Licensing inevitability
II franchise license today; state license in 3–7 years. Every technology that reached civilizational scale got licensed within a generation. THE central moat.
reinforces → cost of capital · FDE moat · demand capture
The Strategic Thesis · Intelligent Internet
Citizens are taxed in data, attention, and displaced labor while the returns route to cap tables that exclude them. This thesis makes the political and economic case for the Champion, a public-benefit company owned by the population of its jurisdiction, licensed as the utility AI actually is.
The case
The argument, in seven parts.
From the mechanics of the current system to the institutional form that replaces it. Read it straight through, or jump to the part you came for.
- 01The DiagnosisAI infrastructure is financed as software when it is utility-class. The neocloud loop, the demand wave, and the displacement shock.
- 02The Institutional AnswerThe Champion: locally licensed, broadly owned, AI-mediated, public-benefit by mission, and licensing inevitability on a 3–7 year fuse.
- 03What the Champion OperatesThe three-layer stack, the reference thesis, Zenith, the 80/20 engine, the layered platform, and the FDE workforce.
- 04Intelligent EconomicsValue relocates from access to verified benefit, contribution, and deployment, and five participation forms beyond equity.
- 05The Capital ArchitectureR1 at $1 pre-money, R2 at markup, R3 in public. Why every investor shows up, and why the 30:1 asymmetry makes it fundable.
- 06The Network~200 sovereign node operators coordinated by a Bitcoin-derived protocol, no federation, no council, nothing to capture.
- 07TimingSix reasons the window closes, Champion #1 in 6–12 months, and the structure that persists for decades.
Part I · The Diagnosis
A category error, financed.
AI infrastructure is being financed and operated as if it were enterprise software. It isn't. Once you accept the mismatch, most of the current weirdness in the market resolves: the "unsustainable" capex, the sovereign-AI drumbeat, the leveraged-compute vehicles, the stalled open-vs-closed debate. The mismatch is the source of every dysfunction, and the answer is not better software. It is the right institutional form for utility-class infrastructure.
Treated as: enterprise software
Short-duration capital
Globally homogeneous products
Concentrated in 4–5 hyperscalers
Quarterly earnings discipline
Open vs. closed as a product question
Actually is: regulated utility
Long-duration, capital intensive
Locally bounded, power, regulation, sovereignty
Natural monopoly within jurisdictions
Universal service implications
Open vs. closed as an ownership question
1,000–10,000×by 2032
The token-demand wave this system must absorb, growth that physically cannot fit in current hyperscaler geographies.
modelled§3
11.7%of US labor
The share of the US labor market whose work AI can already perform, and the displacement has begun, in a form UBI and retraining cannot answer.
reported§4
Part I · The Diagnosis
The hidden mechanics of the current system
The proliferating "neoclouds" are not technology companies, and the chip makers are not just selling picks in a gold rush. Follow the dollars around the loop, and the market's weirdness resolves into structure.
The closed economy
One dollar can lap this loop many times, equity out, chip purchases back, revenue recycled into new equity. Each lap inflates reported revenue and the implied valuation of everything inside.
The loop, whole
One dollar can travel this circuit repeatedly: the chip maker invests equity in AI labs and compute providers; they spend it back on chips; the revenue funds the next round of equity. Each lap inflates reported revenue and the implied valuations of everything in the loop. Click any node to inspect its role.
Closest analog: Standard Oil’s control of railroads, leverage over distribution, not the commodity.
Chip makers want twenty or fifty buyers, not four. Any credible non-hyperscaler buyer of compute is a strategic partner.
Figure 1
The closed loop: chip-maker equity funds labs and compute vehicles, whose chip purchases become chip-maker revenue, partially recycled into further equity. Each lap inflates reported revenue and implied valuations of everything in the loop. Attached: the neocloud leverage mechanic.
The status quo
What neoclouds actually are
The proliferating leveraged-compute vehicles are not technology companies. They are leveraged equipment-leasing structures wearing cloud-native marketing. The mechanic: borrow against GPUs as collateral, sign a multi-year take-or-pay contract with a hyperscaler or AI lab, use that contract as further collateral, repeat.
The hyperscalers benefit even more than the vehicles themselves. They are spending more on AI capex than any companies in history, and putting that capex directly on their own balance sheets would crush free cash flow and compress multiples. So they route demand through these vehicles, committing to long-term capacity contracts that the vehicles then finance against. The result is off-balance-sheet financing for AI infrastructure at massive scale.
Once you see it, the high valuations relative to gross margins make sense: these vehicles are not priced as software companies but as financial vehicles in an environment of structurally suppressed cost of capital for AI compute.
The chip-maker closed economy
Chip makers run a more sophisticated flywheel than the picks-and-shovels framing implies. They invest equity into compute providers and AI labs, giving those entities cash and credibility to commit to chip purchases, which generates chip-maker revenue, which gets partially recycled into further equity investments.
The same dollars flow around a closed loop multiple times. Each lap inflates reported revenue and the implied valuations of everything in the loop. Chip allocation becomes a strategic instrument: labs and clouds aligned with a chip maker's interests get more chips, faster. The closest historical analog is Standard Oil's control of railroads, not the oil itself, but leverage over distribution.
The Champion answer
The structural implication
Chip makers are structurally motivated to seed alternatives to hyperscaler monopsony. They do not want a world where four buyers control their demand; they want twenty or fifty.
Any credible non-hyperscaler buyer of compute is therefore a strategic partner for chip makers, not a competitor. That is the foundational alignment that makes the Champion model work.
$400B+/yr
AI infrastructure capex being deployed globally, more than any companies have ever spent, much of it routed off balance sheet.
reported§28
4–5buyers
The hyperscaler monopsony chip makers are structurally motivated to escape.
stated§2
20–50buyers
The demand base chip makers actually want, which is why credible new buyers of compute get seeded, not fought.
stated§2
They do not want a world where four buyers control their demand. They want twenty or fifty.
The takeawayChip makers are structurally motivated to seed alternatives to hyperscaler monopsony. Any credible non-hyperscaler buyer of compute is a strategic partner, not a competitor, the foundational alignment that makes the Champion model work.
Part I · The Diagnosis
The token demand wave
Capital pools matter only if the underlying demand is growing fast enough to justify them. It is: five multipliers compound to a 1,000–10,000× climb by 2032, and the biggest one only happens if someone deploys it.
The token-demand wave
Five multipliers stack toward the 2032 base case. They overlap, toggling a subset shows an illustrative product; the full set is pinned to the thesis range. modelled
1,000–10,000×token demand by 2032thesis range · multipliers overlap, not a literal product
Hover or toggle a multiplier to see its mechanism.
Consistent with frontier compute roughly doubling every six months since 2020. Illustrative subset products are capped at 10,000×; the hatched band is additive and structurally distinct, embodied demand cannot be optimised away.
Figure 2
The token-demand wave: toggle the five multipliers. Agentic workloads, reasoning, multimodal, and population-scale deployment compound toward 1,000–10,000× by 2032; embodied AI is additive and structurally distinct, you cannot algorithmically optimise away the body.
The status quo
Five demand multipliers
Agentic workloads (~30×): agent loops replace round-trips, with 10–100 internal calls per useful action. Reasoning models (~10×): long internal chains, moving from frontier-only to default. Multimodal expansion (10–100×): video, audio, image, and 3D at scale. Population-scale agent deployment (100–1,000×): continuous use, not chat, the multiplier the Champion model specifically unlocks.
The fifth, embodied AI and robotics, is additive and structurally distinct. Physical labour cannot be made more efficient the way compute can: you cannot algorithmically optimise away the body. Robotics is the most structurally durable demand driver in the stack.
Where it's heading
The geographic forcing function
The absolute compute needed by 2030 cannot fit in current hyperscaler-region datacenters. Power, cooling, and grid capacity in Northern Virginia, Dublin, and Singapore cannot 1,000× from here. Compute has to move to where the power is, and the jurisdictions with the power will demand Champion-shaped structures as the price of access.
Geographic distribution of AI infrastructure is a physical necessity that the demand wave forces. The Gulf states understand this and are already trading power access for Champion-shaped equity. The next round of jurisdictions will too.
The Champion answer
The deployment forcing function
The largest multiplier, population-scale agent deployment, only happens if someone deploys it. Hyperscalers will not deploy agents into seven billion people's daily lives: the political acceptance and regulatory risk are too high.
Champions are the only organizational form that can absorb population-scale deployment risk and reap population-scale demand. The largest pool of incremental demand requires an organizational form that hyperscalers cannot inhabit.
1,000–10,000×
Compounded token-demand growth by 2032, consistent with frontier compute doubling roughly every six months since 2020.
modelled§3
85.2Munfilled jobs
Korn Ferry's projection for 2030: $8.5T in unrealised annual revenue. A demographic fact, not a cyclical condition, and the floor under robotics demand.
reported§3
10–100heavy users
The compute each humanoid robot consumes, expressed in simultaneous heavy human users. Robotics demand has no DeepSeek-style efficiency risk.
modelled§3
Compute has to move to where the power is, and the jurisdictions with the power will demand Champion-shaped structures as the price of access.
