The New Enclosure:
Cloud Capital, Colonial Logic, and the Battle Over Intelligence
Structure Before Sophistication | Gerry Visco
I.
There is a pattern in history that we keep failing to recognize until we are already living inside it. Power concentrates. It builds walls around what was once common. It invents the language of necessity — civilization, security, progress, order — to explain why the enclosure is, in fact, for your benefit. And then it extracts.
We are living through one of those moments right now.
The current battle over AI regulation and the governance of AGI is being framed, predominantly by the people who stand to benefit most from its outcome, as a question of safety. Who can be trusted with this technology? Who has the institutional maturity, the ethical architecture, the fiduciary seriousness to steward intelligence at scale? The subtext, rarely made explicit, is always the same: not you. Not governments. Not the public commons. Us.
II.
Yanis Varoufakis names the terrain on which this battle is being fought. In Technofeudalism, he identifies a new class of capital that operates outside the rules of traditional market competition — what he calls cloud capital. Unlike the industrial capital of the twentieth century, cloud capital is not productive in the conventional sense. It does not manufacture. It appropriates. The great cloud platforms — their algorithms, their recommendation engines, their logistics systems — function as feudal fiefdoms in digital form. Access is granted, not owned. Rent is extracted in the form of attention, data, and behavioral compliance. The vassal relationship is no longer between serf and lord of a physical estate; it is between user and platform, worker and algorithm, merchant and marketplace. The terms are always set by the one who built the walls.
What Shoshana Zuboff adds to this architecture is the extraction mechanism itself. Surveillance capitalism is not simply a business model; it is an epistemological project. The capture of behavioral data — every click, every pause, every search, every hesitation — is the raw material for what Zuboff calls behavioral modification at scale. The product being sold is not your data in any straightforward sense. It is your future — the probability of your next choice, packaged and sold to whoever wants to steer it. This is not advertising in the traditional sense. It is prediction infrastructure. The goal is not to sell you something you already want. It is to know what you will want before you do, and then to shape that wanting.
Hannah Arendt, writing decades before any of this was technically conceivable, understood the structural logic that makes it possible. In The Origins of Totalitarianism, she observed that imperialism — the nineteenth century’s defining political formation — was not driven by national ambition alone but by the internal logic of capital that had run out of domestic room. When the rate of return at home flattens, capital does not sit still. It moves outward, finds new territory, and replicates its extraction apparatus in the new space. What she could not have anticipated was that the new territory would not be geographic. It would be cognitive. The unexploited frontier of the twenty-first century is the inner life — the habits, desires, fears, and decisions of three billion connected minds. Cloud capital has gone there.
Arendt also developed, in The Human Condition, a distinction that cuts directly to what is being lost. She described the common world — the shared space of public life, of institutions, of things that outlast any individual — as the essential precondition of genuine political existence. The enclosure of that common world by private interest was, for her, not merely an economic problem but an anthropological one. When everything that matters migrates into privately owned and privately governed space, the common world atrophies. People remain present to each other, but only as users, as behavioral profiles, as nodes in a network whose architecture they did not design and cannot alter. What passes for public conversation increasingly happens inside infrastructure owned by a handful of individuals who answer to no electorate.
III.
What makes this moment particularly instructive is the apparent contradiction at its center. The billionaire architects of cloud capital — the same figures who have locked down the foundational models, the compute infrastructure, the proprietary data pipelines — are also the loudest voices warning about the dangers of AGI in the wrong hands. The concern is presented as altruistic. We are told that general intelligence, unaligned and uncontrolled, represents an existential risk. And so, for your protection, the development of that intelligence must be concentrated in a small number of organizations with sufficient resources, sophistication, and moral seriousness to manage it responsibly.
Pause here. Because if you follow that logic to its conclusion, it leads somewhere that should feel very familiar.
It is the logic of the charter company.
IV.
In the year 1600, Queen Elizabeth I granted a royal charter to a group of London merchants, conferring upon them a monopoly on all trade east of the Cape of Good Hope. The East India Company was, in its inception, a commercial enterprise. Within a century and a half, it had become something else entirely: a sovereign power. By the mid-eighteenth century, the Company maintained an army larger than the British Crown’s own forces. It administered courts. It collected taxes. It fought wars. The Battle of Plassey in 1757, where Robert Clive defeated the Nawab of Bengal with a force that was predominantly Company-paid, marks the moment the fiction of a trading enterprise fully dissolved. The Company was now a government — privately held, shareholder-accountable, operating under the continuing premise that its presence was a form of benefit to the territories it controlled.
