The Return of the Toll Gate: Techno-feudalism, the Daily Wage Worker, and the Closing of the Digital Commons
This essay does four things. It sets out what actual feudalism was, because most invocations of it are lazy. It sets out the technofeudal thesis fairly, and then sets out the serious case against it, because the thesis is contested and anyone who tells you otherwise is selling something.
An extended essay on what medieval lordship can and cannot teach us about the platform economy
Introduction: The Mill You Are Not Allowed to Build
In an English village in the year 1300, a peasant woman with a sack of grain had a problem that was not really about grain. She could grind it at home with a hand quern, which was slow, or she could take it to the mill. The mill belonged to the lord. It was often the only mill she was permitted to use, because the lord held a legal right called soke, and if she was caught with a hand quern the bailiff could smash it. At the mill she surrendered a portion of her flour, the multure, typically somewhere between a sixteenth and a twelfth of what she brought. She did this every time, for her whole life, and her children did it after her.
The lord had not grown the wheat. He had not carried it. He had built a mill once, and then collected forever.
Seven centuries later, a rider in Kathmandu buys a motorbike on a loan, pays for the fuel, pays for the phone, pays for the data pack, maintains the machine, absorbs the risk of a crash, and delivers food. A platform that owns none of that takes a percentage of every trip. A shop owner in Pokhara who wants to reach customers online pays a marketplace roughly a third of gross revenue in fees before she has paid for a single item of stock. A developer in Bengaluru who writes an app pays a store between fifteen and thirty percent for the privilege of reaching a phone that the customer already owns. A startup founder anywhere on earth who wants to train a model rents the compute from one of three companies.
None of these are the same as the medieval mill. But the shape of the thing rhymes, and the rhyme has become the most interesting argument in political economy right now. Its name is technofeudalism.
It looks hard at what all this means for the person at the bottom of the structure, the daily wage worker, who is the least discussed and most affected party. And it examines the squeeze on small firms, which is where the "oligarchy" question actually bites.
Part I: What Feudalism Actually Was
The trap in the word
Historians fight about the word "feudalism" more than almost any other term in their discipline. Susan Reynolds argued in Fiefs and Vassals (1994) that the neat pyramid taught in schools, king above barons above knights above peasants, was largely constructed by sixteenth-century lawyers and nineteenth-century historians, and that medieval people did not experience their world as a tidy system. Marxists and Weberians use the word differently again. So when someone says "we are returning to feudalism," the first honest question is: which feudalism?
For our purposes the useful referent is not the military-legal system of fiefs and homage. It is manorialism, the economic arrangement that organised most of rural Europe for several centuries. That is where the analogy has real content.
The manor as an engine of extraction
The manor was land divided into two parts. The demesne was worked directly for the lord. The rest was held by peasants in return for obligations. Those obligations came in three flavours, and they mattered enormously.
Labour rent. The villein owed week-work, typically two or three days on the lord's fields, plus boon-work at harvest when the demand spiked. The peasant fed himself from his own strips, maintained his own plough and oxen, and then gave up his time. This point deserves emphasis because it recurs later in this essay: the peasant supplied his own tools and his own subsistence, and the lord captured the output. The lord's capital investment in the peasant's productive capacity was close to zero.
Rent in kind and in money. A share of the harvest, eggs at Easter, a hen at Christmas, and increasingly, as the centuries went on, cash.
Banalités, the monopoly dues. This is the category that matters most for the modern analogy. The lord held exclusive rights over infrastructure the community could not do without: the mill, the bread oven, the wine or oil press, sometimes the bull. Use was compulsory, and each use was taxed. Building your own was an offence. The lord did not compete with rival mills on quality or price. He held a jurisdictional right, backed by force, over a chokepoint.
On top of this sat a dense mesh of fees on life itself: merchet to marry off a daughter, heriot payable on death (often the best beast), chevage if you wished to live away from the manor, tallage levied more or less at will. And crucially, disputes were heard in the manorial court, where the lord or his steward presided. The party extracting the rent was also the party adjudicating whether the extraction was lawful.
The three features that actually define the structure
Strip away the pageantry and manorial feudalism had three defining properties:
- Economic power and political-legal power were fused. The lord did not merely own things. He governed. He made rules, enforced them, and judged violations of them. There was no neutral state standing above the relationship.
- Revenue came from position, not production. The lord's income derived from controlling access to something people needed, land and infrastructure, rather than from producing more efficiently than a rival.
- Exit was constrained. The serf was not a slave and could not be sold as chattel, but mobility was legally restricted, and even where it was not, there was usually nowhere to go.
How it ended, and why that matters
The standard story is that feudalism dissolved into capitalism through technological progress and moral enlightenment. That story is wrong in ways that are directly relevant to the present.
