Tenants of Intelligence

Consider a single working day.

A freelance strategist wakes up and opens her laptop. She drafts a client proposal using an AI writing assistant whose terms of service changed last week in ways she has not yet read. She pays for the model through a subscription platform that can suspend access for policy violations it defines unilaterally. Her bank flags the payment as unusual because she processed three international invoices yesterday and her account risk score shifted. The flag clears after forty minutes. She has a call with a prospective enterprise client at noon. She passed their initial screening, but their procurement system has flagged her firm as too small for their vendor tier. They will discuss it. Her visibility to that client came through a professional platform whose ranking algorithm deprioritized her profile last month when she failed to post consistently. She has been working on the proposal for two weeks. It is good work. She will not be reviewed because of the flag, and seen because of the ranking, and considered because of the vendor threshold, and paid because of the processor, and none of this will feel like control.

She is not employed. She is not supervised. She has no landlord and no boss. She is, in every formal sense, independent.

She is also completely surrounded.

This is the new class condition of the AI economy: not slavery, not employment, not classical entrepreneurship, and not merely the gig precarity analysts have been describing for a decade. It is tenancy inside intelligence environments: a condition in which actors appear independent while leasing the conditions of cognition, visibility, distribution, trust, payment, permission, and execution from systems they do not own and cannot meaningfully negotiate with.

The tenant of intelligence does not rent one tool.

She rents the conditions under which her skill becomes economically real.

That is the core distinction. A person may own her laptop, her credentials, her company, her inventory, her audience, her ideas, and her time. She may still rent the environment in which any of those things can become income, leverage, recognition, or action.

Platform capitalism rented distribution.

Rented intelligence environments rent the conditions of economic agency itself.

That is the difference.

And that difference is becoming a class structure.

The Tenant Is Not Owned. She Is Surrounded.

The old image of dependency was personal.

A worker depended on an employer. A peasant depended on a landlord. A debtor depended on a creditor. A client depended on a patron. Power had a face. It could be hated, negotiated with, appealed to, resisted, or replaced. The relationship was legible enough to become political.

The new dependency is harder to name because it is environmental.

A single actor today may depend simultaneously on a marketplace for customer access, a payment processor for income conversion, an AI model provider for cognitive capacity, a compliance system for market eligibility, a cloud platform for operational existence, a professional network for visibility, and a procurement system for institutional access.

No single system owns her.

Each relationship appears voluntary. Each can be exited in theory. But she cannot exit the environment as such, because the environment is not one system. It is the aggregate of all of them.

This is tenancy without a single landlord.

The political significance of that phrase is not rhetorical. Social movements historically coalesced around visible antagonists: the employer, the landlord, the colonial power, the monopolist. Rented intelligence environments produce no single antagonist. The actor is surrounded by payment systems, ranking algorithms, compliance requirements, procurement filters, model providers, and cloud platforms whose owners are numerous, whose effects are cumulative, and none of whom bears individual responsibility for the aggregate condition they produce together.

Each system can claim it is only enforcing its own rules.

The payment processor manages risk. The platform optimizes relevance. The model provider enforces safety. The procurement system protects the institution. The bank prevents fraud. The compliance vendor satisfies regulation. The cloud provider guarantees uptime.

Individually, each gate may be reasonable.

Together, they define the world through which action must pass.

The old labor question was: who employs whom?

The new intelligence question is: who owns the environment in which action becomes possible?

These are not the same question. The political institutions built to answer the first are poorly equipped to answer the second.

What Counts as an Intelligence Environment?

The phrase intelligence environment does not mean model output alone.

It means the full system through which cognition becomes economically actionable: the model, the cloud, the payment rail, the ranking system, the compliance layer, the procurement filter, the identity system, the data infrastructure, and the interface through which decisions are made.

AI matters because it fuses these layers into a decision environment.

A model does not merely help someone write, code, design, search, classify, or decide. It increasingly sits inside workflows that determine who is visible, who is trusted, who is eligible, who is payable, who is institutionally acceptable, and who can execute at scale.