The takeawayThe largest pool of incremental demand, population-scale agent deployment, requires an organizational form that hyperscalers cannot inhabit. That is a real arbitrage.
Part I · The Diagnosis
The coming shock
The people building the technology have started saying it out loud: the displacement of human labor by AI-first companies has begun, in a form for which the conventional answers: UBI, retraining, redistribution, are structurally inadequate.
The status quo
What the people watching most closely are saying
Emad Mostaque's frame: capital historically needed labor; AI breaks the loop. His framework traces four inversions, land → labor → capital → intelligence. Human cognitive labor doesn't go to zero in value; it goes negative. The timeline he stakes out: roughly 1,000 days.
Dario Amodei's quantification: 50% of entry-level white-collar jobs disrupted within five years, unemployment potentially spiking to 10–20%, with an "almost overnight" transition once business leaders see the savings. The early data tracks the warnings.
Where it's heading
The deeper structural fact
The conventional answers to this shock are inadequate. UBI, retraining, redistribution: none of them change the structural fact that the productive asset has concentrated ownership and the value flow excludes the population whose work is being displaced.
The central political pathology is taxation without representation. The current AI economy taxes citizens in data, attention, displaced labor, and shifted political agency, and routes the revenue to cap tables that exclude them entirely. That mechanism produces the same political response every time it occurs at scale, and right now the response is incoherent, because there is no institutional alternative.
The Champion answer
What Champions convert
The shock without Champions: foreign AI-first companies extract value from a jurisdiction's displaced workers, pay minimal local taxes, provide no local employment, and route all returns to concentrated foreign cap tables. Political crisis without institutional answer.
The shock with Champions: the same productive activity routes value to citizen cap tables, services to citizen lives, governance capacity to citizen-elected representatives, and contribution rewards to citizens who participate. Where does labor get capital when capital no longer needs labor? Labor owns the productive capital.
30–50%
Collapse in entry-level tech hiring in 2025. Wall Street is cutting ~200,000 roles concentrated in entry-level analyst positions.
reported§4
~55,000US layoffs
Attributed directly to AI in 2025. MIT studies show AI can already perform the work of 11.7% of the US labor market.
reported§4
5–10%of incumbent cost
The cost structure at which AI-first companies enter markets. The gap is unbridgeable in any reasonable timeframe.
reported§4
The current AI economy taxes citizens in data, attention, displaced labor, and shifted political agency, and routes the revenue to cap tables that exclude them entirely.
The takeawayWhere does labor get capital when capital no longer needs labor? In the Champion model, labor owns the productive capital, returns flow to citizens because citizens are owners, not because of any transfer mechanism.
Part II · The Institutional Answer
The Champion as civilizational navigator
A new kind of private company, locally domiciled, broadly held, AI-mediated, public-benefit by mission, built around a single economic primitive: sovereign token-demand aggregation.
The aggregation surface
A jurisdiction’s token demand is born fragmented, thousands of small buyers, no leverage. The Champion’s primitive is toown the aggregation.
- Citizens
- Government
- Hospitals
- Schools
- Enterprises
- SMEs
Fragmented jurisdictional demand becomes one owned demand surface with supply-side leverage, compute supply is in surplus, and surplus commodities negotiate poorly.
Figure 3
Sovereign demand aggregation: fragmented jurisdictional demand, citizens, government departments, hospitals, schools, enterprises, SMEs, becomes one owned demand surface with supply-side leverage over a compute market in surplus.
The economic primitive
The market has compute supply in surplus. What it lacks is the institutional form to aggregate jurisdictional demand, to bring citizens, government departments, hospitals, schools, enterprises, and SMEs into a single demand surface that can negotiate with global supply at scale.
Everything else, the financing structure, the protocol-level coordination, the moats, the valuation, follows from this primitive.
Locally licensed, protocol-permissionless
The Champion sits at the intersection of two architectures. Locally, it is a licensed entity: domiciled in its jurisdiction, regulated by its state, operating under its laws. Protocol-side, it is a permissionless node operator, like a Bitcoin miner, opting into the protocol by performing valid work, opting out by ceasing to.
These are not in tension. Bitcoin miners are already licensed by their jurisdictions for the parts of their operations that touch local law while participating in a permissionless protocol globally. Local licensing gives the Champion its moat and political franchise; protocol participation gives it interoperability and economic alignment with every other Champion, with no institutional coordination required, and the right to fork or exit intact.
Sharper fiduciary, not softer
The conventional argument for narrow shareholder-return maximization assumes shareholders want only financial return. For an entity whose shareholders are millions of citizens of a jurisdiction, that assumption fails. The citizens want financial return and their work to exist, their children to have prospects, their language served, their healthcare functional, their society flourishing.
A PBC structure with this mission is the most accurate possible representation of what the shareholders want. It produces superior decadal returns precisely because the owners' welfare and capital appreciation are inseparable. And the competitive advantage falls out directly: a foreign extractor cannot match the mission, because the mission is downstream of the cap table, the cap table downstream of the PBC structure, the structure downstream of the founding act. None of it is replicable without becoming the Champion.
The Champion answer
AI-mediated governance: the historical unlock
The historical reason ownership concentrates is partly that coordinating millions of small shareholders was operationally impossible. The coordination overhead defeated democratic depth, so companies concentrated ownership, and citizens lost standing in the institutions that shaped their lives.
AI mediation collapses that overhead. Radiant provides institutional memory at scale, every decision, input, and outcome recorded with provenance. II-Agents let every stakeholder participate at their own time and granularity. Automated oversight runs in real time. A parliamentarian can ask what a policy would do to the elderly in rural areas of her constituency and get a substantively useful answer in seconds, with citations and counterfactuals; a citizen can ask how a proposed change affects them personally.
Broad ownership becomes operationally viable in a way it has never been. The Champion is not just the navigation form for the intelligence age; it is the governance form for the intelligence age. The same institution that provides universal AI services enables universal AI representation.
$1pre-money
R1 opens the cap table to every local institution and every retail participant on identical terms, the widest local ownership base any company has ever had.
stated§5, §21
100%local
Citizen data processed and stored on locally-owned infrastructure under local law. No foreign entity has access or extraction rights.
stated§5
A public-benefit private company that champions the people of its jurisdiction: of the people, by the people, for the people.
The takeawayThe Champion is both a licensed AI utility in its jurisdiction and a permissionless protocol participant globally, a genuinely new institutional form, engineered to navigate a jurisdiction into the intelligence age.
Part II · The Institutional Answer
Licensing inevitability
Two licenses, two timelines. The II franchise license exists now and organizes the network. The state-issued AI provision license is coming, because every technology that became civilizationally essential got licensed, and AI hits all five preconditions at once.
The licensing arc
Every technology that became civilizationally essential got licensed.Every one. Click an era to inspect its regime.
1920s
Broadcasting
Radio reached every household within a decade, and the state responded: spectrum became a licensed public resource. The licensed broadcasters then dominated the medium for half a century.
- spectrum licenses
- public service obligations
- content standards
1980s
Telecommunications
Telephony became essential infrastructure, and every jurisdiction carved it into licensed operators, with interconnection duties and universal-service mandates as the price of the franchise.
- operator licenses
- interconnection
- universal service obligations
Always
Financial services
Banks, payments, insurance, securities, licensed in every jurisdiction on earth. Money is too consequential for unlicensed provision, and it has never been otherwise.
- banking licenses
- payments
- insurance
- securities
Always
Power generation & transmission
All generation and transmission is licensed. No jurisdiction, anywhere, lets an unlicensed operator run the grid.
- generation licenses
- transmission licenses
2026–2033
AI provision
projectedIntelligence provision follows the same arc, compressed. Operator licenses and sector licenses; universal service defined as a daily inference quota; audit rights; local data requirements; license fees.
- operator licenses
- sector licenses
- universal service = daily inference quota
- audit rights
- local data requirements
- license fees
Five structural conditions, all hold, several more strongly than any prior case
- civilizational reach
- information asymmetry
- sovereign interest
- concentrated provision: 4–5 companies, 2 jurisdictions
- national-security domain
No prior technology hit all five simultaneously. Not whether but when, state licensing arrives 3–7 years after substantial token-economy deployment.
The recurring pattern
technology becomes essentialstate imposes licensingfew licensed operators emergeoperators dominate for decadesthe license is the moat
Figure 4
The pattern is so consistent it is basically a rule: technology becomes essential → state imposes licensing → a small number of licensed operators emerge → those operators dominate for decades → the license is the moat. AI is next, on a 3–7 year fuse.
The two licenses
The II franchise license exists now. Issued by the Intelligent Internet network itself, it grants the right to operate as the locally-domiciled Champion in a given jurisdiction under the II open-source stack: protocol interoperability, common architecture, and exclusive territorial franchise rights. The pattern is how Visa member banks were chartered and how cellular operators received geographic franchises, except II's economic interest is a warrant rather than equity, closer to a GP carry than a parent-subsidiary relationship.