Feudalism in western Europe weakened primarily because of a catastrophe. The Black Death of 1348 to 1349 killed somewhere between a third and a half of the population of England, and comparable proportions elsewhere. Land became abundant and labour became scarce. Wages rose sharply. Peasants discovered that they could bargain, or simply walk to another manor that would ask fewer questions.
The response of the lords was not to accept the market outcome. It was to legislate against it. The Ordinance of Labourers (1349) and the Statute of Labourers (1351) capped wages at pre-plague levels, criminalised leaving employment, and made it an offence to offer higher pay. The first reaction of the incumbent class to labour gaining leverage was to make labour leverage illegal. Hold that thought; we will return to it when we discuss Proposition 22 and worker classification rules.
The statutes largely failed. Enforcement was patchy, the economic pressure was overwhelming, and resentment fed into the Peasants' Revolt of 1381. Over the following century, labour services were commuted into money rents, serfdom in England withered, and a class of wage labourers and tenant farmers emerged.
But here is the part that gets left out, and it is the single most important historical lesson in this essay. The same shock produced the opposite outcome elsewhere. East of the Elbe, in Poland, Prussia, Bohemia, Hungary and Russia, roughly the same period saw serfdom intensify. Landlords, facing booming grain demand from a urbanising western Europe, responded to labour scarcity by binding peasants more tightly to the land, extending labour dues, and capturing the state to enforce it. Historians call this the second serfdom, and in Russia it persisted until 1861.
Robert Brenner's celebrated argument about this divergence is that the outcome was determined not by technology, not by trade, and not by market forces, but by the balance of political and organisational power: how well peasants were organised, how strong village institutions were, and whether the state was captured by landlords or answerable to a broader coalition.
Two regions. Same technology. Same market opportunity. Same demographic shock. One got wage labour and eventually the Industrial Revolution. The other got two more centuries of bondage. The difference was politics.
The enclosure of the commons
One more piece of medieval and early modern history belongs here, because it is the closest historical parallel to what happened to data.
For centuries English villagers held customary rights over common land: grazing, gathering firewood, collecting fallen timber, gleaning after harvest. These were not owned in the modern sense. They were shared entitlements embedded in custom. Beginning in the sixteenth century and accelerating through parliamentary enclosure between roughly 1750 and 1830, some seven million acres were fenced, consolidated and converted into exclusive private property.
The enclosers had a genuinely strong efficiency argument. Enclosed land was more productive. Output rose. And the people who had lived off the commons lost the basis of their independence and became, in Karl Polanyi's account in The Great Transformation, the labour supply for the factories.
A resource that everyone used and nobody owned became a resource that one party owned and everyone had to pay to use, and the justification was efficiency, and the efficiency claim was true, and the dispossession was real at the same time. That sentence is the entire argument about data, and it is worth reading twice.
Part II: The Technofeudal Thesis
Who is arguing this
The idea has a lineage. McKenzie Wark's Capital Is Dead (2019) asked whether what came after capitalism might be worse rather than better. The French economist Cédric Durand published Techno-féodalisme in 2020, arguing that digital giants had stopped competing on production and had become collectors of tolls on access. Jodi Dean developed a parallel argument about neofeudalism. Yanis Varoufakis, the former Greek finance minister, gave the idea mass circulation with Technofeudalism: What Killed Capitalism (2023).
The core claim
Varoufakis argues that capitalism had two defining pillars: markets as the mechanism coordinating economic activity, and profit as the reward for successful production. He claims both have been displaced. Markets have been replaced by platforms, which are not marketplaces but privately governed territories where one party sets the terms of exchange. Profit has been replaced by rent, specifically what he calls cloud rent, extracted by the owners of what he terms cloud capital.
Cloud capital is his name for the accumulated stack of servers, data centres, networks, data and algorithms that mediate an increasing share of economic and social life. His claim is not that these companies are unusually greedy. It is a structural claim: that a firm which owns the arena in which others trade occupies a categorically different position from a firm competing inside the arena.
Evgeny Morozov, though a critic of the feudal framing, supplied a phrase that captures the mechanism well: platforms exercise pre-market power. They shape the conditions of exchange before any transaction occurs, by determining ranking, visibility, matching, pricing rules and eligibility. A firm with normal market power raises prices. A firm with pre-market power decides who gets to appear in the market at all.
Varoufakis also emphasises the post-2008 monetary environment. Enormous quantitative easing and near-zero interest rates for over a decade, in his telling, funnelled capital into asset accumulation by the platform owners rather than into productive investment, which is why asset valuations soared while wage growth and productivity for everyone else stagnated.