That is why the tenant condition is deeper than platform dependency.

The platform era rented access to audiences.

The AI era rents access to cognition, execution, and legitimacy.

Distribution was the first layer. Intelligence is the generalization.

Visibility Is Rented

The first form of intelligence tenancy predates AI.

The distribution tenant owns the thing being offered but rents the path to the audience. She may own the product, the skill, the inventory, the content, or the service. But demand is mediated by ranking, recommendation, advertising, search placement, marketplace rules, account health, reviews, and algorithmic circulation.

Her economic life becomes a study in conditional visibility.

She is not forbidden from selling.

She is merely not shown.

This is a different kind of exclusion than the old market produced. It does not announce itself as rejection. It appears as lower reach, higher ad costs, reduced ranking, weaker conversion, demonetization, category suppression, account review, unclear policy violations, or a dashboard that quietly stops turning green.

The European Union’s Digital Markets Act designates large platforms that provide core platform services, including search engines, app stores, social networks, advertising services, browsers, operating systems, and marketplaces, as gatekeepers: entities through which businesses must often pass to reach users. [1]

The DMA’s framing captures the structure. The distribution tenant is not merely outcompeted. She is not surfaced. The gate is upstream of the competition.

Platform capitalism created this condition. AI intensifies it by moving eligibility further upstream still, from ranking inside the platform to pre-market classification before ranking begins.

A seller who understood how to compete for search position now must also satisfy agent eligibility criteria that may never be visible to her. A product may be ignored because its warranty is not machine-readable. A restaurant may disappear because its menu data is inconsistent. A contractor may never reach the buyer because the agent cannot verify the required insurance credential. A small vendor may lose not because the buyer disliked the offer, but because the buyer’s delegate never treated the vendor as eligible for comparison.

The field does not merely change shape.

It begins to disappear from view.

In the old market, the seller competed for attention.

In the platform market, the seller competed for ranking.

In the agentic market, the seller competes to enter the agent’s eligibility set before the buyer forms a preference at all.

That is not simply a different marketing problem.

It is a different market.

Cognition Is Rented

AI creates a deeper tenancy because it does not only mediate distribution.

It mediates thinking.

The cognition tenant rents judgment support, drafting, summarization, coding, analysis, design, translation, customer service, planning, search, memory, workflow automation, and decision assistance from systems she does not own.

She becomes, in a practical sense, more capable.

She also becomes more dependent.

That paradox is the defining feature of cognition tenancy. The actor does not feel constrained. She feels augmented. The tools are good. The productivity gains are real. The constraint operates at the level of the environment, not the task.

A professional using AI to draft faster, research faster, summarize faster, and respond faster may experience genuine liberation from tedious work. A small business using AI support may do things that once required a staff. A startup using model APIs may build products it could not have built five years earlier.

The immediate experience is empowerment.

The structural condition is dependency.

A firm that reorganizes around rented intelligence changes its internal muscles. Employees stop performing some tasks directly and start supervising systems. Workflows adapt. Hiring changes. Training changes. Memory moves into tools. Judgment becomes entangled with outputs produced elsewhere.

The organization becomes more efficient and less self-sufficient at the same time.

The source of capability sits upstream.

A model provider can change pricing, deprecate a model, alter safety policies, throttle usage, bundle a competing feature, modify the context window, adjust the API, restrict a category, or change enterprise terms. Each change passes through the tenant’s operation as an environmental condition rather than a negotiated term.

The dependency is not only financial.

It is cognitive.

Stanford’s AI Index 2025 reports that corporate AI investment reached $252.3 billion in 2024, with private investment rising 44.5 percent. [2] That capital is not building a neutral intelligence utility evenly owned by society. It is constructing an infrastructure layer that most actors will access as tenants.

The scale of the buildout is, structurally, the scale of future tenancy.

This is where open-source AI creates a real but limited objection.

