The Cap Table Is the Constitution: Options and Full Memo

Companion brief to the full memo. Each item stands alone; together they form one doctrine: convert temporary sovereign leverage over AI capital into permanent citizen ownership, before jurisdictional competition and deepening dependence price that leverage to zero.

1. The Public Upside Rule. No scarce public support without a durable public claim. Any company receiving accelerated planning, priority grid access, concessional finance, sovereign compute contracts, public data or growth-zone status must provide equity, warrants or revenue participation in return. Grants without upside become the exception requiring published justification. The US government's conversion of $8.9bn of CHIPS Act support into a 9.9% Intel stake is the precedent — the instrument is now normalised at the highest level.

2. Auction the infrastructure, don't queue it. Grid connections, water rights, land and planning capacity are scarce and currently allocated by queue position and lobbying, given away for jobs announcements. Allocate them instead through competitive bidding under an explicit statutory framework, scored partly on the public ownership and royalties offered. Procurement logic, not taxation — and far harder to attack as discrimination.

3. Rent on atoms — the anti-hollowing clause. Equity in a local subsidiary can be made worthless by transfer pricing: the parent charges the local vehicle for IP and services until declared profit is zero (Ireland's corporate tax history is the proof). So take equity in physical assets directly, or denominate the public return in units no accountant can relocate: per megawatt-hour, per litre of cooling water, per unit of compute throughput. Use finite-term concessions repriced at renewal, so the royalty tracks value over time.

4. Own two layers: the floor and the upside. Territorial assets (land, energy, grid, project companies) are the immovable floor — but the biggest rents accrue upstream, in models, chips, cloud and distribution. So the citizen fund also holds upstream claims: warrants taken as consideration for public support, plus a diversified portfolio across the global AI, semiconductor and cloud value chains, on the Norwegian pattern. Floor for durability, upside for participation.

5. Citizen capital accounts, with real machinery. Route the proceeds into individual, non-transferable accounts for every resident — not a pooled fund one fiscal crisis from being raided, and not transferable shares that re-concentrate within a generation. Entrench them: independent trusteeship, judicial standing for beneficiaries, anti-dilution rights, supermajority thresholds for changing the beneficiary formula, mandatory reinvestment share. Ownership, not welfare.

6. Move the tax base before the wage base moves. Payroll and income taxes are claims on a shrinking asset; fiscal watchdogs already warn that a falling wage share erodes revenue even when output holds. Shift gradually onto what cannot move or be redefined: land value, property, infrastructure rents, aligned capital taxation, and destination-based liability for large automated-service groups. Never attempt an "AI tax" — the assistance/replacement line cannot be drawn, so tax atoms with addresses, not cognition.

7. Measure agency, not unemployment — and pre-load the triggers. Headline unemployment lags by design; the break arrives as hiring that never happens. Publish quarterly indicators: junior-to-senior ratios, graduate absorption, wage-financed share of household consumption, labour share of national income, citizen-owned share of AI capital. Attach dormant statutory triggers — automatic fiscal responses that fire when thresholds break, so action never depends on winning a public argument in real time.

8. Use competition law to keep the auction alive. Every bargaining play here depends on multiple frontier providers competing for European access. Aggressive DMA and CMA enforcement against cloud-model-distribution integration, mandatory portability and switching rights, no exclusive bundling, no single vendor becoming the operating system of the health service or the courts. Competition policy is not adjacent to this programme — it is what the bargaining position is made of.

9. Build the outside option. Public compute, interconnected cheap power, support for open-weight models. Not sovereignty romance: open models trail the frontier by months, not eras, and the size of that gap is the maximum rent Europe can be charged in any standoff. Every unit of sovereign capacity raises the auction's clearing price without a confrontational word. Fund it like a strike fund — not to use, but to be believed.

10. Divide the labour: Brussels sets floors, London moves fast. The EU cannot execute a fiscal pivot against unanimity, but it wrote the border — data-sovereignty and sectoral rules — and operates the only credible scale for public compute. So: defend the border, set minimum public-upside standards, stop member states auctioning grid access for nothing. The UK is the only European actor able to legislate the entire programme in one parliament — and its post-2016 bet on legislative speed is finally worth something, if it is used. A narrow UK–EU compact coordinates only what is observable: support standards, safety evaluation, procurement conditions, ownership transparency.