The state-issued AI provision license is coming. No major jurisdiction currently licenses AI provision at scale. That changes as token-economy effects ripple through labor markets, financial flows, and governance. The state license is what creates the multi-decade regulated-utility position the framework's economics rest on. Champions stack both.
Where it's heading
Why AI follows the pattern
Five structural conditions all hold for AI, several more strongly than in any prior case: civilizational reach; information asymmetry (AI systems make decisions citizens cannot inspect); sovereign interest (credit, healthcare, education, legal interpretation); provision concentrated in four to five companies in two jurisdictions; and recognition as a national-security domain in every major government.
No prior technology hit all five simultaneously. State licensing is not a question of whether, but when and in what form. The structure is predictable from prior regimes: operator licenses for token provision at scale, sector-specific licenses for sensitive workloads, universal service obligations, already pre-figured as the daily inference quota, audit rights, local data requirements, license fees.
Why licensing and not the alternatives
States could instead nationalize, criminalize, or fragment. Licensing has been the dominant response in every prior comparable case because it solves the state's actual problem: allowing private operation while retaining public accountability and durable regulatory authority. Nationalization captures the asset but destroys operational efficiency; criminalization forgoes the benefits; fragmentation produces regulatory chaos.
The framework's honest boundary: Champions are robust in licensing regimes, partly robust in hybrids, and not robust under pure nationalization or criminalization. Deployment concentrates in the jurisdictions whose politics favor the licensing response, most major economies, but not all.
The Champion answer
Why Champions get the licenses
Champions are advantaged in license acquisition for four reasons: local domicile is usually a license requirement, and Champions are locally domiciled by construction. Broad ownership creates political acceptability, the R1 cap table is, among other things, a license-acquisition asset. Compliance-native operation means designing for one regulatory regime, not a hundred simultaneously. And operating history through the formative period supplies credibility, citizen-relationship density, and visible public-benefit delivery by the time state licensing arrives.
A hyperscaler taking equity in a Champion is dramatically more politically acceptable than a hyperscaler seeking a direct operating license. This is why R2 capital works, and why licensed-infrastructure multiples, stable and predictable, are exactly what pensions, insurance, and sovereigns want at scale.
10%warrant
II's stake in each Champion, struck at R1, exercisable at R3, no voting rights until exercise. Alignment without control.
stated§6
3–7years
The window over which token-economy disruption forces every major jurisdiction to establish regulatory authority over AI provision.
modelled§6
The competitive question shifts from "can a Champion compete with hyperscalers?" to "can the hyperscaler obtain a license?" For many jurisdictions the answer is: only through a Champion.
The takeawayWhen state licensing regimes harden, the Champions that operated credibly through the formative window get the licenses. Other providers route through Champions, partner with them, or exit the jurisdiction.
Part II · The Institutional Answer
The historical analog
Electric utilities, 1900–1935. Telecom privatizations, 1984 onward. Critical infrastructure organized as locally-owned, broadly-held, sovereign-aligned entities, staged domestic anchor → strategic → public, becoming dominant enterprises for decades.
The template, five times over
British Telecom (1984): domestic retail anchor, then strategic, then international listing. NTT (1987) and Deutsche Telekom (1996): the same template, becoming multi-decade institutional fixtures. Reliance Jio (2016–2024): $20B+ of strategic capital from globally diverse partners, ~$100B+ of value within eight years of founding. Saudi Aramco (2019): the same staging at $1.7T.
The Jio template is the most directly relevant: a domestically-organized entity absorbed tens of billions from US tech, Asian sovereigns, and Western private equity; built national-scale infrastructure in years rather than decades; and reset the cost basis of telecom for an entire subcontinent, all without state ownership. That is the operational and capital template. What the institution actually runs is the three-layer stack that follows.
Staged capital, sovereign alignment, the precedents
| Era | Example | Template |
|---|---|---|
| 1984 | British Telecom | Domestic retail anchor, then strategic, then international listing |
| 1987 | NTT | Same template; multi-decade institutional fixture |
| 1996 | Deutsche Telekom | Same template |
| 2016–2024 | Reliance Jio | $20B+ strategic capital from Facebook, Google, KKR, Silver Lake, Gulf sovereigns. ~$100B+ value. Reset telecom economics of a subcontinent in five years. |
| 2019 | Saudi Aramco | $1.7T valuation. Domestic anchor + strategic + international float + captive demand + sovereign alignment. |
$20B+strategic capital
What Reliance Jio absorbed from Facebook, Google, KKR, Silver Lake, and Gulf sovereigns while resetting the telecom economics of a subcontinent in five years.
reported§7
$1.7Tvaluation
Saudi Aramco: domestic anchor + strategic + international float + captive demand + sovereign alignment.
reported§7
The takeawayNone of these are software comps. All of them work because of structure, not technology, and Reliance Jio is the direct operational template: national-scale infrastructure, tens of billions in strategic capital, ~$100B+ of value in eight years, without state ownership.
Part III · What the Champion Operates
The operating architecture
II provides the open-source stack Champions deploy. Three layers, each with a distinct role: one interface to act, one substrate to remember, one deployment layer to control, materialising into five products.
The operating stack
Three layers, one machine
Action on top, memory in the middle, ownership underneath. Select a layer to open it.
- Reasoningfirst-principles, verifiable, self-improving, carries the most weight
- Memorypersistent context across sessions, departments, decades
- Tools
- Browsers
- Sandboxes
- Files
- Schedules
- Workflows
- Outputmultimodal production and coordinated task execution
Chat was built for talking. II-Agent is built for shipping.
- II-42sparse retrieval foundation: dense embedding → SAE encoder → sparse atoms → inverted index → BM25+SAE fusion; you can see WHY a result was returned
- Commonsgoverns what agents are allowed to know: boundaries, provenance, review, freshness
- CommonGroundpreserves what agents produce: records, evidence, decisions, handoffs, the return path
- Contribution / Elevationreusable work passes review and re-enters Commons as governed shared knowledge
Search before acting. Preserve what is produced. Return what is reusable.
- Private Cloud
- Sovereign Cloud
- Edge
- On-Prem
- Global Scale
The charter without the machine is a mission statement. The machine without the charter is another open-source AI platform.
The product family
- II-Agentone agent, every kind of work
- Factorya sentence in, a complete production out, text, image, video, voice, music
- Geniilives where you live: messages, voice, desktop, browser
- Boardlya company OS for solo founders, coordinated by a CEO Agent
- Zenithcontinuous self-improvement of all four
Every component open-source under OSI-approved licenses.
Figure 5
The three-layer stack and the product family. II-Agent coordinates eight capabilities in one managed execution environment; Radiant decomposes into II-42, Commons, CommonGround, and Contribution / Elevation; the Champion layer deploys across Private Cloud, Sovereign Cloud, Edge, On-Prem, and Global Scale.
II-Agent: the action layer
The universal AI touchpoint the user owns: model-agnostic, brandable, localisable, composable. One managed execution environment where a trusted coordinator researches, designs, builds, and ships with you, connecting any model, any data source, any tool, within the user's own context.
The capability that carries the most weight is reasoning: first-principles, verifiable, self-improving, orchestrating frontier capability where needed and locally-fine-tuned open models for the bulk. The other seven are the operational nervous system. Chat was built for talking. II-Agent is built for shipping.
Radiant: the layer outsiders miss
Everyone knows what a model is. Almost nobody thinks about what happens to the work the model produces after the session ends. Radiant is the durable reference and coordination substrate: it begins with the minimum durable reference future humans and agents must be able to search, inspect, verify, and reuse, sources, evidence, decisions, constraints, approvals, handoffs, results.
II-42 turns dense semantic capability into sparse, indexable retrieval with diagnostic visibility: you can see why a result was returned. Commons governs what agents are allowed to know, provenance, boundaries, review, freshness. CommonGround preserves what agents produce. Contribution / Elevation closes the loop: reusable work passes review and re-enters Commons as governed shared knowledge. Search before acting; preserve what is produced; return what is reusable.
Champion: the ownership and deployment layer
Sovereign infrastructure that deploys, secures, and keeps the stack sovereign, locally controlled across Private Cloud, Sovereign Cloud, Edge, On-Prem, and Global Scale. This is the technology layer the PBC actually operates: one is the charter, the other is the machine.
A charter without the machine is a mission statement. The machine without the charter is another open-source AI platform, substitutable by next year's fork. Together they are sovereign AI that a jurisdiction actually owns and controls.
The Champion answer
The citizen wedge
The stack materialises into five products: II-Agent for work, Factory for creative production, Genii for daily life, Boardly as a company OS, and Zenith improving all four continuously. Together they cover the full surface of a citizen's interaction with AI, all feeding through Radiant, all accumulating governed reference inside the Champion.
The whitepaper makes the obligation concrete: every citizen receives an II-Account, a non-custodial identity and agent the citizen owns, and every Champion must honor a daily inference quota. Universal access is not a service line; it is a protocol-enforced obligation. Population-scale agent adoption is the demand base on which everything else compounds.