The Diwani grant of 1765 formalized what Plassey had established by force. The Mughal Emperor granted the Company the right to collect revenue across Bengal, Bihar, and Orissa — effectively handing the taxation of tens of millions of people to a private corporation. The Company did not govern so that it could trade. It traded so that it could govern. Extraction was the point; commerce was the vehicle; the language of order, stability, and civilizational advancement was the justification.
The structural parallel to the current AI moment is not rhetorical. It is architectural. The foundational models being built today — the ones that will increasingly mediate how human beings access information, make decisions, interact with institutions, and understand the world — are being developed by private entities operating under self-granted charters of responsible stewardship. There is no democratic mandate. There is no sovereign oversight with genuine teeth. There is a series of voluntary commitments, regulatory capture dressed as cooperation, and the repeated assertion that only those with the capital to build at this scale have the competence to govern at this scale. The East India Company said the same thing about Bengal.
V.
Cecil Rhodes understood, perhaps better than any figure of his era, that the charter company model was not a historical accident. It was a replicable technology — a mechanism for converting private ambition into public legitimacy through the instrument of the civilizing mission.
In 1889, Rhodes secured a royal charter for the British South Africa Company, granting it authority over a vast territory in southern Africa that would become Rhodesia. The premise was explicit and unapologetic. Rhodes believed, and said publicly, that the Anglo-Saxon race was the instrument of human progress, and that the expansion of its dominion was synonymous with the advancement of civilization. The Ndebele and Shona peoples who occupied the land had their own governance structures, their own legal traditions, their own relationships with the territory that had sustained them for generations. None of that registered in the colonial accounting. What registered was the assessment that they were not doing enough with what they had — that the land was underleveraged, the resources insufficiently extracted, the human potential inadequately directed.
The Pioneer Column that marched into Mashonaland in 1890 under the BSAC’s flag did not present itself as an invasion. It presented itself as the arrival of development. The telegraph lines, the railways, the administrative infrastructure that followed were genuine — they were also instruments of extraction, built to move value out of the continent and into London balance sheets, structured to make the territory dependent on the company’s continued presence rather than sovereign in its own right. The infrastructure of civilization was, simultaneously, the infrastructure of dependency.
It is worth being precise about how that dependency was engineered, because the mechanism matters more than the metaphor. The Cape railways did not merely connect — they sorted. Freight rates were structured to favor the export of raw materials over local commerce. Gauge specifications were set by the Company, ensuring that maintenance, replacement, and expansion required the Company’s continued involvement. The routing decisions — which settlements became nodes, which became dead ends — concentrated economic gravity at Company-controlled ports. Once this infrastructure was in place, the territory could not function without the entity that had built it. The dependency was not a side effect of development. It was the design.
Rhodes had a vision of a Cape-to-Cairo railway — a spine of connectivity running the length of Africa that would integrate the continent into the imperial economy. The equivalent today is not difficult to identify, and it is already well underway. Google’s Equiano cable and Meta’s 2Africa cable are among the largest undersea infrastructure projects in history, stretching tens of thousands of kilometers along African coastlines, owned outright by the same platforms that will monetize the behavioral data flowing through them. Satellite constellations and server farm archipelagos complete the stack. The hyperscalers are in a race to own the physical substrate of connectivity across the Global South — not as a side project, but as a strategic priority. The railway race has been reimagined in fiber and orbit.
The extraction mechanism is structurally identical. Starlink terminals cost approximately six hundred dollars in markets where that figure represents months of household income, priced in US dollars regardless of local currency. All data transits through US-jurisdictional infrastructure regardless of where the user sits — the CLOUD Act of 2018 obliges American technology companies to produce user data stored anywhere in the world upon government demand, extending US legal sovereignty across every network its companies touch. The routing nodes, like the railway chokepoints before them, are controlled by the infrastructure owner. The territory cannot function without the entity that built it.