Open models reduce one dimension of dependency. They allow more actors to access model capability without relying entirely on proprietary frontier systems. They matter. They may prevent the intelligence layer from becoming completely closed.

But open weights are not open estate.

An open-weight model running on a laptop is not the same as a secure, audited, scalable, insured, compliant, supported, enterprise-grade system that a hospital, bank, government agency, defense contractor, school system, or multinational corporation can adopt.

The model may be downloadable.

The environment is not.

Running intelligence at scale still requires compute, hosting, uptime guarantees, security, latency management, observability, compliance, maintenance, integration support, and institutional accountability. Each layer pushes the actor back toward cloud providers, managed services, model platforms, security vendors, and procurement-approved infrastructure.

Open source dissolves one gate.

The estate still stands.

Permission Is Rented

Visibility and cognition are not enough.

The actor must also be permitted to act.

This is the third and most politically invisible layer of tenancy: permission tenancy. It includes compliance, payment, procurement, identity, risk classification, and institutional eligibility.

A business, creator, contractor, startup, school, hospital, nonprofit, or professional does not only need to be good at the work. It must be acceptable to systems that classify risk before the work is ever evaluated on merit.

Permission tenancy is powerful because it presents itself as neutrality.

No one says: we govern your economic life.

They say: we enforce safety, trust, quality, fraud prevention, brand suitability, procurement standards, financial compliance, data protection, cybersecurity, and user protection.

Much of this is necessary. Large digital markets cannot function without trust systems. Payment networks cannot ignore fraud. Healthcare systems cannot ignore privacy. Financial systems cannot ignore money laundering. Marketplaces cannot ignore counterfeit goods. Platforms cannot ignore abuse. Public agencies cannot buy software without security standards.

But necessity does not make the system apolitical.

Who defines risk? Who defines trust? Who defines acceptable use? Who sets the threshold between legitimate business and prohibited activity? Who decides which vendor is too small, which account is suspicious, which model is compliant, which seller is safe, which worker is qualified, and which firm is eligible to enter the institutional choice set?

These decisions increasingly sit inside private or quasi-private systems with public consequences.

The compliance tenant experiences governance as administration.

An account review. A rejected ad. A frozen payment. A verification loop. A policy notice. A missing approval. A procurement checklist. A risk score. A support ticket that takes eleven days to close.

The language is procedural.

The effect is existential.

AI governance intensifies this condition because trustworthiness itself becomes formalized into documentation, audits, certifications, model cards, security questionnaires, risk management frameworks, procurement requirements, and compliance evidence.

This is rational at the frontier. High-risk AI systems should be documented. Public agencies should know what they are buying. Hospitals, banks, schools, and governments should not deploy opaque systems without accountability.

But once trustworthiness becomes a condition of institutional adoption, the ability to satisfy governance frameworks becomes part of market access.

The architecture cascades.

Agencies impose requirements on vendors. Vendors impose requirements on subcontractors. Subcontractors impose requirements on suppliers. Enterprises impose requirements on AI tools. Platforms impose requirements on sellers. Payment systems impose requirements on merchants.

The compliance architecture of the AI economy is not one wall.

It is a nested set of conditions, each imposed by a different actor, each presented as reasonable, each bearing most heavily on the smallest and least resourced actors.

The procurement tenant competes before the competition begins.

A small company may have a better product. A consultant may have better judgment. A startup may have better software. But institutional buyers do not only buy products. They buy from approved vendors, risk-cleared entities, compliant suppliers, security-audited platforms, insured providers, credentialed firms, and administratively legible organizations.

The firm that cannot pass procurement does not lose on price.

It fails to enter the choice set.

The institutional buyer never encounters the bid.

The same structure appears in payments.

The payment tenant discovers that selling is not the same as being allowed to receive money.

A business can have customers, demand, inventory, a website, a brand, and a working product. But if payment processors classify it as high-risk, if fraud systems flag its activity, if reserves are imposed, if payouts are delayed, if chargeback ratios rise, if an account is frozen, or if a platform removes monetization, the business can become economically stranded with its product intact and its customers willing.