Who the serfs are, in this account
The thesis identifies three distinct groups performing something like unpaid or underpaid labour service:
The user. Every scroll, click, search, review, photograph and message trains the systems and enriches the dataset. Varoufakis's provocative framing is that billions of people perform unwaged labour that directly builds the owner's capital stock. This is his weakest empirical claim and his strongest rhetorical one.
The vassal capitalist. This is the crucial and underappreciated category. Sellers, app developers, publishers, hotels, restaurants and small manufacturers are genuine capitalists. They own capital, employ workers, take risk and earn profit. But an increasing share of what they earn is transferred upward to the platform as a condition of access. They are not workers. They are subordinate lords who hold their land from a superior lord and owe him a cut.
The platform worker. The rider, the driver, the warehouse picker, the data annotator, the content moderator. Formally independent, practically directed by an algorithm, bearing their own capital costs.
The numbers behind the claim
The empirical case rests on measurable concentration at chokepoints.
Cloud infrastructure. Amazon Web Services, Microsoft Azure and Google Cloud together hold roughly 63 to 68 percent of the global cloud infrastructure market as of early 2026, with AWS around 28 to 31 percent, Azure around 21 to 25 percent and Google Cloud around 11 to 14 percent depending on the analyst and the exact segment measured. Global cloud infrastructure spending reached about 129 billion dollars in the first quarter of 2026 alone, growing around 35 percent year on year. That combined share has been remarkably stable for two years even as the market has exploded, which is precisely what you would expect from structural rather than competitive advantage.
AI compute. Amazon, Microsoft, Alphabet and Meta collectively planned around 725 billion dollars of capital expenditure in 2026, up roughly 77 percent from the prior year's record. This is one of the largest private infrastructure build-outs in economic history, and it is being carried out by four companies. Whatever else it is, it is a moat made of concrete and silicon. Varoufakis has himself drawn the distinction: AI services may well be a bubble, but the underlying infrastructure of chips and data centres is structural. Software can be copied. Power substations cannot.
Marketplace take rates. Amazon's referral fee runs from 8 percent to 45 percent by category, most commonly 15 percent. Stack fulfilment, storage, surcharges and the advertising that is now practically compulsory for visibility, and the total effective take reaches roughly 34 percent by one widely cited estimate and up to 50 percent by another. In April 2026 a group of Amazon sellers organised a 24-hour advertising boycott in protest at margin compression. The core dynamic is fee creep: increments small enough to absorb individually, cumulatively transformative.
App stores. The standard 30 percent commission has finally cracked under legal pressure, but the crack is instructive. Google settled with Epic in March 2026 and now charges 20 percent or less on many transactions, and opened alternative billing in the United States, United Kingdom and European Union from 30 June 2026. But Google announced on 22 July 2026 that from 1 October 2026, US developers using external payment must report those transactions and pay Google service fees anyway. Apple, having been found in contempt in April 2025 and having lost at the Ninth Circuit in December 2025, proposed in August 2026 that it be permitted to charge 5 to 15 percent on purchases made through external links. Epic objects that this is far outside the bounds of the court's ruling, and Apple is seeking Supreme Court review.
Read that sequence carefully. Years of litigation, a contempt finding, an appellate loss, and the outcome under negotiation is not whether a toll is charged on transactions occurring outside the store, but how large that toll should be. The right to charge for access to your own customers, on a device the customer bought outright, is being treated as the baseline from which negotiation starts.
The jurisdictional argument
The strongest version of the technofeudal thesis is not about money at all. It is about governance.
Platforms write the rules. They enforce the rules. They adjudicate disputes about the rules. A seller suspended from a marketplace, a developer whose app is rejected, a creator demonetised, a driver deactivated: each appeals to a process owned and operated by the party that took the action. There is discovery of no evidence, no independent adjudicator, no precedent, often no stated reason, and frequently no human being.
That is the manorial court. Not metaphorically. Structurally. The identity of the rule-maker, the enforcer and the judge is the specific institutional feature that distinguishes lordship from ordinary commerce, and it is present.
Part III: The Case Against the Thesis, Taken Seriously
An essay that only prosecutes is propaganda. The technofeudal thesis has serious critics making serious arguments, and several of them are, in my reading, winning on points.
It is monopoly capitalism, not a new mode of production
This is the central counterargument and it is strong. Orthodox economists have a perfectly good term for above-normal returns arising from lack of competition: economic rent. Rent-seeking is not the opposite of capitalism. It is a permanent feature of it, extensively theorised by Ricardo, Marx and every industrial organisation economist since. Railroads, telegraphs, Standard Oil, the Bell System and IBM all occupied chokepoints and extracted from those positions. We did not call them feudal. We called them monopolies and, sometimes, we broke them up.