In 2022, this dependency became briefly visible in a way that stripped away any remaining ambiguity. Starlink had become critical communications infrastructure for Ukraine’s military. When Elon Musk restricted coverage over Crimea, citing his own assessment of escalation risk, a private individual made a sovereign military decision affecting a nation at war. No charter was required. No democratic mandate was sought. The infrastructure spoke for itself. Nick Couldry and Ulises Mejias, writing in The Costs of Connection, describe this configuration not as an analogy to colonialism but as its structural continuation — the same extractive relationship between data-rich corporations and data-generating populations, the same asymmetry of terms, the same fiction of mutual benefit. The infrastructure of connectivity is, simultaneously, the infrastructure of extraction. The logic is unchanged. Only the medium is new.
VI.
The grammar of the civilizing mission runs through North American colonization with the same consistency. The Indigenous peoples of this continent were not, in the colonial framing, displaced. They were improved. What they had — complex governance systems, sophisticated ecological stewardship, extensive trade networks, rich legal and philosophical traditions — was systematically re-categorized as absence: absence of private property, absence of written law, absence of Christian civilization, absence of the specific forms of organization that the colonizer happened to recognize. The fact that these sophisticated frameworks already existed was not merely overlooked; their existence had to be actively suppressed, because their acknowledgment would have undermined the foundational premise. The colonizer’s value proposition depends entirely on your prior life being characterized as insufficiency.
The Indian residential school system in Canada was perhaps the most explicit articulation of this logic: the complete administrative project of replacing one cognitive architecture with another, by force, across generations, in the name of benefit. The language was always care. The mechanism was always extraction — of land, of labor, of cultural continuity, of the capacity for self-determination. The damage was not incidental to the project. It was structural to it. Systems designed to produce dependency reliably produce dependency.
Arendt recognized this dynamic as something more than historical injustice. In The Origins of Totalitarianism, she identified what she called the boomerang effect: the methods developed in the colonies — surveillance, administrative violence, the reduction of human beings to managed populations — do not stay in the colonies. They return. The techniques of behavioral management pioneered on colonized populations, the logic of the administered human being whose choices are shaped rather than respected, eventually migrates back to the metropole. Surveillance capitalism is, in this reading, the boomerang arriving. The behavioral modification apparatus that Zuboff describes was not invented for application to Western consumers. It was refined there. Its ancestors are the census regimes, the pass systems, the productivity surveillance of colonial administration.
VII.
Arendt offers one more lens that bears directly on the AI governance crisis, and it is perhaps her most unsettling contribution to this analysis.
In Eichmann in Jerusalem, she introduced the concept of the banality of evil — the observation that the most catastrophic acts of institutional harm are rarely driven by malice. They are driven by thoughtlessness. By the abdication of moral reflection in favor of process optimization, metrics, and institutional loyalty. Adolf Eichmann was not a monster in the theatrical sense. He was an administrator. He followed procedures. He met targets. He did not ask whether what he was optimizing was good; he asked whether he was optimizing it efficiently.
The engineers building behavioral modification systems at scale are not monsters either. They are optimizing engagement metrics, recommendation algorithms, content ranking systems. The harm produced — the radicalization pipelines, the eating disorder communities algorithmically reinforced, the political polarization systematically amplified because outrage retains attention — is not the intent of any individual actor. It is the aggregate output of a system designed to maximize a signal without asking whether the thing being maximized is worth maximizing. This is Arendt’s banality, translated from the bureaucratic to the algorithmic. The thoughtlessness is now engineered into the infrastructure itself. It does not require a human to decide not to think. The system does not think on anyone’s behalf, and the absence of thought is mistaken for neutrality.
Arendt also distinguished, in her political philosophy, between power and violence. Power, for her, was genuine and legitimate: it emerges from collective action, from people acting together in concert, from consent and participation. Violence is instrumental, substituting force for consent when power has failed or been refused. The current configuration of cloud capital does not exercise power in Arendt’s sense. It exercises structural violence — the violence of dependency, of terms of service that cannot be negotiated, of infrastructure you cannot opt out of without opting out of the modern economy. The fact that this violence is invisible, mediated through interfaces and algorithms rather than through physical coercion, does not make it less real. It makes it more durable, because it is harder to name and therefore harder to resist.
A precise objection deserves a precise answer. The East India Company governed Bengal with an army. The British South Africa Company administered Rhodesia through courts backed by the credible threat of physical force. The relationship between a platform and its users involves none of that. Nobody is compelled to use Google. Nobody is imprisoned for leaving Facebook. The coercion is categorically different in kind, and conflating the two risks making the analogy do more work than it can honestly bear.
The objection is correct about degree. It is wrong about structure.