8capabilities
Reasoning, memory, tools, browsers, sandboxes, files, schedules, workflows + output, one action plane replacing the dozen disconnected tools a knowledge worker juggles.
stated§8
100–1,000×per citizen
Token-demand growth as the agent moves from chat to continuous deployment across government services, healthcare, finance, education, commerce.
modelled§8
Frontier models are upstream. Radiant is what turns them into accountable institutional systems. Without it, AI is a demo. With it, AI is infrastructure.
The takeawayCitizens use one agent ecosystem, and the Champion owns the relationship. Token demand per citizen rises 100–1,000× as the agent moves from chat to continuous deployment across civic workflows, the demand multiplier the Champion model specifically unlocks.
Part III · What the Champion Operates
Why reference is the scarce layer
As intelligence commoditises, the scarce asset shifts from intelligence itself to the governed, durable, shared reference that intelligence acts through. Intelligence is abundant, it can be copied, rented, or replaced. Reference has to be earned.
The return path
Production returns to the commons
The loop that makes AI a capability you accumulate instead of a tool you rent. Select a node.
Search Commons
Before any work begins, the agent searches Commons, the governed shared knowledge base, so every task starts from what the institution already knows.
The status quo
Work happens → context evaporates inside sessions and platform memory. Knowledge enters production; production does not return.
What was ephemeral becomes lasting. Every task makes the next task easier.
Figure 6
The return path: agents search Commons, work through II-Agent, leave records in CommonGround, and return verified outcomes through Contribution / Elevation. On proprietary platforms, the same work evaporates inside sessions, knowledge enters production; production does not return.
The status quo
What disappears today
Agents give papers, code, policy, and experience the capacity to act. But today, work happens and its search paths, judgments, failures, and handoffs vanish inside sessions and platform memory. The work is done; the context is lost; nobody inherits it.
This is the knowledge-level version of the extraction the thesis diagnoses at the institutional level: proprietary platforms capture the productive output of agentic work and let it evaporate.
The Champion answer
The return path, and why it compounds
Radiant opens the return path: what was ephemeral becomes lasting; what was private to a session becomes available to the next agent, the next team, the next year. Agentic AI converts from a tool you rent into a capability you accumulate, owned by the Champion's population-scale cap table, subject to its PBC mission.
This is why Champions compound rather than commoditise. Jurisdictional reference, records, evidence, decisions, handoffs, reusable results produced by local work, is produced locally, governed locally, and cannot be replicated by a foreign competitor. Without Radiant, Champions are locally-hosted AI: defensible on licensing grounds, substitutable on technology grounds. With it, the reference advantage compounds year over year.
A caution, stated plainly
Holding the reference is also the thing that most concentrates power, and openness is necessary but not sufficient to prevent that. An open stack that everyone converges on is still a single reference.
The safeguard is plurality, more than one maintained stack, switchable by the populations that depend on it, not transparency alone. On day one the Champion runs a single open stack; reference plurality is the direction of travel, and the network is designed so the cost of forking to an independent reference falls over time rather than rising.
5years
A Champion operating this long holds five years of governed institutional memory, decision provenance, and reusable capability no new entrant can match, regardless of model quality.
stated§9
Intelligence knows. Work acts. Shared reference endures.
The takeawayThe question that matters in the long run is not "who has the best model?" Models commoditise. The question is "who holds the reference?"
Part III · What the Champion Operates
Zenith and self-improvement
The obvious objection: open-source stacks fall behind closed frontier labs. Zenith is the answer, a system that proposes, tests, verifies, and ships its own improvements, with human approval at the gate, continuously.
Zenith · the self-improvement engine
The loop that improves the loop
Ten steps, one human gate. The highlight cycles on its own, or click a step to walk it yourself.
1/10Observewatch production traces
FrontierSWE · first benchmark evidence
bar length = dominance · lower rank is better
The harness itself is the competitive advantage. Open-source engineers plus Zenith compounds; frontier-lab engineering scales linearly.
Figure 7
The Zenith loop and the FrontierSWE result: the same model moves from fifth to first when the harness changes, average rank 5.53 → 2.06, dominance 68% → 92%.
The status quo
The question the thesis must answer
The thesis claims Champions will run competent intelligence on open models at infrastructure margins. If the open stack can't keep up with closed frontier labs, Champions lose the technology argument and compete on licensing and sovereignty alone, defensible, but weaker than the thesis needs.
The Champion answer
What Zenith does
Zenith adds continuous coding to the software delivery loop: observe usage and failure patterns, evaluate against quality criteria, propose improvements, code the changes, seek approval, test in isolation, verify against production behaviour. If something breaks, roll back; if it holds, sandbox for safety and ship.
It also answers a question the thesis identified but hadn't previously solved: how does one open stack adapt to ~200 Champions across different languages, regulatory regimes, and cultural contexts? Each Champion's deployment adapts to its jurisdiction through Zenith's improvement cycle, locally-tuned capability from a shared open-source base, without each Champion maintaining a frontier-scale engineering team.
5th → 1st
GPT-5.5 on FrontierSWE, moved from its default harness to II-Agent + Zenith, finishing ahead of Claude Code, Codex, Cursor CLI, and Gemini CLI.
reported§10
68% → 92%dominance
The first evidence that the continuous-improvement loop produces real capability uplift, not incremental polish. The harness itself is the competitive advantage.
reported§10
Not "open-source engineers vs. frontier lab engineers" but "open-source engineers plus Zenith vs. frontier lab engineers." The former compounds; the latter scales linearly.
The takeawayChampions running Zenith on competent open models converge on closed-frontier capability for competent workloads, without frontier-lab-scale investment. The 80/20 engine becomes more defensible over time, not less.
Part III · What the Champion Operates
Most AI work doesn't need frontier capability
The distinction that does most of the work in the thesis: frontier intelligence for the cases that genuinely need it; competent intelligence, interpretable, auditable, locally deployable, for the vast majority of useful work.
Two different products
Frontier intelligence is for novel scientific reasoning, complex multi-step analysis at the edge of capability, and specialized domains where marginal accuracy earns its cost, a meaningful but minority share of useful AI work. Competent intelligence is for the vast majority: drafting, summarizing, routing, classification, customer service, translation, education assistance, healthcare admin, government services.
Conflating them produces strategic confusion. The 80/20 engine, made precise: Champions run competent intelligence at infrastructure margins on the open stack and route to frontier only when it earns its keep, a split of roughly 80/20, or even 95/5.
The open stack is not optional for the public sector
Government workloads require interpretable models on auditable data with verifiable provenance. The closed hyperscaler stack, closed source on closed models on closed data, with no bargaining power, cannot meet those requirements. The open Champion stack meets them by construction: model provenance hashes recorded on-chain, training data from auditable corpora with consent flows, frontier access through standardized interfaces that commoditize the API providers.
The protocol's standardized model interfaces mean each Champion can route workloads to any of several frontier providers and swap on price and quality. Frontier labs aligned with human flourishing are natural partners; misaligned providers find their offerings commoditized away.
The Champion answer
Complements, not competitors
Champions are customers of frontier labs, not competitors, specifically of frontier labs oriented toward human flourishing and human-AI collaboration rather than autonomous AGI. Those labs build increasingly capable assistants; Champions deploy them to populations.
No single frontier lab can match the deployment footprint: ~200 Champions in steady state aggregate enormous competent-intelligence capacity, distributed, locally fine-tuned, regulatorily protected, protocol-coordinated. A frontier lab's organizational form is optimized for capability research, not population-scale deployment. Both forms operate at scale; neither displaces the other.
80–95%of workload
Common, deterministic, locally-served: competent intelligence on locally-deployed open models at infrastructure margins.
stated§18
5–20×cheaper
The Champion's per-token cost base for competent workloads, open weights, locally hosted, financed at jurisdiction-specific cost of capital. Structural, not transient.
modelled§13
$200 → $2,000+/citizen/yr
Per-citizen revenue scaling as the layered platform builds out on top of the 80/20 engine.
modelled§18
Open-source stack on open models on open data, with access to frontier AI through collective bargaining power.
The takeawayChampions capture utility-margin volume on the bulk and distribution-margin premium on the frontier slice. That margin structure is what makes free or near-free universal baseline service economically viable.
Part III · What the Champion Operates
The layered platform
The citizen relationship is the foundation. On top of it: FDE depth, then agent-mediated financial services, then robotics, then frontier technologies. Each layer is a multi-hundred-billion market, and each layer pays for the next.
The build sequence
Each layer pays for the next
Six layers, bottom up. The order is not optional, climb the staircase to see why.
Year 1
Universal Citizen Agent
II-Agent, Factory, Genii deployed jurisdiction-wide. Population-scale demand base. The foundation.
Built on 0 layers beneath it.
Revenue per citizen · per year
$200/yr
$200$2,000+
You cannot skip to robotics without the citizen relationship; you cannot get the citizen relationship without the agent.
That compounding is what makes a major-economy Champion worth$300B–$1T at maturity.
Figure 8
The five-tier build sequence: agent → FDE → financial services → robotics → frontier. Selecting a layer shows why the order is not optional, and how revenue per citizen ramps from $200 toward $2,000+ per year.