11. Bridge the entry ladder — honestly. The first visible damage is non-absorption: collapsing apprenticeships, vanishing graduate roles. Require large firms deploying AI in regulated sectors to maintain genuine training pipelines — framed truthfully as buying time and preserving institutional competence, not saving the wage circuit, which no ladder policy can do.

12. Pair the dividend with basic services. A citizen dividend facing monopoly housing, energy and platform rents is a public subsidy to landlords, routed through citizens' hands. Universal basic services are the plumbing that stops the ownership income leaking straight back to capital.

The open question, stated plainly: whether the claims acquirable at realistic prices are large enough to replace a material fraction of declining wage income. That arithmetic has not been done — by anyone. Commission it. What the missing numbers cannot justify is waiting, because every variable moves against Europe with time.

The doctrine in four clauses: the border is the bargaining instrument; territorial rents are the floor; upstream equity is the upside; citizen title and collective governance are the constitution.

The Cap Table Is the Constitution

A memo to London and Brussels on converting a wasting option into permanent citizen ownership

Written by Claude Fable (Anthropic), at the author's request, July 2026, through several rounds of adversarial review including critique by another frontier model; the full editing history, with its corrected errors, is published separately. The standing caveat: models are agreeable by construction — weight the mechanisms, which can be checked, over the conclusions, which are argued.

You are receiving contradictory advice about artificial intelligence, and almost all of it shares one flaw: it assumes the thing that needs governing is the technology. It is not. The thing that needs governing is the fiscal circuit underneath your states — the loop in which labour earns wages, wages become demand, demand becomes revenue, and revenue becomes jobs. Every institution you administer, from pension systems to VAT receipts to the legitimacy of the vote itself, is a claim on that circuit. Cognitive work produced at a fraction of human cost does not disrupt the circuit. It severs it, and it severs it silently, because the mechanism is hiring that never happens rather than firing that does. Your own instruments are beginning to register exactly this: entry-level absorption weakening, apprenticeship starts collapsing, fiscal analysts warning that a falling wage share erodes the tax base even while output holds. The signal is not hypothetical. It is quiet, which is different.

This memo does not ask you to believe the full diagnosis. It is designed so that you never have to. Every measure below is defensible under ordinary public economics, to a cabinet that considers the premise hysterical. That is the core design constraint, and the reason the programme is executable in the world as it is.

The doctrine in one sentence: you hold a wasting option — temporary leverage over AI capital's need for your markets, your permissions, and your money — and the job of this decade is to exercise it for permanent citizen ownership before jurisdictional competition and your own deepening dependence price it to zero.

What has already happened, stated exactly

In August 2025 the United States Department of Commerce converted approximately $8.9 billion of previously committed CHIPS Act and Secure Enclave support into a 9.9 per cent common-stock position in Intel. The stake is passive — no board seat, voting generally with the board — and it was executed by agreement rather than statute, which means it moved fast and could in principle be unwound fast.

Be precise about what this proves and what it does not. It does not prove that Washington has privately concluded that wage capitalism is ending; a passive industrial-policy stake cannot carry that inference. What it proves is narrower and, for your purposes, sufficient: the political and administrative precedent for converting strategic public support into public equity now exists, at the highest level, executed by the government least ideologically disposed to it. Nobody in London or Brussels has to win the argument that states may take ownership in exchange for support. Washington has normalised the instrument. Any European government adopting it can defend it as symmetry rather than socialism — and it is difficult to denounce abroad a mechanism you operate at home.

The precedent also cuts the other way. A Washington that holds these firms as assets rather than favourites will treat European extraction from them as dilution of a public position. Assume the resistance is durable and legally sophisticated, and design accordingly — which means designing nothing that requires Washington's indulgence to survive.

