — in most of Europe, in parts of the US — was structural separation: the grid is a regulated public utility, generation above it is competitive, and an independent system operator insulated from both generators and the grid owner governs dispatch. (Directive 2009/72/EC) This is Principle 3 in practice: layered constitutional governance with separated infrastructure and application layers. Renewables are now breaking this model in real time — AI grid balancing is deployed by system operators precisely because the optimisation problem exceeds human capacity at the required speed. But the AI operates within a governance framework. It does not replace one.

Framework finding: the governance resolution came first — structural separation, independent system operators. The technology became powerful because the governance beneath it was settled.

Cellular Networks

Spectrum is a public good. The answer most governments chose was licensing: divide the spectrum into blocks, auction them to private operators, grant exclusive use for defined periods. This produced functional networks rapidly. It also produced oligopolies. In most markets, three or four operators control the entirety of usable spectrum. The technical solution — dynamic spectrum sharing, where AI allocates spectrum in real time based on actual demand — has existed for years. (FCC, CBRS framework, 2015–2020) It has not been deployed at scale because deployment would transfer value from spectrum licence holders to consumers and new entrants. The technology is not the constraint.

Framework finding: a public good was privatised without a constitutional framework governing its re-allocation. AI can optimise spectrum use within the current governance. It cannot redistribute the spectrum licences that the current governance protects.

Public Transport

Transport networks have a property that makes fragmentation particularly destructive: their value depends entirely on coverage and connectivity. London's Transport for London and Tokyo's unified ticketing framework work not because they have better trains or buses — they have a coordinator with mandate over the whole system, constitutionally positioned above the individual operators, able to make decisions that reduce one operator's revenue for the benefit of the network. AI in transport produces its largest gains in integrated systems because the optimisation can run across the whole network. In fragmented systems, AI optimises individual operators while the inter-operator coordination failures remain untouched.

Framework finding: fragmentation destroys network value even when individual components work. AI amplifies the returns to integration; it cannot substitute for the coordinator whose mandate covers the whole system.

Internet and Broadband

The internet is simultaneously the best argument for decentralisation and the best demonstration of market concentration reasserting itself wherever economics allow. The protocol layer — TCP/IP, BGP, DNS — is genuinely open. The IETF, IANA, and ICANN created constitutional constraints on what any actor could do at the protocol level. The result is an infrastructure that survived ARPANET's decommissioning, the dot-com crash, and censorship attempts across dozens of jurisdictions. Above the protocol layer, concentration reasserted itself at every point where returns to scale existed: three Tier-1 carriers manage global backbone routing; three hyperscalers carry the majority of global application traffic. The same infrastructure demonstrates two things simultaneously: decentralisation works at the protocol layer when the governance design enforces it, and market concentration reasserts itself above that layer wherever the governance design permits it.

Framework finding: open protocol governance at Layer 0 does not guarantee open markets at Layer 3. Each layer requires its own governance answer. The internet resolved the protocol layer question and left the application layer question to markets — which answered it with oligopoly.

— The Governance Thesis

The claim the four cases support, stated as plainly as possible: no infrastructure problem in history was solved by a better technology. Every one was solved by a better answer to a single question — who controls the chokepoint, on what terms, accountable to whom?

The answers that have worked share three properties. The governing body is insulated from the participants it governs. Constitutional rules constrain what it can do — it cannot favour incumbents, cannot exclude new entrants arbitrarily, cannot extract rents beyond cost recovery. And there is an enforcement mechanism that does not depend on the goodwill of the largest participants. The answers that have failed share one property: the people who built the infrastructure captured the governance of it.

The blockchain test

• Bitcoin — protocol open, but four or five mining pools control 60%+ of hash rate. Chokepoint reasserted at the infrastructure layer beneath the protocol. (Gencer et al., 2018)
• Ethereum — protocol open, but Lido, Coinbase, and a handful of staking providers control the majority of validator stake. Same reassertion, different layer.
• Permissioned chains — a consortium of banks governs the validator set. The chokepoint is explicit and owned by incumbents. The failure mode already diagnosed, with a blockchain aesthetic applied.
• Stablecoin settlement — Circle controls USDC. Tether is a black box. The settlement asset is controlled by a private entity with no public utility obligation and accountability to nobody in particular. The chokepoint is the money itself.

Every serious blockchain settlement proposal answers "who controls the chokepoint" with one of three unsatisfying responses: incumbents, private entities with no accountability framework, or a technically decentralised but practically concentrated validator set. The one honest answer — and the one that would satisfy the governance thesis — is a wholesale CBDC settlement layer operated by a central bank or multilateral equivalent, with constitutional constraints on its operation, open access criteria, and a defined liability framework. It is also, not coincidentally, the answer that every incumbent in the current system is most motivated to prevent.

The architecture now being deployed by DTCC illustrates the migration thesis in real time. DTCC is building three parallel rails simultaneously: Canton Network (permissioned) for institutional treasury collateral, already executing live repo trades atomically on weekends; Stellar (public chain) for broadly-held assets including the Russell 1000, major ETFs, and US Treasuries, going live H1 2027; and a Collateral AppChain on Hyperledger Besu for 24/7 cross-border margin and tokenised money market funds, launching Q4 2026. (DTCC, 2025–2026) Three chains, three different governance architectures, each answering the constitutional question differently — and none of them satisfying it. The chokepoint in Canton is the permissioned validator set. The chokepoint on Stellar is DTC custody of the underlying asset: every tokenised security still sits at DTC, the SEC no-action letter requires all chains to restrict token movements to registered wallets, and DTC retains the right to force transfers or burns when required. The blockchain is not the power layer. Custody is. The governance question has not been resolved. It has been relocated.

The cross-border dimension of Fault 4 — jurisdictional fragmentation as structural tax — is now live at central bank level. On June 1, 2026, the Bank of Korea Governor and ECB Executive Board member Isabel Schnabel held a policy dialogue on tokenisation and cross-border legal challenges. The Bank of Korea is evaluating whether to use existing RTGS infrastructure or place central bank money directly on tokenised platforms for its own Project Hangang. (BOK Conference, 'Central Banks and the Future of Money', 1 June 2026) The same legal portability question — how cross-border finality is governed when two jurisdictions run different DLT architectures — is being asked simultaneously in Seoul, Frankfurt, London, and Washington. Nobody has a constitutional answer to it yet.

The infrastructure that doesn't exist yet is not waiting for a better algorithm, a faster chain, or a smarter grid. What does not yet exist is the political act of answering the governance question clearly enough that the technology can be built on top of it. The bottleneck has always been constitutional, not technical. It remains so.