Robotics requires local infrastructure
A humanoid robot in Indonesia needs maintenance in Indonesia, training data from Indonesian environments, integration with Indonesian regulations and worker-safety regimes, support in Bahasa Indonesia, and parts supply chains for Indonesian conditions. It does not deploy from a hyperscaler datacenter in Virginia.
The economic shock of robotics depends entirely on local deployment infrastructure that does not currently exist anywhere. Whoever builds it captures the value, and the Champion's existing FDE corps, citizen relationships, regulatory standing, and integration depth make it the natural deployment partner.
Robots are agents: one workforce, one platform
A humanoid robot is an embodied AI agent. The same foundation model that processes scheduling tasks as a software agent controls physical manipulation in a warehouse. Champions deploy agents across the entire spectrum, pure software to fully physical, through the same FDE corps, fleet intelligence platform, financial services layer, and customer relationship.
The FDE who configured the hospital's scheduling agent on Monday demonstrates meal delivery to the humanoid robot on Tuesday and replaces its worn hands on Friday. At maturity, the robotics layer generates more Champion revenue than all other layers combined.
The Champion answer
Why robotics is the most durable revenue layer
As AI inference costs drop toward the marginal cost of electricity, digital-only AI revenue compresses and the software agent layer commoditises. But robotics has irreducible physical costs that do not compress: the body, the maintenance, the parts, the deployment, the insurance, the FDE labour.
A robot that costs $6,000 to build and $1M+ to operate over its lifetime concentrates value in the physical deployment layer regardless of what happens to the cost of intelligence.
The build sequence, each layer pays for the next
| Phase | Layer | What it does |
|---|---|---|
| Year 1 | Universal Citizen Agent | II-Agent, Factory, and Genii deployed jurisdiction-wide. Population-scale demand base. The foundation. |
| Years 1–3 | FDE Deployment Depth | Forward-deployed engineers embedded in government, enterprise, SME. Switching costs compound. |
| Years 2–4 | Agent-Mediated Financial Services & Commerce | Banking, payments, lending through the agent. Revenue per citizen scales ~10×. |
| Years 2–4 | Local Media & Knowledge Infrastructure | Public broadcasting partnerships, knowledge organization, archival, accessibility. |
| Years 3–5 | Robotics Deployment | Local deployment partner for global robotics manufacturers. Largest medium-term value pool. |
| Years 5+ | Frontier Technologies | Quantum, advanced biotech computing, next-gen capabilities through the same channel. |
$300B–$1Tat maturity
What the compounding build sequence makes a major-economy Champion worth.
modelled§14
$1–5Tmarket
The robotics deployment gap: Champion-shaped, because physical AI deployment is irreducibly local.
modelled§26
300–400Mworker-equivalents
What 100 million deployed humanoids add to the productive economy at 22 hours a day, a ceiling above the $60–70T human labour market.
modelled§14
The cheaper the brain, the more robots get deployed, the more deployment infrastructure is needed. Intelligence cost decline accelerates robotics deployment, which accelerates Champion revenue.
The takeawayYou cannot skip to robotics without the citizen relationship, and you cannot get the citizen relationship without the agent. Without the build sequence the maturity numbers look optimistic. With it they look conservative.
Where robotics value settles
The auto industry already ran this experiment.
Where the value settles
The profit pool is downstream
A century of auto-industry data, replayed for humanoids. Hover or tap a segment.
A century of auto-industry profit data
OEM assembly
~18% of the profit pool at 3–8% operating margins. A century of capital intensity for the thinnest slice.
The humanoid replay
$6,000the robot, hardware converging on $5,000–20,000 commodity pricing from smartphone/EV supply chains; AI compute just 1–3% of unit cost
$1M+lifetime Robotics-as-a-Service revenue generated per deployed unit
Who captures the lifetime value
OEM · <1%deployment infrastructure · 99%+
Over $10B has flowed into humanoid OEMs in 2024–2026. Approximately zero into the downstream infrastructure that captures60%+ of lifetime value.
Robots work 22 hours a day. 100 million deployed humanoids add the equivalent of 300–400 million workers, a market larger than the $60–70T human labour market it displaces.
Figure 9
A century of auto-industry profit pools: OEM assembly captures ~18% at 3–8% margins; the downstream layers, dealers, captive finance, aftermarket, fleet, insurance, collectively capture 60%+. Humanoid hardware converges on commodity pricing while deployment captures the lifetime value.
Part III · What the Champion Operates
Proof of benefit, and the workforce that delivers it
Government is not just the regulator; it is the largest single AI customer in any jurisdiction, structurally captive to locally-domiciled providers. And AI does not deploy itself: it deploys through forward-deployed engineers, the largest source of new white-collar work the transition creates.
Why public-sector AI is structurally captive
Five constraints lock public-sector AI to locally-domiciled providers: national-security workloads cannot route through foreign infrastructure; citizen data cannot leave the jurisdiction in most regimes; sovereign decision-making cannot be opaque, interpretability is required; procurement rules favor local providers, often explicitly; and language and cultural specificity matter more than global hyperscalers can serve.
Champions that deploy compute for verified public benefit also mine Foundation Coin through the protocol's consensus mechanism, an additional revenue stream on top of the contract itself. Proof-of-benefit demand strengthens the licensing argument (government licenses the entity it already procures from), anchors R1 and R2 capital, and builds the public-interest political franchise: citizens experience the Champion through better government services.
The FDE math
Palantir built a $300B+ business not on better software but on the willingness to embed engineers inside customer operations for years at a time. That model, high-touch, high-margin-per-customer, multi-year switching costs, turns out to be the right shape for any AI deployment at institutional scale. The Champion thesis applies it at population scale.
One FDE-led deployment generates $5–50M in annual revenue, takes 6–18 months to mature, and creates 5–10 years of switching costs, roughly one engineer per $2–10M of mature annual revenue. A major-economy Champion at $5–20B revenue needs 500–5,000 FDEs deployed locally. Hyperscalers cannot field this: embedding 2,000 engineers in government ministries is the opposite of self-serve software economics. A Champion can, because FDE economics are its economics.
The Champion answer
FDEs as the workforce-development answer
FDEs are not just a moat. They teach the institutions they deploy into, civil servants, doctors, lawyers, teachers, factory workers, how to work alongside AI. They are simultaneously the deployment infrastructure, the education infrastructure, and the transition infrastructure. Job creation, job transformation, and AI deployment are three views of the same activity.
In the robotics era the FDE's role expands from integrator to task demonstrator: an FDE in a care home shows the robot how to deliver meals by walking the route once. The FDE's domain knowledge, this corridor, these residents, these preferences, becomes training data no simulation can replicate. The FDE corps becomes more valuable in the robotics era, not less. A jurisdiction debating whether to host a Champion is also debating whether to host this scale of new skilled employment.
$1–10B+/yr
What a major-economy government becomes as an AI customer at maturity. Aggregate across major economies: $50–150B annually.
modelled§15
10–25%of tokens
Proof-of-benefit workloads as a share of total consumption in major economies, in early years, possibly the majority of Champion revenue.
modelled§15
100k–1MFDE positions
Across ~200 Champions at full network density: knowledge-intensive, well-paid, locally-rooted jobs that cannot be done remotely from a foreign hyperscaler.
modelled§16
The Champion's deployment workforce is the AI age's equivalent of the post-WWII civil engineering workforce that built electrical grids and telephone networks.
The takeawayGovernment anchor demand answers how a Champion survives its early years, the same answer that worked for telecom, utilities, and broadcasting. And every government interaction feeds Radiant: five years of governed records no competitor can replicate.
Part IV · Intelligent Economics
Intelligent Economics
When intelligence becomes abundant, value relocates, from scarce access to verified benefit, contribution, and deployment. The frame that lets sophisticated investors value Champions correctly, and the reason the financial services layer is bigger than it looks.
The value frame
Old economy: value extracted from scarce access, pay for the API call, the seat license, the subscription. Intelligent Economics: value created by verified benefit, contribution, and deployment, pay for the outcome, the contribution to the network, the institutional integration.
The protocol has its own economic engine: a Bitcoin-derived consensus mechanism in which compute deployed for verifiable public benefit mines the protocol's reward token, with local-jurisdiction currency layers respecting local sovereignty over taxation, privacy, and data residency. Champions earn protocol rewards by doing exactly the work each is built to do.
Financial services as transition infrastructure
What traditionally required 50,000 employees at a national bank, underwriting, compliance, servicing, support, fraud detection, regulatory reporting, requires AI agents plus a few hundred FDEs when agents handle 90% of back-office operations. Nubank (100M+ customers, ~8,000 employees), Revolut (50M+), and Toss (30M+) already prove the operating model. The Champion inherits it from formation.
The structural need is equally clear: as AI-first firms capture the productive economy, the population needs new mechanisms for capturing and storing value. Income increasingly arrives through agent-mediated work, contribution rewards, and ownership returns rather than salaries, and traditional banks built on traditional employment models will struggle to serve it.