The payment layer is not plumbing.

It is permission to convert activity into income.

This also changes what it means for AI to mediate commerce. An agent that only recommends still sits near the surface of the market. An agent that can authorize payment enters the transaction itself. Once the delegate can pay, the infrastructure beneath that payment begins to determine which sellers can be selected before the buyer sees anything at all.

Visa’s Intelligent Commerce program and Mastercard’s Agent Pay infrastructure both describe payment rails built for agentic transactions: systems in which the delegate, not necessarily the human, authorizes and executes the purchase. [3] [4]

When the agent pays, the payment system’s eligibility criteria become market entry criteria.

A seller not cleared by the agent’s payment infrastructure is not merely ranked lower. She is absent from the transaction before the buyer forms a preference.

The old market asked whether someone wanted to buy.

The permission layer asks whether the system will allow the sale to become income.

This is why payment, compliance, and procurement belong in the same political-economic category. They are not separate inconveniences. They are the permission architecture of the intelligence economy.

They decide whether visibility can become contact, whether contact can become trust, whether trust can become a transaction, and whether a transaction can become income.

Sovereignty Is Rented

The tenant condition does not stop at individuals or firms.

It scales to states.

A government can retain full legal sovereignty, including flags, courts, elections, borders, ministries, police, budgets, and diplomatic recognition, while becoming computationally dependent on foreign infrastructure in ways that make that sovereignty increasingly formal.

This is the sovereign tenant: the state that governs through systems it does not own.

A government may announce national AI strategies, digital transformation plans, sovereign cloud initiatives, domestic compute programs, and AI education campaigns. But its real operational capacity may depend on foreign chips, foreign cloud providers, foreign model architectures, foreign cybersecurity tools, foreign data-center operators, and foreign technical expertise.

The state remains legally sovereign.

Its execution layer is rented.

A study cited by Deloitte’s Technology Predictions 2026 finds that only 34 countries host any public AI compute, only 24 of those have access to training-level compute, and most rely on cloud or chip infrastructure controlled by a small number of foreign actors. The same analysis reports that 90 percent of all AI compute is managed by U.S. and Chinese companies. [5]

Cloud concentration reinforces the condition. Amazon Web Services, Microsoft Azure, and Google Cloud together mediate the majority of the world’s cloud infrastructure. [6]

Data residency does not solve this by itself. The U.S. CLOUD Act clarified that providers subject to U.S. jurisdiction can be compelled under valid U.S. legal process to disclose responsive data within their possession, custody, or control regardless of where that data is stored. A European agency using a U.S.-headquartered cloud provider may therefore keep data in Europe while still renting part of its legal exposure from outside Europe. [7]

Nations are aware of this. They increasingly speak the language of sovereign AI, sovereign cloud, domestic compute, data residency, national model capacity, and strategic autonomy.

But political will can announce sovereignty.

It cannot instantly reproduce the estate.

The result is sovereignty as a service.

Large technology firms have responded to national sovereignty demands not by withdrawing, but by productizing them: localized cloud regions, data residency commitments, public-sector licensing tiers, sovereign AI packages, compliance guarantees, and national infrastructure partnerships.

The nation purchases the experience of sovereignty while remaining dependent on the infrastructure of the firms providing it.

This does not make sovereignty fake.

It makes sovereignty layered.

A state can govern people and still rent the systems through which governance is executed. It can regulate platforms while depending on them. It can proclaim AI independence while importing chips, models, and cloud capacity. It can write laws for systems it cannot fully replace.

The question is not only whether a state has legal authority.

The question is whether it owns the infrastructure required to act on that authority.

The sovereign tenant is not necessarily a weak state or a failed state. It may be a prosperous democracy with strong institutions, a capable bureaucracy, and a legitimate government. It is simply a state that has not yet reckoned with the fact that governance increasingly runs on infrastructure it does not own and cannot unilaterally modify.