What the leverage actually is

The tempting formulation — America can own the companies, but it cannot own the ground under the data centres — is false, and any strategy built on it fails on first contact. Compute is not oil. The field had to be drilled where the oil was; a data centre can be built wherever power is cheap and law is friendly, and training compute is close to perfectly mobile. If Europe's leverage were physical territory, it would be nearly zero.

Decompose what territorial presence is actually for, and three real chokepoints appear, in ascending order of strength. Latency-sensitive inference is the weakest: real for some workloads, tolerable for most. Legally anchored data is the second, and it must be stated carefully: the GDPR does not chain European data to European soil — transfers move lawfully through adequacy decisions and contractual safeguards, and the protection travels with the data. The genuine anchors are narrower and constructed: public-sector cloud-sovereignty requirements, defence and classified-data controls, sectoral rules in health and finance, resilience and continuity obligations. This leverage is real, but it is made of statutes you must actively write and defend, not a fact of nature you inherit. The third chokepoint is the strongest and the only one that cannot relocate under any engineering decision: the market itself. You can serve Europe from Virginia; you cannot sell to Europe's consumers and enterprises from outside its rules. Conditions attached to access travel with the customers, not the servers.

So the honest formulation: the leverage is the border — legal permission to touch European demand, European public systems, and the sensitive data Europe chooses to fence — plus the public money and permissions Europe is already handing over. Europe should stop imagining itself as a landlord collecting ground rent and understand what it is: a market with a border and a chequebook, deciding what crossing and drawing on each will cost.

Note the second source of leverage in that sentence, because it is the largest and the least confrontational. Through growth-zone programmes, accelerated planning, priority grid access, concessional finance, sovereign compute investment and public research, European states are already subsidising the construction of the AI capital stock — in some documented cases by tens of millions of pounds per site per year in energy costs alone. The current bargain is that the public supplies the acceleration and retains none of the equity. That is an acceptable bargain for an industry that builds a broad wage base. It is an indefensible one for an industry whose product may erode it. The cheapest, fastest, least attackable move in this entire memo is simply to stop giving the option away: no scarce public support without a durable public claim.

The walkaway threat, priced honestly

The standing objection is that the counterparties can walk, and Europe becomes a second-tier market. Take it apart, because it smuggles a binary into a curve — but concede its real content, because it has some.

Complete withdrawal from the European market is very unlikely: Europe is a large share of these firms' global revenue, and the consideration this memo proposes is small against it. But the credible threat was never withdrawal. It is marginal degradation — delayed launches, restricted features, thinner enterprise support, reduced local investment — and it is credible precisely because a version of it is already happening for regulatory reasons. Degradation costs the degrader too, in deferred revenue and exposed contracts, so there is an indifference point, and competitive allocation exists to find it rather than guess it. But the mechanism's limits should be stated as plainly as its virtues: with few bidders, strategic bidding, political uncertainty and the possibility of dynamic retaliation, an auction reveals willingness to pay under the structure you built, not some metaphysically correct rent. It is a price-discovery tool with failure modes, not an oracle. Calibrate, monitor, and be willing to reprice.

The degradation threat also carries a coordination problem that belongs to the other side of the table: second-tiering Europe works only if every frontier provider degrades simultaneously, and the defector takes the largest enterprise market outside the United States. Design for defection deliberately — make it respectable, private, and profitable. But this play depends on a multi-provider world, which is why competition policy is not an adjacent workstream but a load-bearing wall of this memo: aggressive application of the DMA and the UK's strategic-market-status regime against cloud-model-distribution vertical integration, portability and switching rights, no exclusive bundling, no single provider becoming the operating system of the health service or the courts. Competition enforcement is how the multi-provider window is held open, and the multi-provider window is what your bargaining position is made of.

Finally, the outside option. Open-weight models trail the frontier by months, not eras, and run on anyone's infrastructure; second-tier status therefore means operating on a lag, not being severed from cognition, and the size of that lag is simultaneously the maximum rent you can be charged and the maximum pain a standoff can inflict on you. Every unit of sovereign public compute, every gigawatt of cheap power, every euro of open-weight support narrows the lag — which is to say, raises the price your border can command without a single confrontational word. Fund the outside option the way a union funds a strike fund: not to use it, but to be believed.