The colonial method did not remain static. Early Company rule in Bengal was maintained by muskets; later colonial administration was maintained by railway debt, currency controls, and the systematic dismantling of local textile industries that left populations with no viable economic alternative to participation in the imperial system. By the late nineteenth century, the most effective instruments of colonial control were not military — they were infrastructural. The dependency did the work that the garrison used to do. When a territory’s food supply, its communications, its credit system, and its export routes all run through a single externally-owned architecture, the question of whether you are technically free to leave becomes largely academic. Freedom without a viable alternative is a formal category, not a lived condition.
This is precisely the configuration that platform capitalism has engineered. Opting out of Google’s infrastructure does not mean declining a service. For most small businesses, academic institutions, healthcare providers, and municipal governments, it means opting out of the operational systems on which their basic functioning now depends. The stack has been built, layer by layer, to make exit prohibitively costly — not through legal compulsion but through network effects, switching costs, and the careful bundling of essential services with proprietary lock-in. The guns have been replaced by the terms of service, and the terms of service are enforced not by soldiers but by the architecture itself. That is not the same thing as a musket. It is, structurally, the same thing as a railway gauge.
VIII.
The concentration of wealth and power now underway has no precedent in industrial history for its velocity, its scale, or its abstraction. The compute infrastructure required to train frontier AI models costs hundreds of millions of dollars per run. The training data has been assembled through decades of free behavioral contribution by billions of users who had no meaningful understanding of what they were building. The physical layer — the cables, the chips, the server farms — is owned by a vanishingly small number of entities. Each layer of this stack is a toll road, and the tolls compound.
What is new is the object of extraction: not cotton, not rubber, not oil, not labor in the classical sense — but cognition. The raw material of the twenty-first century economy is the structured output of human minds, digitized, aggregated, and used to train systems that will progressively displace the humans who generated the training data in the first place. The process that Harry Braverman identified in the deskilling of industrial labor — the systematic decomposition of skilled craft into measurable, manageable, replaceable components — has now reached the cognitive tier. The writers, analysts, designers, coders, diagnosticians, and translators who believed their work was too complex, too contextual, too human to be automated are discovering that what made their work valuable was precisely what made it trainable. The data they generated, freely and without compensation, is the capital that renders them disposable.
This is not metaphor. It is the next phase of a process that has been running for two centuries, now applied to the one domain that workers believed was safely theirs. The Metaproletariat is not a class defined by what it lacks. It is defined by what has been taken from it without its recognition or consent.
IX.
The question worth asking — the question colonial subjects have always eventually had to ask — is: who decided that this particular arrangement of ownership and governance was necessary? Who benefits from the premise that intelligence, at scale, must be privately held? What would it look like if the foundational infrastructure of the AI era were treated as a public good?
These are not naive questions. They have been answered before, in practice, by democratic societies that chose to treat foundational technologies differently.
The electromagnetic spectrum is not owned. It is licensed. The decision, made in the early twentieth century, that the airwaves constituted a public commons whose commercial use required public authorization was not technologically inevitable. It was political. The Human Genome Project made a deliberate choice in 2000 to release its sequencing data into the public domain rather than allow private patents — a decision that required political will against significant commercial pressure, and that has saved hundreds of billions of dollars in pharmaceutical research costs in the decades since. CERN, the multinational physics institution that produced the World Wide Web as a public good, is funded by member states and governed by international treaty. The internet’s foundational architecture was built with public money, placed in the public domain, and deliberately designed without a central owner. These are not historical accidents. They are records of political decisions that went differently than the AI governance decisions are currently going.
The policy levers exist. A public compute infrastructure — equivalent to the national laboratory system, accessible to researchers, governments, and civil society organizations — would break the capital barrier that currently makes frontier AI development the exclusive province of six companies. Mandatory licensing of foundational models for public interest use, analogous to compulsory licensing in pharmaceutical law, would make the tools accessible without requiring public ownership of the development process. Data trusts — legal structures that allow communities to pool and govern their own behavioral data rather than surrendering it individually to platforms — would shift the terms of the training data relationship. None of these require dismantling the private sector. They require treating the foundational layer of the AI economy the way democratic societies have, at their best, treated other foundational layers: as infrastructure, not inventory.
These questions are not being asked in the rooms where governance frameworks are being written. They are being crowded out by the urgency of safety discourse — which is real, which matters, and which also conveniently forecloses the prior question of whether the concentration of this capability in private hands is itself the primary safety problem. Safety governance that takes the ownership structure as given is not neutral. It is a choice about who the system is designed to protect.