If everything in the technofeudal analysis can be described using the existing vocabulary of monopoly, market power and rent extraction, the new word may be adding heat rather than light. The American Affairs framing of the debate as "technofeudalism versus total capitalism" captures it: perhaps this is not capitalism's successor but capitalism at its most concentrated.
Feudal lords did not spend a hundred billion dollars a year on R&D
Morozov's objection cuts deep. The defining behaviour of a feudal lord was passive collection. He held a position and drew income from it without needing to innovate. Whatever else the hyperscalers are doing, they are not passive. Alphabet, Amazon, Microsoft and Meta spend staggering sums on research and infrastructure, and they do so under visible competitive pressure. The 725 billion dollar capex figure that supports the "moat" argument equally undermines the "idle rentier" argument. Lords who must spend three quarters of a trillion dollars a year to keep their position are not lords in any meaningful medieval sense. They are capitalists in an arms race.
The thesis idealises a capitalism that never existed
To say capitalism has been replaced by something worse, you must specify what it was before. If the reference point is a competitive market of small producers coordinating through price signals, that world has been rare and brief wherever it existed at all. Varoufakis has responded to this criticism directly, arguing it is misplaced and that the point is a structural change in the dynamic of accumulation rather than a moral comparison. Readers can judge for themselves whether that response is sufficient.
The legal reality of serfdom is absent, and this matters enormously
A serf could be whipped for leaving. A rider can quit at any moment and no one will pursue them. Being economically constrained and being legally bound are not the same thing, and collapsing the distinction insults the people who actually lived under bondage. Exit from a platform is expensive, difficult and often impractical. It is not illegal. That difference is not a technicality; it is the difference that made the entire history of emancipation possible.
Courts sometimes find these markets contestable, and they may be right
In November 2025, Judge James Boasberg ruled against the Federal Trade Commission in its long-running case against Meta, finding that the agency had failed to prove Meta currently holds monopoly power in personal social networking, largely because TikTok and YouTube compete for the same attention. The court noted that the great majority of time on Facebook and Instagram is now spent watching algorithmically recommended video from strangers rather than content from friends. The FTC appealed in January 2026.
You may think that ruling was wrong. Plenty of serious people do. But it is a reminder that a company can look invincible and then lose an enormous share of attention to a competitor from a different country within a few years. Feudal lords did not face TikTok.
The honest verdict
My assessment: technofeudalism is a bad literal description and an extremely useful analytical lens.
It is wrong that capitalism has ended. Competition is fierce, the incumbents invest ferociously, and the legal bondage that defined serfdom is absent.
It is right that a growing share of economic activity is mediated by privately governed chokepoints where the operator sets rules, adjudicates disputes, and extracts a share of every transaction; that this share is increasing over time; and that our institutions, built for the regulation of competing producers, are poorly designed to handle firms that own the arena itself.
Use the analogy for what it illuminates. Drop it when it starts doing your thinking for you.
Part IV: The Daily Wage Worker
This is the section that most writing on technofeudalism skips, because most of it is written from inside wealthy economies by people thinking about advertising and antitrust. But the person with the least room to manoeuvre is the daily wage worker, and there are three distinct populations to consider.
Tier one: the worker whose employer became an interface
The rider, the ride-hailing driver, the delivery courier, the on-demand cleaner. Formally, an independent partner. Practically, a directed worker.
The structural parallel to manorialism here is precise, and it is not the one people usually reach for. It is not that the worker is bound. It is the allocation of capital costs.
The medieval peasant owned his own plough, his own oxen, his own tools, and fed himself from his own strips of land. The lord contributed the land and the legal right, and captured a share of the output. He did not have to buy the peasant's plough or feed the peasant's family.
The platform worker owns the motorbike, pays the loan on it, buys the fuel, buys the phone, pays for the data, pays for repairs, absorbs the depreciation, and absorbs the risk of injury and of a bad day with no orders. The platform contributes the matching layer and the brand, and takes a share of every transaction. The economics of who supplies the means of production, and who captures the surplus from it, are genuinely and remarkably similar. This is the strongest single point in the whole technofeudal analogy and it deserves more attention than the "users as serfs" framing that gets the headlines.
Layer on top of this the specific features of algorithmic management:
- Opaque pay formulas. Rates change without notice or explanation. Two workers doing identical trips may be paid differently, and neither can discover why.
- Gamified compliance. Streaks, quests, surge windows and acceptance-rate thresholds that function as scheduling control while preserving the legal fiction of flexibility.
- Rating as discipline. A customer's momentary irritation, entered as a number, becomes a permanent mark that can throttle access to work.
- Deactivation without due process. The single most consequential feature. A person's entire livelihood can be terminated by an automated system, often with no stated cause and no meaningful appeal.