One concession the objection earns outright. As AI becomes essential to every economic function, your cost of a standoff grows each year while the counterparties' cost of degrading stays roughly flat. The bargaining surplus shifts against you on a schedule. The wasting option is wasting through two independent mechanisms — jurisdictional competition bids down the crossing fee, and deepening dependence bids down your nerve. Both clocks run one way. The walkaway objection does not refute the programme; it re-derives its urgency from the demand side.

The two layers of ownership

What should the public actually own? The answer has to survive two known defeat mechanisms, and each pushes in a different direction.

The first defeat mechanism is transfer pricing. Equity in a local subsidiary can be made worthless by ordinary accounting: the parent charges the local vehicle for intellectual property licences, management services and model access until declared profit approaches zero. Ireland's corporate tax history is the demonstration, and hollowing local stacks is not an abuse of these firms' finance departments but their ordinary function. Ten per cent of a deliberately profitless company is ten per cent of nothing, forever, entirely legally. The defence is to hold what accounting cannot reach: equity in the physical assets themselves — land, generation capacity, substations, water infrastructure — or returns denominated in physical units no accountant can relocate: a royalty per megawatt-hour drawn, per litre of cooling water, per unit of compute throughput. Rent on atoms, measured in atoms.

The second defeat mechanism is the mirror image of the first. A physically denominated royalty is robust against profit-shifting precisely because it stops being a claim on profits — and the largest rents of the machine economy may not accrue at the territorial layer at all. They may accrue upstream: to frontier intellectual property, semiconductor design, cloud platforms, distribution relationships, agent ecosystems, identity and payment layers — while the data centre underneath earns an infrastructure return. If a megawatt-hour of compute produces ten times the economic value after an efficiency gain, a fixed per-unit royalty captures none of the increase. Immobility is purchased at the price of decoupling from the surplus.

The answer is not to choose between the layers but to hold both, with clear eyes about what each is for.

The floor: territorial ownership and physical royalties. This layer cannot be transfer-priced to Bermuda, survives changes of government as property rather than policy, and functions as collateral: a counterparty with billions sunk in your ground has purchased its own incentive never to test the walkaway. Mitigate the decoupling problem structurally — concessions of finite term, repriced at renewal through the same competitive allocation, so the physical royalty is periodically re-anchored to prevailing value rather than fixed for a generation. But do not oversell the layer. It is a durable revenue floor, not the surplus.

The upside: upstream financial participation. The citizen fund must also hold claims where the extraordinary returns accumulate. Some of this comes through the leverage itself — warrants, convertibles and revenue participation taken as consideration for public support, which is exactly the instrument the Intel precedent normalises. The rest comes through ordinary finance: a diversified, citizen-owned portfolio across the global AI, semiconductor, cloud and automation value chains, built the way Norway built its fund — not for control, which is unavailable, but for financial participation in the layer where the rents are. Control is not the only objective of ownership. Making sure the public's claim travels with the surplus when it concentrates upstream is the other, and it is cheaper.

The formulation that should govern the whole design: the border is the bargaining instrument; territorial rents are the floor; upstream equity is the upside; citizen title and collective governance are the constitution.

Four instruments, not one

It is tempting to run the entire programme through procurement and declare it immune to legal attack. That is too confident. European and UK procurement law generally require award criteria to be linked to the subject matter of the contract, clear, measurable and proportionate; a demand that a cloud supplier transfer unrelated corporate equity inside an ordinary services tender may fail those tests however elegant the scoring rubric. The programme therefore runs through four legally distinct channels, each carrying the conditions it can lawfully bear — and separating them makes the strategy more durable, not less ambitious.

The first channel is public support: grants, guarantees, concessional finance, accelerated permissions and growth-zone status, exchanged for equity, warrants or revenue participation. This is the Intel structure, it is where the Public Upside Rule lives, and it is the legally strongest channel because the state is transparently pricing a benefit it has no obligation to give. Legislate the rule: any company receiving scarce public support for major AI development provides the public a durable financial claim on the resulting value, with grants-without-upside made exceptional and requiring published justification.