The Champion answer
Structurally Champion-shaped
Champions have the trust (PBC, broadly owned, locally accountable), the reach (every citizen already has an II-Agent), the physical presence (FDE corps), the regulatory fit (financial services licensing regimes Champions are suited to occupy), and auditability by construction for credit and underwriting decisions.
This converts the financial services layer from "another revenue stream" into the economic continuity infrastructure for the entire transition: the institutional rails for a population moving from wage-labor to ownership-and-contribution.
$5T
The network-level difference between valuing Champions as plain utilities and valuing them as the demand-side rails of Intelligent Economics.
modelled§17
100M+customers
Nubank serves them with ~8,000 employees. AI-native financial services at population scale is a solved operational problem, the Champion inherits the solution.
reported§19
90%of back-office
What agents handle in AI-native financial delivery, what needed 50,000 employees at a national bank needs agents plus a few hundred FDEs.
modelled§19
Champions are the transition banks for the intelligence age.
The takeawayWithout the Intelligent Economics framing, Champions look like utilities at 8–15× EBITDA. With it, they are the demand-side rails of a new economic organization at 10–25× blended multiples. The difference is $5T.
Part IV · Intelligent Economics
Beyond equity
A Champion's stakeholders have richer relationships than equity captures. Citizens are owners and users and contributors and governance participants and recipients of universal service, five real economic relationships, and equity captures only the first.
One citizen5/5
holds every active form at once
Equity ownershipinstitutional capital
Contribution rightsproductive human capital
Network access rightscross-border commercial capital
Governance participationpolitical capital
Universal accessregulatory & citizenship capital
5 capital pools ·5 moats ·5 sources of political durability
A single citizen can hold all five simultaneously.Each rewards a different kind of relationship.
Figure 10
The five participation forms, equity, contribution rights, network access, governance participation, universal access, compose. A single citizen can hold all five simultaneously; each rewards a different kind of relationship and taps a different capital pool.
The five forms
Equity ownership, voting, dividends, capital gains. Necessary but not sufficient. Contribution rights, a verified claim on the value your contributions generate: data with consent, attention, feedback, training signal, deployments; the protocol's consensus mechanism is the implementation. Network access rights, use any Champion's agent when traveling, deploy through any participating jurisdiction, run workloads on cross-Champion compute markets.
Governance participation, citizens' voice on universal service, worker participation on deployment and FDE policy, public-interest seats, and protocol-level proposal-and-implementation for the few decisions that touch the shared protocol. Different from equity voting: equity voting decides management; governance participation decides how the institution operates. Universal access, every citizen has access regardless of ownership or contribution, as a protocol-enforced daily inference quota. Not free, but not gated by ownership, the pattern of electricity, telecom, and broadcasting under licensed regimes.
The Champion answer
How the forms compose
A given citizen can hold all five simultaneously: equity in the Champion, contribution rights from active participation, network access for protocol-wide services, governance voice on operations, universal access as a citizenship right.
Each form accesses a different capital pool: equity → institutional capital; contribution rights → productive human capital; network access → cross-border commercial capital; governance → political capital; universal access → regulatory and citizenship capital. Across the network the forms compose further, a network equity index at R3, cross-Champion contribution rights, tiered network access, and universal network services guaranteeing baseline access across all participating jurisdictions.
Contribution becomes capital.
The takeawayFive capital pools instead of one. Five moats instead of one. Five sources of political durability instead of one.
Part V · The Capital Architecture
The three-round structure
Three rounds in sequence, each serving a different purpose, each attracting different capital, each creating aligned upside for a different stakeholder. The staging is load-bearing: no single round can do what the three together accomplish.
Nobody is trapped, nobody is squeezed.
R1 · Local capital$1 pre-money
- Who
- Citizens, retail, universities, pensions, foundations, sovereigns, local strategics, same price, same terms for everyone.
- Purpose
- Establish the widest local cap table any company has ever had in its jurisdiction.
- Key insight
- Not price discovery, political franchise establishment.
R2 · Strategic & long-term capital5–10× markup
- Who
- Sovereign wealth funds, pensions, insurance general accounts, chip makers, compute providers, local strategics. Hyperscalers may participate but are not load-bearing.
- Purpose
- Funds compute commitments, citizen agent rollout, FDE buildout.
- Key insight
- The markup means R1 holders sit on paper return, politically essential.
R3 · Public listingglobally accessible
- Who
- Pension funds, mutual funds, ETFs, retail worldwide, sovereigns at index scale.
- Purpose
- The Champion becomes a durable public institution.
- Key insight
- A new asset class is created.
Compute-for-benefit: equity on identical terms, plus a matched compute credit deployable to the investor's public mission.
University endowment
Puts $50M into R1 → R1 equity on identical terms plus $50M of inference & training compute credits deployable to its research, teaching, and library digitization.
Sovereign fund
A $500M commitment works the same way, deployable to public-health analytics, infrastructure planning, language preservation.
An additional elected utility return, not a different equity deal.
Figure 11
Three rounds, from the jurisdiction up: R1 local at $1 pre-money on identical terms for all; R2 strategic patient capital at 5–10× markup; R3 public listing, globally accessible. Public-benefit R1 investors can elect the compute-for-benefit return.
The compute-for-benefit return
R1 is more than political franchise for public-benefit institutions. In addition to equity terms identical for everyone, public-benefit institutions investing in R1 can elect to receive equivalent-dollar compute credits back, deployable to their own public-benefit work, an additional utility return alongside the equity stake, not a different equity deal.
A university endowment puts $50M into R1 and receives R1 equity on the same terms as every other participant, plus $50M of inference and training compute for its research, teaching, and library digitization. A sovereign fund's $500M works the same way, deployable to public-health analytics, infrastructure planning, language preservation, citizen-services modernization. The cap table becomes broader and more durable because the institutions that shape long-term political legitimacy have direct economic reason to participate.
Why each investor shows up
Sovereign wealth funds get decades-duration AI exposure plus sovereignty and economic development, supporting $300–800B of cumulative R2 capacity over the next decade. Pensions and insurance get long-duration, infrastructure-shaped, AI-correlated assets that fit a liability profile no other vehicle fills, against $95T+ of combined AUM underweight AI. Chip makers get demand diversification away from hyperscaler monopsony, worth $5–20B of effective subsidy per major Champion. Local strategics lock in their position in the new layer. Citizens and retail are already in at R1; the R2 markup pays the political franchise.
The Champion answer
Hyperscalers are optional
The Champion model does not depend on hyperscaler capital, cooperation, or benevolence. R2 closes on patient capital alone, with substantial oversubscription. Where hyperscalers do participate, the G42 template is real: Microsoft's $1.5B into the UAE's sovereign-aligned champion, with multi-year compute relationship and US government sign-off.
If hyperscalers cooperate, the Champion gets additional capital and supply. If they compete, the Champion still has more than enough patient capital, and the licensing arbitrage works in its favor. The structure does not require competitor goodwill, a critical robustness property.
$1pre-money
R1: same price, same terms for everyone. Universities subscribe on the same terms as retail; sovereign funds on the same terms as citizens.
stated§21
5–10×markup
R2: the markup puts R1 holders on paper return. Politically essential.
modelled§21
$5–20Beffective subsidy
Per major Champion, from chip makers and compute providers seeking demand diversification: guaranteed supply, deferred payment terms, minority equity.
modelled§22
Not price discovery. Political franchise establishment.
The takeawayEach markup creates aligned upside for a different stakeholder: R1 → R2 rewards citizens; R2 → R3 rewards strategics. Nobody is trapped, nobody is squeezed.
Part V · The Capital Architecture
Why this is fundable
There is roughly thirty times more institutional capital seeking AI equity exposure than there is AI equity available to absorb it. The capital is not the constraint. The structure to absorb the capital is the constraint. Champions are that structure.
Institutional pools seeking AI exposure · $Treported
Mutual funds & ETFs
~$70T
constrained by liquidity and benchmarks; massive latent passive demandPension funds
~$55–60T
long-duration liabilities; unable to reach AI targets through the tiny public AI universeRetail & DC pensions
~$40T
almost entirely locked out of private AI todayInsurance general accounts
~$35–40T
underweight infrastructure and private creditSovereign wealth funds
~$12–13T
most strategic pool: $600B–$1.2T in target AI allocations, growing 30%+ annuallyFamily offices
~$5–7T
most flexible mandate; buying every AI secondary at 30–60% premiumsEndowments & foundations
~$2T
small but reputationally influential as price-settersBar length on a square-root scale so the smallest pool stays legible; labels carry the reported ranges.
Demand vs supply · annualmodelled
Unsatisfied institutional demand
$3–6T stock · $500B–$1T incremental / yr
~30:1demand to supply
Current annual primary AI fundraising supply
$80–120B / yr
The capital is not the constraint. The structure to absorb the capital is the constraint. Champions are that structure.
Figure 12
The pools and the pipe: $3–6T of unsatisfied institutional demand and $500B–$1T of annual incremental demand, against $80–120B of current annual primary AI fundraising. The asymmetry is roughly 30:1.