Legal sovereignty remains.

Computational sovereignty is rented.

This is the tenant condition at civilizational scale.

The New Class Map

The class structure of the AI economy does not map cleanly onto old categories.

Capital versus labor still matters. Owners versus workers still matters. Employers versus employees still matters. Wages, exploitation, bargaining power, and ownership of productive assets still matter.

But the map is incomplete.

A different axis is forming: ownership versus tenancy inside intelligence environments.

At the top sit infrastructure owners: the firms and states that own the compute estate, chip supply chains, cloud platforms, model layers, payment rails, identity systems, and distribution systems. Their power is not primarily the power of employment. It is the power of the environment. They do not need to hire the actors who depend on them. They need only to maintain the systems through which those actors must pass.

Below infrastructure owners sit interface owners: the firms that own platforms, marketplaces, app stores, professional networks, agentic interfaces, productivity suites, and content systems that sit between infrastructure and end users. Interface owners may themselves be tenants of the infrastructure layer while acting as landlords to those beneath them.

Below interface owners sit intelligence tenants: the individuals, firms, small businesses, creators, professionals, contractors, institutions, and governments that operate inside rented intelligence environments. They appear independent. Their formal status may be entrepreneurial, professional, organizational, or sovereign. Their practical condition is tenancy.

Below tenants sit the excluded: actors who cannot satisfy the eligibility conditions of the intelligence environment at all, too small to pass procurement, too informal to satisfy compliance, too marginal to reach payment rails, too unfamiliar to become machine-legible, too politically inconvenient to be trusted, too resource-poor to format themselves for the systems that govern access.

They are not outcompeted.

They are classified outside the arena before competition begins.

The distance between these positions is not primarily a function of skill or effort. Skill and effort still matter, but only after access has been granted to the environmental conditions that make skill and effort economically real.

A brilliant worker who cannot pass the hiring screen is not evaluated.

A capable firm that cannot pass procurement is not considered.

A good product that cannot be parsed by agents is not compared.

A real audience that cannot be monetized through payment rails is not income.

A sovereign state that cannot secure compute is not fully operational.

The class question is no longer only: who owns the means of production?

It is also: who owns the environment in which production, cognition, trust, payment, and access become possible?

That is the class structure.

Exit Becomes Theoretical

Every tenancy preserves the fantasy of exit.

If a platform changes its rules, use another platform. If a payment processor freezes funds, switch processors. If an AI model becomes expensive, use another model. If a marketplace suppresses reach, build your own website. If an app store rejects you, distribute elsewhere. If procurement blocks you, find other customers. If a cloud provider becomes politically risky, migrate.

Sometimes this works.

Often it does not.

Exit becomes theoretical when the environment itself is layered. Leaving one dependency means entering another. The seller who leaves a marketplace must buy ads. The creator who leaves a platform must rebuild distribution. The startup that leaves one cloud must migrate to another. The worker who avoids one hiring platform must pass through a different screening system. The business that leaves one payment processor must satisfy another risk engine.

The tenant can change rooms.

The building remains owned by someone else.

The trap is not only individual lock-in. It is collective convergence. Different platforms may compete, but they increasingly require the same forms of legibility: verified identity, structured data, payment compliance, cloud integration, security documentation, policy conformity, machine-readable reputation, and algorithmic visibility.

Exit from one system often means re-entering the same architecture through another door.

This is different from monopoly in the simple sense.

The problem is not always that there is only one provider. The problem is that the whole field of providers shares the same structural position: they own the layers through which access is mediated.

Competition between landlords does not abolish tenancy.

It can improve terms. It can lower prices. It can create options. It can reduce abuse. It matters.

But it does not eliminate the class condition.

The question is not only whether the tenant can choose among systems.

The question is whether she must rent the conditions of action at all.

Independence as Experience, Dependency as Structure

The tenant condition is politically difficult to name for the same reason it is economically comfortable to inhabit.