An OECD employer survey in 2025, covering just over six thousand managers across six countries, found that around 90 percent of US managers and on average about 79 percent of managers in the surveyed European countries worked in firms using at least one algorithmic management tool. This is not a gig-economy curiosity. It is becoming how work is organised generally.
Tier two: the informal daily labourer whose market is being intermediated
This is the population that essays about the platform economy almost never discuss, and in South Asia, Africa and much of Latin America it is the large majority of the workforce.
The mason waiting at a labour chowk. The electrician. The domestic worker. The porter. The small transport operator. Historically, their market was local, informal, socially embedded and relationship-based. It was insecure and often exploitative, and nobody should romanticise it. But it had one property worth naming: no intermediary owned the interface between the worker and the customer.
As booking apps, aggregators and marketplaces expand into these trades, that changes. Access to customers becomes access to a platform. A worker who was previously known in a neighbourhood becomes a profile with a rating. Bargaining, which was informal but was at least bilateral, becomes a posted rate set by a third party. Once the intermediary reaches sufficient density, the worker's option to work outside it shrinks, and the terms are no longer negotiable.
The danger for these workers is not that platforms are worse than what came before. In many cases they are better: more reliable payment, wider reach, some protection against being cheated by a client, and, importantly, access to work for people excluded by caste, gender, or social networks from traditional channels. The danger is that the improvement arrives permanently bundled with a permanent toll and a permanent loss of the right to set your own terms, and that the toll is adjustable upward by a party you cannot negotiate with.
Tier three: the invisible worker in the machine
Data annotators, content moderators, transcription workers, AI trainers, and the enormous population of remote freelancers on global bidding platforms.
Nepal is an instructive case. The Online Labour Index indicates that around 49 percent of Nepal's online freelance workforce is engaged in software and technology development. That is a genuinely high-skill export. But the structure of global freelancing platforms means Nepali professionals bid against professionals worldwide, in a market where the platform sets the fee, controls the reputation system that makes a worker employable, holds the funds, and where payment rails are frequently more restricted for workers in low-income countries than for clients in high-income ones. Value is created locally and a share of it is captured by an intermediary in another jurisdiction, permanently.
For content moderation and data annotation specifically, the work is routed to lower-wage countries precisely because it is cheap there, it is psychologically demanding in the case of moderation, it is often mediated by outsourcing firms that insulate the platform from liability, and the workers who make advanced AI systems usable are almost entirely absent from the story told about those systems.
The Nepali picture, concretely
Nepal offers an unusually clear view of these dynamics because its platform economy grew fast and its labour law did not move at all.
By 2024 estimates from the Asian Productivity Organization, the gig economy contributed roughly 7 percent of Nepal's GDP. Pathao alone engages over 200,000 people, including around 190,000 motorbike riders, more than 10,000 cab drivers and over 5,000 delivery personnel. Foodmandu, the oldest surviving platform in the country, operates with roughly 400 to 500 active riders and states that it complies with Nepali labour law including minimum wage, insurance and enrolment of eligible workers in the Social Security Fund. Yango, inDrive, Tootle and the payment platforms fill out the landscape.
Three structural problems stand out.
Legal invisibility. Nepal's principal labour statutes were built around an identifiable employer and a contract of employment. Platform workers fit neither category cleanly, so they fall into a gap. Ride-sharing operated for years in a grey zone where riders could be fined under transport legislation that predates the business model entirely, which means the worker absorbs the regulatory risk of an arrangement they did not design.
Social protection that does not fit the work. As Foodmandu's founder Manohar Adhikari has publicly noted, enrolling short-term workers in the Social Security Fund often provides little immediate benefit, because contributions are locked until the age of sixty. A person who works six months and moves on has paid into something they cannot touch. A scheme designed for lifetime formal employment does not map onto episodic work, and workers respond rationally by avoiding it.
No floor and no recourse. Riders bear fuel, maintenance, loan repayments and traffic fines, with no guaranteed minimum income, no injury cover in many cases, and no process for contesting a deactivation. In July 2026, the death of a ride-sharing driver in Kathmandu prompted a public reckoning with exactly these conditions and with how little legal standing the workers have. The riders quoted in the coverage described a daily calculus dominated by enforcement risk and income uncertainty. Whatever one concludes about the broader economics, that is a policy failure with a name and a date attached.
The international response, which just arrived
Something significant happened in June 2026 that has been under-reported outside labour circles.