Per-Champion economics, by tier
At maturity, a Champion's revenue draws from every layer of the platform: universal AI services and proof-of-benefit deployments as the base; enterprise FDE contracts and financial services for depth; robotics and local media for scale; frontier access, cross-border network services, and contribution rewards at the top. Illustrative revenue per citizen: $200–$2,000 a year.
The revenue base is durable because it rests on accumulated jurisdictional reference: a Champion's Radiant layer grows more valuable each year as governed records, decisions, and reusable results compound inside it. Revenue at maturity is a projection from an asset that appreciates with use.
Political durability varies by jurisdiction, say so
In stable regimes with mature regulatory frameworks (US states, UK, Germany, Japan, France, Korea, Singapore, Australia, Canada, Switzerland), the political-durability argument works as described, over multi-decade horizons. In stable regimes with less mature frameworks (UAE, Saudi Arabia, Israel), durability rests on state continuity. In less stable jurisdictions, the R1 cap table provides legitimacy but does not immunize against political attack.
R2 and R3 valuations should reflect this. At full network density: ~200 Champions, a mix of national and sub-national operators, 100% open-source foundation, aggregate network scale is comparable in order of magnitude to the global telecom sector.
Per-Champion economics by tier
| Tier | Examples | R2 post-money | At maturity |
|---|---|---|---|
| Major economy | UK, Germany, Japan, California | $25–40B | $300B–$1T |
| Mid-tier economy | Vietnam, Indonesia, Brazil, Nigeria | $500M–$5B | $20–$100B |
| Frontier markets | Smaller emerging markets | $100–$500M | $5–$20B |
~30:1
Unsatisfied institutional demand for AI equity versus available supply.
modelled§23
$500M–$5Bper Champion
Domestic retail and pension capital accessible at terms unavailable in any other vehicle, captive, price-insensitive, no-alternative-route capital.
modelled§23
$500B–$1.5Tover the decade
Public-market R3 capacity, on top of annual institutional flows, sovereign AI allocations, and chip-maker strategic equity.
modelled§24
The constraint is not capital availability but the pace of Champion formation.
The takeawayThe primary investable asset across all three rounds is ordinary regulated equity in Champions, allocated through the same channels as any listed infrastructure operator. No crypto-adjacent regime is triggered at the institutional layer.
Run the numbers
The case, in figures you can move.
Every number here is drawn from the thesis's own evidence register. Move the inputs and watch the economics respond. Modelled figures are the framework's projections, not results.
The operational revenue model
The 80/20 engine
Competent intelligence on the open stack for the bulk of the workload; frontier routed only when it earns its keep.
23.5blended cost index
all-frontier = 100modelled
all-frontier = 100modelled
5–20% · the rest is competent work
5×–20× cheaper per token than frontier APIs
The workload, splitmodelled
competent · open stack, infrastructure margins frontier · routed, distribution margin layered on frontier APIs
23.5blended cost index
vs everything-through-frontier-APIs = 100≈ 4.3× cheaper than all-frontiermodelled
This margin structure is what makes free or near-free universal baseline service economically viable, and scales per-citizen revenue from$200 toward $2,000+/yr as the platform layers build out.
What a Champion is worth
Per-Champion economics
Pick a tier, size the served population and the mature per-citizen revenue, and read the implied valuation band.
$36Billustrative annual revenue
at maturitymodelled
at maturitymodelled
5M–300M · defaults per tier
$200–$2000/yr · baseline service → platform layers
Implied valuation at maturitymodelled
Base case 10–18×
$360B–$648B
Upside 15–25×
$540B–$900B
Thesis range Major economy
shared linear scale$300B–$1T
Illustrative, drawn from comparable regulated infrastructure and platform businesses. Actual multiples reflect jurisdiction, regulatory standing, and operating maturity. All figures modelled.
Deployment is a workforce
The FDE engine, sized.
$7.50Bmature annual revenue · one Champion modelled
a major-economy Champion at $5–20B revenue needs500–5,000 FDEs
Network view modelled
across ~200 Champions at full density →300,000FDE positions worldwide
One FDE-led deployment: $5–50M annual revenue,6–18 months to mature, 5–10 yearsof switching costs. These are knowledge-intensive, locally-rooted jobs that cannot be done remotely from a foreign hyperscaler, the largest source of new white-collar work the AI transition creates.
What R1 participation is worth
One ticket, same terms.
Participant typeinstitution: university · public pension · sovereign fund · foundation
$70kpaper value at R2 modelled
$280killustrative value at R3 modelled
+ compute-for-benefit returnequal-value compute credits at R1,$10k of inference & training compute deployable to your own public-benefit work, alongside identical equity terms.
Indicative ranges from the thesis; actual markups reflect operating progress, market conditions, and jurisdiction.
R1 is $1 pre-money, same terms for every participant, citizens, universities, pensions, sovereigns. Not price discovery; political franchise establishment.
Part VI · The Network
The protocol layer
Individually, Champions are good businesses. As a network they are architecturally new: equity-financed PBCs operating in their jurisdictions, coordinated by a Bitcoin-derived permissionless protocol, no federation, no council, no membership, nothing to capture.
Independent operators converging on a Schelling point: the consensus rule that defines public-benefit work. No council, no membership, nothing to capture.
Permissionlessness has real geopolitical exposure: Bitcoin's era of majority foreign-state-aligned hashrate at higher stakes. Local licensing and fork-resilience mitigate; neither eliminates it.
Figure 13
~200 sovereign node operators converging on a Schelling point, the consensus rule that defines public-benefit work, with no center. Toggle the federation comparison: a federation can be lobbied, captured, or fragmented; a protocol has no institutional center to attack.
Quality assurance is in the protocol; alignment is harder
The consensus rule ties reward to verifiable public benefit: deployed work must meet defined quality thresholds, accuracy, audit compliance, service levels. Automated oversight rechecks compliance; fraudulent or sub-threshold work earns nothing, and bad actors get slashed and lose the franchise license.
That is quality assurance at the deployment layer, not full alignment. A subtly misaligned model that produces verifiable benefit on the audited metrics also gets rewarded, the consensus rule cannot distinguish, because it only sees what the benefit-class definitions tell it to see. Upstream alignment of frontier capability remains the frontier labs' hard, unsolved work; deployment alignment, the gap between QA and deeper alignment, is ongoing research each Champion owns under its licensing regime. The framework's contribution is real but bounded.
Permissionlessness has a real cost
Bitcoin's permissionlessness allowed state-aligned mining pools to hold majority hashrate for years, creating sustained geopolitical anxiety. This protocol carries much more economically significant infrastructure, and will face the concern at higher stakes: adversarial state-aligned Champions may dominate sub-areas of network activity; de facto deployment standards will reflect the priorities of the largest deployers.
The response is honest but partial: local licensing limits the cross-jurisdiction reach of adversarial Champions, and fork-resilience provides a safety valve if protocol direction is captured. Both are real mitigants; neither eliminates the exposure. Participants are buying into a permissionless system, not a Western-aligned cartel.
Stewardship on the Linux model
II, the small core team that designed the protocol, maintains the open-source reference implementation, the way Linus Torvalds and a few maintainers steward the kernel that millions depend on. No membership, no governance body. Champions adopt and adapt the reference under OSI-approved licenses, can pay II's subsidiaries for setup and ongoing support the way enterprises pay Red Hat, and can stop paying whenever the services stop adding value.
To make this a commitment rather than an aspiration: the transition to multi-maintainer stewardship is bound to objective triggers, a defined threshold of independent contributors and operating Champions, and II's aggregate network stake is subject to a stated cap. Both are written into the franchise terms.
The Champion answer
How II sustains itself
Three complementary streams. The 10% warrant in each Champion, struck at R1, exercisable at R3, no voting rights until exercise, rewards II for each Champion's individual success without governance authority during the formative period. Foundation Coin holdings, the protocol's native asset, mined through proof of benefit, reward II for network-level success. Services revenue rewards II for delivering useful work, on terms each Champion can walk away from.
Together they make II one of the most valuable companies in the framework without requiring it to control a single Champion.
~200node operators
At full deployment: one or several per major economy, one per smaller country, state-level Champions in federal nations, city-level in some jurisdictions.
stated§25
12at launch
The whitepaper launches with 12 Champions and scales toward at least one per sovereign nation.
stated§25
~$30B
What II's 10% warrant on a single $300B major-economy Champion would be worth at R3, the warrant portfolio is II's primary hard asset.
modelled§25
Champions are to AI infrastructure what miners are to Bitcoin: independent operators running a common protocol, each licensed in its own jurisdiction, each free to opt out.
The takeawayThe Champion layer is the primary investable asset and functions whether or not the protocol layer ever achieves financial sophistication. The protocol is operational substrate, necessary for the network to cohere, not the surface institutional capital allocates on.
Part VI · The Network
The five headline arbitrages
The Champion model captures multiple structural advantages that compound: licensing inevitability, protocol coordination, cost of capital, FDE switching costs, and sovereign demand aggregation, jointly very hard to dislodge once established.