It does not feel like dependency. It feels like a tool, a platform, a service, a workflow, a product feature. The interface improves. The capability expands. The cost per unit falls. The work gets done.

The freelance strategist from the opening of this essay is, in many respects, more capable than she would have been a decade ago. She drafts faster, reaches more clients, processes payments internationally, manages operations that would have required a small team, and uses systems that expand her practical reach.

The dependency is real, but it does not announce itself.

It becomes visible only when the environment changes.

When the AI API bill arrives larger than projected. When the payment processor’s reserve policy changes. When the platform downgrades the account tier. When the procurement questionnaire asks for documentation the firm has never needed before. When the ranking algorithm updates and revenue falls. When the model provider changes policy. When the compliance vendor flags an account. When the cloud bill becomes the margin structure.

Each of those events arrives as a product change, a policy update, a terms-of-service amendment, a risk classification, or an administrative inconvenience.

None announces: you are a tenant and the landlord has changed the lease.

The old dependency was legible because it was personal.

The new dependency is illegible because it is environmental.

The actor can name the payment processor. She cannot name the condition of being payment-dependent. She can identify the platform. She cannot identify the condition of being platform-surrounded. She can complain about the model provider. She cannot easily contest the fact that cognition itself has become rented infrastructure.

This illegibility is politically consequential.

Workers organized against employers because employers were identifiable. Tenants organized against landlords because landlords were nameable. The tenant of intelligence has no equivalent focal point. The condition that shapes her economic life is distributed across systems whose individual operators may be acting reasonably, whose aggregate effect is structural, and whose architecture was never designed to be contested.

Independence becomes the experience.

Dependency becomes the structure.

And the structure is what matters, for income, leverage, voice, sovereignty, and the ability to claim a stake in what the AI economy produces.

Naming that structure is the first step toward contesting it. The tenant condition cannot be politically addressed as long as it is experienced only as a succession of product features and administrative inconveniences. It requires a political vocabulary that does not yet exist at scale.

That vocabulary begins here: not with employment, not with ownership, not with the old labor question of who works for whom.

With the new intelligence question of who owns the world through which everyone else must move.

Sources

[1] European Commission, Digital Markets Act Gatekeepers Portal. The DMA designates large platforms as gatekeepers when they provide core platform services that function as important gateways between businesses and users. https://digital-markets-act.ec.europa.eu/gatekeepers-portal_en

[2] Stanford HAI, AI Index Report 2025. Corporate AI investment reached $252.3 billion in 2024, with private investment rising 44.5 percent year over year. https://hai.stanford.edu/ai-index/2025-ai-index-report/economy

[3] Visa, Intelligent Commerce. Visa describes infrastructure for AI agents, platforms, and developers to integrate into global commerce with secure and reliable payment experiences. https://corporate.visa.com/en/products/intelligent-commerce.html

[4] Mastercard, Agent Pay. Mastercard describes agentic payments using registered agents and governed, traceable network tokens. https://www.mastercard.com/us/en/business/artificial-intelligence/mastercard-agent-pay.html

[5] Deloitte Technology Predictions 2026, citing Oxford Internet Institute research on global AI compute concentration. Only 34 countries host any public AI compute; only 24 have access to training-level compute; and 90 percent of all AI compute is managed by U.S. and Chinese companies. https://www.deloitte.com/us/en/insights/industry/technology/technology-media-and-telecom-predictions/2026/tech-sovereignty.html

[6] Cambridge Core, “Sovereign AI in 2025.” The article reports global cloud market share estimates for Amazon Web Services, Microsoft Azure, and Google Cloud. https://www.cambridge.org/core/journals/natural-language-processing/article/sovereign-ai-in-2025/C51560626AF518BDF280891C406E9553

[7] U.S. Department of Justice, “The Purpose and Impact of the CLOUD Act.” The DOJ explains that the CLOUD Act clarified that providers subject to U.S. jurisdiction must disclose responsive data under valid U.S. legal process regardless of where the data is stored. https://www.justice.gov/criminal/media/999616/dl