On 12 June 2026, the International Labour Conference adopted the Decent Work in the Platform Economy Convention, 2026 (No. 193), by 406 votes to 8 with 36 abstentions. It is the first binding international labour standard dedicated to platform work, covering an estimated 150 million-plus workers worldwide. Its central design choice is the important one: rather than fighting the classification war directly, it requires that core protections apply regardless of employment classification. Minimum wage, occupational safety and health, social protection, and freedom of association apply whether a worker is called an employee, a contractor or a partner.
It also addresses algorithmic management in binding international law for the first time. Platforms must disclose how automated systems allocate work, set pay, generate ratings and deactivate accounts, and must provide explanations and genuine human review of consequential decisions. It contains protections against discriminatory or unlawful suspension and deactivation, and against misclassification.
Two caveats matter. First, the Convention is not self-executing; each state must ratify and then legislate. Second, the voting pattern is revealing. The United States voted against. India, which has one of the largest platform workforces on earth, abstained, having argued through the process that a non-binding recommendation would be preferable. Meanwhile the US Department of Labor proposed a rule in February 2026 that would refocus worker classification on control and opportunity for profit or loss in a way that makes contractor classification easier, and the EU's Platform Work Directive, which introduces a presumption of employment and shifts the burden of proof onto platforms, faces a member state transposition deadline of December 2026.
So the international legal position is not converging. It is splitting, sharply, with Europe moving toward presumption of employment, the United States moving away from it, and much of the global South, including Nepal, yet to legislate at all.
This is the Brenner divergence happening in real time. Same technology, same business models, same global market, and outcomes diverging entirely on the basis of political organisation and state capacity. It is the strongest evidence available that none of this is technologically determined.
Part V: The Small Company Problem
The user asked specifically about smaller companies not getting room to grow. This is where the oligarchy question is most concrete and most measurable.
Self-preferencing: the referee who also plays
If you operate the marketplace and also sell in it, you have information no competitor has and a placement power no competitor can match. You see every seller's conversion rate, price elasticity, search volume and margin. You decide which results appear first.
This is no longer contested as a matter of fact. On 23 July 2026, the European Commission fined Google 890 million euros in two decisions under the Digital Markets Act: 460 million for self-preferencing its own shopping, hotel, travel and sports services above third-party rivals in Search, and 430 million for restricting app developers on Google Play from telling their customers about cheaper purchase channels elsewhere. Google was given 60 days to comply or face periodic penalty payments of up to 5 percent of worldwide turnover. It was the largest DMA penalty to date, exceeding the combined 500 million euro fine against Apple and 200 million euro fine against Meta from April 2025, and it pushed Google's cumulative EU liabilities above 10 billion euros. In September 2025 the Commission had separately fined Google 2.95 billion euros over advertising technology, and in early July 2026 Google lost its final appeal against the 4.1 billion euro Android fine.
In the United States, Judge Leonie Brinkema found in April 2025 that Google had unlawfully monopolised the publisher ad server and ad exchange markets. Judge Amit Mehta had found in August 2024 that Google unlawfully maintained a monopoly in general search.
The pattern of findings is consistent across two continents and multiple courts.
But look at what the remedies actually did
This is the more important observation for anyone asking where we are heading.
Mehta's remedies decision in September 2025 declined to order divestiture of Chrome, declined to ban payments for default placement, and imposed limited data-sharing obligations. Those remedies took effect on 3 February 2026. On the same day, the Department of Justice and 35 states filed notices of appeal seeking stronger structural relief; Google had already appealed the data-sharing requirements. The D.C. Circuit is unlikely to resolve this before 2027, and the case was originally filed in 2020.
Note the timeline: a case filed in 2020, liability found in 2024, remedies in 2025, in effect in 2026, appeals running to 2027 or beyond. A startup that was harmed by the conduct in 2020 does not exist in any form that a 2027 remedy can help. Antitrust operates on a legal timescale and platform markets operate on a product timescale, and the gap between them is where small companies die.
The Commission's 890 million euro fine amounts to less than one percent of Google's recent quarterly profit. Enforcement is real, but as a deterrent priced against the value of the position being defended, the arithmetic does not obviously favour compliance.
The kill zone
Venture investors use the phrase "kill zone" for the space around a dominant platform's core business where funding dries up because everyone knows how the story ends: the incumbent copies the feature, bundles it free, gives it default placement, and the startup dies with a superior product.
The chilling effect is upstream of any illegal act. Capital does not get allocated. Founders do not start the company. The harm is invisible in enforcement statistics because it consists of businesses that were never created, and no legal system in the world is good at remedying that.
Acquisition then becomes the rational exit, which routes talent and technology into the incumbent, which is how a market with visible startup activity can nevertheless consolidate over time. Whether this specific dynamic is illegal is genuinely contested: Judge Boasberg's November 2025 ruling in the Meta case concluded that the FTC had not proven present monopoly power despite the acquisitions of Instagram and WhatsApp.