Protocol coordination
~200 Champions coordinate through a Bitcoin-derived protocol: no center to capture, no governance to lobby. Coordination overhead near zero.
reinforces → all four, no institutional overhead
Cost of capital
Patient capital at jurisdiction-specific cost, 300–700 bps below hyperscaler software capital. Worth $300M–$3.5B per year per Champion in NPV terms.
reinforces → FDE deployment depth
FDE switching costs
Locally-rooted engineers embedded in customer workflows; each deployment creates 5–10 years of switching costs. Foreign providers cannot field this without destroying their unit economics.
reinforces → demand capture, compounding annually
Sovereign demand aggregation
Each Champion captures the full 100–1,000× per-citizen token-demand growth in its jurisdiction; every interaction accumulates as governed reference in Radiant.
reinforced by → licensing · protocol · FDE moat
The advantage is not any single arbitrage. It is that all five reinforce each other and are jointly very hard to dislodge.
Figure 14
Five arbitrages and their reinforcement edges. Licensing inevitability is the central moat; protocol coordination means everything happens without institutional overhead; each edge is a mechanism, not a metaphor.
Navigation as governance: top-down + bottom-up
A society entering the intelligence age needs both top-down policy guidance and bottom-up citizen-and-business signal. Historically these were separate and slow: policy on year-scale timelines; citizen signal through elections, surveys, and journalism, each capturing a fraction of the input with substantial lag.
Champions combine both, in real time, with AI mediation. Top-down: policy reaches the Champion through legislation and regulation, encoded into operating parameters with a full audit trail in Radiant, and II-Agent helps officials draft policy with impact projection before it is finalized. Bottom-up: every citizen and business has an II-Agent, and the Champion synthesizes signal at population scale without losing minority voices. A government asking "how is this policy actually landing?" gets an answer in days, with granularity, not crude polling.
The boundary that keeps it safe
One boundary makes this safe rather than dangerous: the navigation function mediates the deciding; it must never author the criterion. II-Agent advises the humans who hold the policy; citizen synthesis surfaces preference to the representatives who hold the decision; the definition of what counts as benefit, and of who counts, stays with people and stays plural.
On day one this is a governance norm with human approval at every gate. Making it a structural invariant, no write-path from the advisory layer to the criterion, is active work, not a solved problem.
The Champion answer
Why this form is necessary, not just an opportunity
No prior institutional form has combined these. Governments have top-down authority but lack continuous bottom-up signal at scale. Corporations have customer signal but lack public-interest mission. Champions have both, made coherent by the PBC mission and AI-mediated governance.
The challenges of the intelligence age cannot be navigated by governments alone, markets alone, or technology alone. Champions are the institutional form that lets a society navigate all of them coherently.
300–700bps
Champion cost-of-capital advantage over hyperscaler software capital, worth $300M–$3.5B per year per Champion in NPV terms on $10–50B of infrastructure.
modelled§26
5–10years
Switching costs created by each FDE-led deployment. They compound annually as deployments mature.
modelled§26
The navigation function mediates the deciding; it must never author the criterion. The definition of what counts as benefit, and of who counts, stays with people and stays plural.
The takeawayThe structural advantage is not any single arbitrage. It is that all five reinforce each other, licensing protects the capital advantage, capital funds FDE depth, FDE depth compounds switching costs, and the protocol coordinates it all without institutional overhead.
Part VII · Timing
Why now
The formative window is 6–12 months for Champion #1, with a first wave of 5–15 Champions inside 18 months. Once the template is operating, the network follows in parallel, and the structure that gets built during the formative window persists for decades.
The deployment pace
From Champion #1 to the licensed provider of record, ten years, five windows. Select a window to inspect it.
1 championFirst 6–12 months
phase 01 · First 6–12 months
Champion #1 launches
II provides setup services to local operating leadership; R1 opens at $1 pre-money to all local participants.
phase 02 · Months 6–24
First wave, in parallel
First wave of 5–15 Champions in parallel. Named initial jurisdictions: New York, California, Florida, Utah, Washington, and the UK.
phase 03 · Years 2–3
Effectively universal
Effectively universal across major economies and US states.
phase 04 · Years 3–5
Full network density
Full network density of ~200 Champions; state licensing regimes start to form.
phase 05 · Years 5–10
The license locks in
Licensing regimes harden globally; Champions become the licensed providers; long-term regulatory protection locks in.
Build-out that took traditional infrastructure operators 5–10 years compresses via II’s replicable setup template, AI tooling, and existing AI capex routed through Champion structures.
Figure 15
The deployment pace: Champion #1 in 6–12 months; a first wave across New York, California, Florida, Utah, Washington, and the UK inside 24 months; effectively universal across major economies in years 2–3; ~200 network density by year 5; licensing regimes harden through year 10.
The status quo
Six reasons the window closes
One: hyperscaler capex is unsustainable. $400B+ annual spend cannot continue on existing balance sheets, and the Champion is the most politically acceptable off-balance-sheet form. Two: sovereignty pressure is intensifying across the EU, India, the Gulf, ASEAN, and Latin America. Three: chip and compute supply chains are actively seeking demand diversification, the supply chain is aligned with Champions.
Four: the robotics deployment gap is now validated by the model companies themselves. On May 11, 2026, OpenAI launched a $4B deployment company with 150 forward-deployed engineers, and Anthropic announced a $1.5B deployment venture with Goldman Sachs and Blackstone the same week, the same FDE language, the same deployment-gap diagnosis this thesis is built on. But centralised, vendor-locked, software-only, consulting economics: the wrong architecture. Five: the displacement shock is starting now, and governments will need an institutional answer within 3–5 years. Six: II franchise licenses are available immediately; state licenses follow as token-economy disruption forces them.
The setup sequence per jurisdiction
The "where are the founders?" objection has a clean answer: II's setup services include recruiting local operating leadership, a CEO with infrastructure-operator credibility, a CTO fluent in the open-source stack, heads of regulatory affairs and capital markets, and a small core team that AI tooling makes far smaller and faster to assemble than historical comparables. The right CEOs and CTOs exist in every major jurisdiction.
The Champion incorporates as a PBC on the II template; II issues the franchise license; R1 opens at $1 pre-money to every local institution and retail participant; R2 opens at markup for patient strategic capital. The Champion then operates, building citizen-relationship density, FDE workforce, and visible public-benefit delivery, and is positioned to receive the state license when political pressure forces regimes to form.
The Champion answer
Why the build-out compresses
Work that takes traditional infrastructure operators 5–10 years compresses through three factors: II's replicable setup template (no greenfield design work per Champion), AI tooling that makes small operating teams highly productive, and the $400B+ of existing AI infrastructure capex that Champions route through their structures rather than building from scratch.
Most major economies have Champions within 3 years. Full ~200-Champion network density is a 3–5 year project. The multi-decade dominant position is the steady-state output of the whole sequence.
6–12months
Champion #1's launch window: II provides setup services; R1 opens at $1 pre-money to every local participant.
stated§28
$4B + $1.5Bin one week
OpenAI's deployment company and Anthropic's venture with Goldman Sachs and Blackstone, announced May 11, 2026, the deployment gap acknowledged as market fact, in the thesis's own FDE language.
reported§28
3–7years
The window over which political pressure forces state licensing regimes, and formative-window Champions are the natural licensees.
modelled§28
They told the market simultaneously that the model is not enough and deployment infrastructure is the bottleneck. They built the wrong architecture. Champions are the right one.
The takeawayThe displacement shock is the reason that converts the framework from "an interesting structural opportunity" to "the institutional answer to the central political problem of the next decade."
The Case, Reassembled
The structure persists for decades.
One correction runs through the whole argument: AI infrastructure is utility-class, not software. Accept that, and the institutional form follows, and with it the answer to the central political problem of the next decade.
The argument, in one pass
AI infrastructure is financed as software but is actually utility-class. That mismatch is the source of every dysfunction in the market, from unsustainable hyperscaler capex to leveraged neocloud vehicles to the stalled open-vs-closed debate, which was always an ownership question miscast as a product one. Token demand is climbing 1,000–10,000× into geographies that cannot hold it, and the displacement of labor has begun in a form UBI and retraining cannot answer.
The Champion is the institutional resolution: a public-benefit private company with a population-scale cap table, AI-mediated governance, and a licensed utility position, running an open stack whose reference layer compounds, financed by three rounds that reward citizens first, coordinated by a protocol with no center to capture. Government anchors the early demand; FDEs carry deployment and become the transition's largest source of new skilled work; robotics settles the durable value in the deployment layer the Champion already operates.
The window is open now: franchise licenses today, state licenses on a 3–7 year fuse, and the first wave of jurisdictions already named. What gets built during the formative window persists for decades. That is the case.
Citizens are not recipients of redistribution. They are shareholders in the entity doing the productive work.
The takeawayThe diagnosis names the mismatch. The Champion resolves it: locally owned, broadly held, protocol-coordinated, licensed into durability. Where does labor get capital when capital no longer needs labor? Labor owns the productive capital.