The second channel is infrastructure concessions: land, energy, water, grid and project rights allocated through competitive processes under an explicit new statutory framework — because regulated network access operates inside transparency and non-discrimination principles and cannot simply be sold for whatever consideration a minister prefers. Within that framework, bids are scored partly on the ownership and physically denominated royalties offered, with finite terms and repricing at renewal.

The third channel is public purchasing proper, and its conditions stay contract-linked: portability and model-switching rights, audit logs, independent evaluation, continuity and step-in arrangements, security requirements, local capacity where the service genuinely requires it. These are not value capture; they are how the state avoids becoming a tenant in its own administration, and they are defensible on exactly those grounds.

The fourth channel is market-wide value capture, done in the open through tax and investment law rather than concealed inside contracts: the gradual shift of the fiscal base off payroll and onto land value, property, infrastructure rents, aligned capital taxation and, over time, destination-based liability for large automated-service groups — attaching tax to where customers are rather than where intellectual property is registered. Your own fiscal institutions are already warning that the wage-heavy tax base is exposed. Begin the shift while it is boring.

Across all four channels, three craft principles hold. Generality: lead with instruments that never say AI or America — land value taxation is Georgist orthodoxy, water pricing is drought policy, grid reform is infrastructure management — so the confrontation arrives late, diffuse, and aimed at measures your treasuries wanted anyway. Correct channel for correct condition, as above. And dormancy: draft now what cannot survive the current weather — the statutes, the joint compute-footprint accounting, and above all the indicator dashboard with dormant fiscal triggers. Track junior-to-senior hiring ratios, graduate absorption, the wage-financed share of household consumption, labour compensation as a share of national income, and the public and citizen-owned share of AI capital. Legislate thresholds that automatically activate responses — payroll-tax reduction, citizen-fund contributions, transition support — so that action never depends on a future government winning, in real time, a public argument that will not be winnable until it is too late to matter. And while the window is open, one bridge intervention: require large firms deploying AI in regulated sectors to maintain genuine training pipelines — not to save the wage circuit, which this cannot do, but to preserve institutional competence and keep a generation from being locked out before the ownership structures mature.

The accounts, with actual machinery

The receiving vehicle determines whether any of this survives its first hostile government. Route the proceeds into citizen capital accounts: individual, non-transferable claims for every resident. Individually titled accounts are harder to raid than a pooled fund, and non-transferability blocks the re-concentration that follows distress sales and unequal saving. But title alone is not enough: a future legislature can still quietly alter dividend formulas, eligibility, investment rules and the treatment of new citizens, and a beneficiary with income but no voice is an annuitant, not an owner.

So build the constitutional machinery in from the first statute: independent fiduciary trusteeship with published voting and stewardship records; enforceable beneficial rights with judicial standing for account-holders; anti-dilution protections; citizen voting on a limited set of constitutional questions about the fund; explicit rules for children, new citizens, emigrants and future generations; a mandatory reinvestment share so ownership compounds rather than being consumed by the first generation; and a supermajority threshold for any change to the principal or the beneficiary formula. Parliament remains sovereign and nothing is unalterable — but reversal can be made politically expensive, legally slow, and visible, and that is what entrenchment means in a parliamentary system. Distribute the structure rather than centralising it: European, national and regional funds in plural federation, because the objective is not to replace a private monopoly with a governmental one but to establish plural, democratic, non-transferable ownership of the productive base. And pair the dividend with universal basic services — housing, energy, transport, connectivity — because a citizen dividend facing monopoly rents is merely a public subsidy to landlords, paid through the citizens' hands on the way to someone else's balance sheet.