The compounding stack
The real squeeze on a small firm is not any single toll. It is that the tolls stack, and that the tolled party has no counterparty to negotiate with at any layer.
A small online business in Kathmandu or Kigali or Karachi typically pays: a marketplace commission or an advertising spend to be discoverable; a cloud bill to one of three providers; a payment processing fee; an app store commission if it has an app; and increasingly an AI inference bill. Each is individually defensible as a fee for a service. Cumulatively, a meaningful fraction of gross revenue leaves the business before it has paid a single employee, and every one of those rates is set unilaterally and can be revised upward.
That is the structure the technofeudal thesis is pointing at. Not one lord, but a stack of them, each holding a chokepoint, each collecting a multure.
The AI layer, which is the next enclosure
Everything above describes the settled state of the last decade. The AI build-out is doing it again, faster, and with higher barriers.
Frontier model training requires capital at a scale that only a handful of firms and states can supply. The 725 billion dollars of 2026 capex from four companies is not a number that a competitive market of small entrants produces. The chips are supplied by an extremely concentrated vendor base. The data centres require power infrastructure that takes years to permit and build. The result is that most AI startups are not competitors to the hyperscalers; they are tenants on hyperscaler land, paying rent in the form of compute bills, often with the hyperscaler as an investor as well as a supplier.
Regulators have noticed. The European Commission's specification decisions of 16 July 2026 require Google to share anonymised search data with rival search engines and AI services from January 2027, and to open eleven Android system-level features to competing AI assistants by the next major Android release in July 2027. The Commission has also opened market investigations into cloud computing services, and has signalled concern that AI-generated summaries at the top of search results may constitute a new expression of the same self-preferencing problem.
That last point matters enormously for small publishers and independent websites. For twenty-five years the implicit bargain of the open web was that a search engine indexed your content and sent you traffic in return. An answer generated directly in the search interface uses the content and returns nothing. Whatever one thinks of the legal question, the economic question is straightforward: the business model that funded independent online publishing depended on a referral flow that is being reabsorbed by the intermediary.
That is enclosure. A commons that everyone contributed to and everyone drew from, converted into an input for a private system, with a genuine efficiency gain and a genuine dispossession occurring simultaneously, exactly as in eighteenth century England.
Part VI: What History Actually Predicts
Historical analogies are not predictions. But history does supply a few well-tested regularities, and they are more useful than forecasting.
1. Extractive positions end when the extracted party gains leverage, not when the extractor develops a conscience. Serfdom weakened in the west because a demographic catastrophe made labour scarce. Nothing in the moral character of the landlord class changed.
2. The incumbent's first response to leverage is to make leverage illegal. The Statute of Labourers criminalised leaving employment and offering higher wages. The modern equivalents are visible: ballot measures that exempt platforms from labour law, mandatory arbitration clauses that prevent collective claims, classification rules revised to reduce coverage, and lobbying against binding international standards.
3. Identical conditions produce opposite outcomes depending on political organisation. Western Europe got emancipation. Eastern Europe got the second serfdom. The variable was the balance of power between organised producers, landlords, and the state. This is the single most important thing to take from the medieval record, and it is a hopeful lesson as much as a warning: outcomes here are not technologically determined, and they are still open.
4. Efficiency gains and dispossession are not mutually exclusive. Enclosure raised agricultural output and destroyed customary rights. Both are true. Anyone who tells you platforms are purely extractive is ignoring the first half; anyone who tells you the efficiency gain settles the question is ignoring the second.
5. New structures are usually named late. Nobody living in 1450 announced the end of feudalism. The people arguing today about whether this is capitalism or something new are engaged in a legitimate exercise, but the label is likely to be settled retrospectively, and probably by neither camp.
Part VII: Three Plausible Trajectories
Entrenchment. Enforcement continues but stays priced as a cost of doing business. Fines land at well under one percent of relevant profits, appeals consume the years in which markets are actually decided, and remedies arrive after the affected competitors are gone. Take rates ratchet upward in increments. AI concentrates further because its capital requirements are higher than anything before it. Platform work grows as a share of employment without acquiring the protections of employment. Most of the world outside the EU does not legislate at all.
Regulated gatekeeping. The European path generalises. Ex-ante rules replace after-the-fact litigation, because the DMA's core insight is that in fast markets you must specify obligations in advance rather than prove violations for six years. Interoperability and data portability become mandatory. ILO Convention 193 is ratified by a critical mass of states and floors become classification-neutral. The platforms remain very large and very profitable, but they become regulated utilities with defined obligations rather than private governments. This does not end concentration. It ends unaccountable concentration, which is a different and more achievable goal.