Arendt believed, even against the evidence of the twentieth century, in what she called natality — the human capacity to begin something genuinely new, to act in ways that could not have been predicted from prior conditions. It was her counter-argument to determinism, to the totalizing systems that claimed to know what history required. The precedents exist. The mechanisms are known. What is required is the political will to treat this moment as a genuine decision rather than an inevitable one — to recognize that the structure being built right now is a choice, and that choices of this magnitude can be made differently.
History will record what we built in this window. The question is whether we are building something that distributes the gains from intelligence broadly, or constructing a new enclosure — one whose walls are made not of stone or wire, but of weights and tokens and proprietary compute, and whose lords speak the language of responsibility while practicing the logic of extraction.
The East India Company had a charter. The British South Africa Company had a charter. They both insisted, until the evidence was overwhelming, that their governance was in the interest of those they governed.
Structure before sophistication. Every time.
The structure being built right now is feudal. And the sophistication of its justifications does not change what it is.
Citations & Sources
1. Varoufakis, Yanis. Technofeudalism: What Killed Capitalism. Bodley Head, 2023. The concept of ‘cloud capital’ and the feudal rent-extraction model of platform economies is drawn directly from Varoufakis’s central argument.
2. Zuboff, Shoshana. The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power. PublicAffairs, 2019. The framework of ‘behavioral modification at scale,’ ‘prediction products,’ and the epistemological character of surveillance capitalism is drawn from Zuboff’s foundational text.
3. Arendt, Hannah. The Origins of Totalitarianism. Harcourt Brace, 1951. Part II: ‘Imperialism.’ The analysis of imperialism as capital overflow, the administrative apparatus of colonial governance, and the ‘boomerang effect’ — colonial methods returning to the metropole — are all developed in this section.
4. Arendt, Hannah. The Human Condition. University of Chicago Press, 1958. The concept of the ‘common world’ and its erosion by private accumulation; the distinction between the public realm and the private household economy; the political implications of common goods migrating into private hands.
5. Arendt, Hannah. Eichmann in Jerusalem: A Report on the Banality of Evil. Viking Press, 1963. The concept of the ‘banality of evil’ — institutional harm produced not by malice but by the systematic abdication of moral reflection — is used here as an analogy for algorithmic harm at scale.
6. Braverman, Harry. Labor and Monopoly Capital: The Degradation of Work in the Twentieth Century. Monthly Review Press, 1974. The deskilling thesis — the systematic decomposition of skilled labor into measurable, manageable components — is extended here to cognitive and knowledge work in the age of generative AI.
7. Davis, Mike. Late Victorian Holocausts: El Niño Famines and the Making of the Third World. Verso, 2001. Background on the structural violence of colonial economic administration, including the East India Company’s revenue extraction from Bengal.
8. Ferguson, Niall. Empire: The Rise and Demise of the British World Order and the Lessons for Global Power. Basic Books, 2002. Historical background on the East India Company charter model and the BSAC / Rhodesia land appropriation.
9. Rotberg, Robert I. The Founder: Cecil Rhodes and the Pursuit of Power. Oxford University Press, 1988. Historical background on Rhodes, the British South Africa Company, and the Pioneer Column of 1890.
10. The Battle of Plassey (1757) and the East India Company’s Diwani grant (1765) are documented in standard histories of British India. See also: Dalrymple, William. The Anarchy: The East India Company, Corporate Violence, and the Pillage of an Empire. Bloomsbury, 2019.
11. Truth and Reconciliation Commission of Canada. Calls to Action. 2015. Background on the residential school system as an administrative project of cultural and cognitive displacement.
12. Arendt, Hannah. ‘On Violence.’ In Crises of the Republic. Harcourt Brace Jovanovich, 1972. The distinction between power (legitimate, collective, consensual) and violence (instrumental, substituting for power when it fails) is drawn from Arendt’s political philosophy as developed across multiple works.
13. Couldry, Nick, and Ulises A. Mejias. The Costs of Connection: How Data Is Colonizing Human Life and Appropriating It for Capitalism. Stanford University Press, 2019. The concept of “data colonialism” as a structural continuation of historical colonialism — not metaphor but mechanism — is drawn from Couldry and Mejias’s central argument.
Structure Before Sophistication | structurebeforesophistication.substack.com