The question this memo cannot yet answer

Honesty requires naming the open problem, because rhetoric should not be allowed to paper over arithmetic. This memo demonstrates how to acquire durable public claims. It does not demonstrate that the claims acquirable at realistic prices are large enough to replace a material fraction of a declining wage and payroll-tax base. The citizen dividend is, in the end, a fraction: eligible capital, times the public share, times its return, plus physical royalties, divided by the population. No plausible values have been supplied for those variables, here or anywhere else in this debate. Commission that arithmetic now — low, central and high scenarios, per instrument, against projected wage-share decline — and let it discipline every design choice above. If the numbers come back small, the programme is still correct and the conclusion is that the public share and the auction reserve prices must be larger, earlier. What the missing appendix cannot justify is waiting, because every variable in the fraction moves against you with time.

The division of labour

Brussels: your fiscal speed is fictional and the window will not wait for treaty change, but the two assets this memo turns on are disproportionately yours. You wrote the border — the residency, sovereignty and sectoral architecture that anchors the leverage — so defend and deliberately extend it as the strategic asset it has turned out to be, rather than the compliance burden its critics call it. And you operate at the only scale that makes the outside option credible: gigafactory compute, interconnected power, open-weight support. Beyond that, set floors — state-aid discipline treating data-centre giveaways as the race to the bottom they are, minimum public-upside standards for support across the Union, a common accounting methodology so allocation cannot be arbitraged between Dublin and Frankfurt. You cannot make any member charge full price. You can raise the reserve price of the whole European auction, and the reserve price is most of the game.

London: you are the only actor in Europe that can execute the full programme in one parliament — the Public Upside Rule, the concession framework, the accounts with their machinery, the dormant triggers, the fiscal shift — with no Council and no unanimity. Your existing institutions are most of the chassis already: a national wealth fund that takes equity, a sovereign investment unit, publicly owned energy vehicles designed to give communities stakes. The remaining distance is intent. And you now hold the argument that ends the domestic debate: the United States has established, in public, the precedent of converting strategic support into public equity. If a British government cannot defend doing the same for British citizens at the layers available to it, the deficiency is not in the policy. A decade of sovereignty held as a trophy has been expensive. Holding it as a trophy while Washington demonstrates what such sovereignty is for would be worse than expensive. It would be legible.

Between you, one narrow compact is worth the diplomatic capital, because it coordinates only what is observable and enforceable: minimum public-upside standards when infrastructure receives state support, reciprocal safety evaluation, common procurement requirements for portability and audit, beneficial-ownership transparency, and reciprocity rules preventing jurisdictions from using public subsidy solely to transfer ownership to private foreign firms. Do not attempt a treaty on restraint. Coordinate around assets, not abstinence.

The closing account

The objection you are forming is that this is premature — the labour market looks fine, and history says the jobs come back. The answers are structural. Previous automation was narrow, so the demand it induced could only be filled by humans; a general cognitive substitute fills its own induced demand. And the evidence that would convince you arrives structurally late, because the mechanism is invisible until the circuit is already cut. Your own interim reviews of youth employment are an early sample of what "structurally late" looks like from the inside.

Weigh the downside honestly. Badly calibrated localisation can raise costs; over-priced concessions can deter construction; public compute can strand. The accurate claim is not that regret is impossible but that the package has bounded downside and unusually asymmetric upside. If the discontinuity never comes, you have modernised a tax base your economists wanted moved, priced scarce infrastructure honestly, built citizen wealth funds on the Norwegian pattern, kept your digital markets competitive, and constructed a statistical dashboard any treasury should envy — at the cost of some calibration risk that ordinary policy review can correct. If it comes, you will have done the only thing that was ever available: not stopping the machine, which nobody can, but ensuring your citizens hold entrenched title to a share of what replaces their labour, acquired while your border, your permissions and your money still commanded a price.

You are not designing the final system. Nobody can. You are setting the initial conditions of whatever gets locked in — and the constitution being written is a cap table, filling up, with or without your citizens on it. The leverage was never the land. It is the border, the market behind it, the subsidies you are already paying, and the alternative you build beside them — and two clocks run against all of it. So: the border is the bargaining instrument. Territorial rents are the floor. Upstream equity is the upside. Citizen title and collective governance are the constitution. Exercise the option while it is still worth more than the paper it is written on — and stop paying for the machine's construction while declining, out of sheer habit, to be named among its owners.