Fragmentation. Geopolitics splits the stack. Sovereign clouds, national AI programmes, data localisation and divergent regulatory regimes produce several regional blocs. Note carefully that this is not obviously an improvement for the worker or the small firm: replacing three global lords with eight regional ones, several of them state-affiliated, changes who collects the rent rather than whether it is collected. For small open economies, fragmentation may simply mean choosing which bloc's infrastructure to depend on.
The realistic future is a mixture, varying sharply by jurisdiction. And that variation is the point.
Part VIII: What Would Actually Change the Structure
Diagnosis without prescription is entertainment. Here is what follows from the analysis.
For the structure of markets
Prefer ex-ante rules to after-the-fact litigation. Six-year cases cannot govern eighteen-month product cycles. Whatever one thinks of the DMA's substance, its procedural design is correct: state the obligations in advance and attach a compliance clock.
Mandate interoperability and portability, and mean it. The single most powerful lever against a chokepoint is making it possible to leave without losing everything. Number portability transformed mobile telecoms. The equivalent for social graphs, seller reputations, worker ratings and business data would do more for competition than any fine. A rider's five-year rating should belong to the rider and travel with them.
Separate the referee from the player. Where a firm both operates a marketplace and competes in it, either the structural conflict is removed or the conduct is constrained by enforceable, auditable, non-discriminatory ranking rules.
Fund public digital infrastructure. India's experience with public payment rails demonstrates that a state can build a shared layer that private firms compete on top of, rather than allowing a private firm to own the layer itself. This is the single most transferable lesson for smaller economies: some infrastructure should be a road, not a toll gate.
For the worker
Decouple protections from classification. This is Convention 193's design and it is correct. Fighting the employee-versus-contractor war country by country has produced twenty years of litigation and very little protection. Attaching a floor to the work rather than to the label bypasses the entire fight.
Establish algorithmic due process. A stated reason, a human review, and a real appeal before a livelihood is terminated. This is the manorial court problem, and the remedy is the same one that ended the manorial court: separate the adjudicator from the interested party.
Require pay transparency. A worker must be able to see how their pay is calculated. Discriminatory personalised pricing of labour is invisible without disclosure.
Build portable, pro-rated benefits. Contributions that follow the worker across platforms and are accessible on a timescale that matches how people actually work. The Nepali critique of locking short-term workers' Social Security Fund contributions until age sixty is exactly right and is fixable through a design change, not a philosophical revolution.
Protect the right to organise. Every historical gain by extracted classes came from organisation. Riders' associations, freelancer collectives and data-worker unions are the mechanism by which any of the above becomes enforceable rather than aspirational.
For small and medium enterprises, especially in smaller economies
Own the customer relationship you can own. Every direct channel, whether an email list, a physical location or a phone number, is a channel that no one can tax or switch off. Platform-only businesses are tenants; the rent is set by someone else.
Treat platform dependence as a measurable risk. If a single channel supplies most of your revenue, that is not a growth strategy, it is a concentration exposure. Price it accordingly.
Push for domestic rails. Local payment infrastructure, local data centres where economically viable, and cooperative or publicly backed platforms in sectors where a single private intermediary would otherwise capture the whole market. This is industrial policy, and it is what every country that successfully industrialised actually did.
Cooperative ownership is not utopian. Driver-owned and rider-owned platforms exist and work at small scale. They struggle to raise capital against venture-funded competitors, which is a solvable financing problem rather than an inherent flaw in the model. Public procurement and development finance can solve it where political will exists.
Conclusion: The Multure and the Mill
Return to the woman with the sack of grain.
The most important thing about her situation was never the size of the multure. Whether the lord took a twelfth or a twentieth mattered to her household budget, but it was not what defined her position. What defined it was that she could not build her own mill, could not use a competitor's, and could not appeal the rate to anyone who was not the lord. Price was a symptom. Jurisdiction was the disease.
That is the right frame for the present. The argument is not really about whether 15 percent or 30 percent is a fair commission, or whether a rider's effective hourly earnings clear a minimum wage. Those are important fights, and they should be fought. But underneath them sits the structural question: who writes the rules of the arena, who enforces them, who judges disputes about them, and to whom can any of it be appealed?
Wherever the answer to all four is the same party, you have something that deserves the name lordship, whatever else we decide to call the economic system that contains it. That is true of a manor in 1300 and it is true of a platform in 2026.
But the medieval record also contains the reason not to despair. Feudal relations did not persist because they were technologically necessary. They persisted where the people subject to them were poorly organised and the state was captured, and they collapsed where the balance of power shifted. Western Europe and Eastern Europe faced the same plague, the same grain markets and the same technology, and arrived at emancipation and re-enserfment respectively. The variable was politics all the way down.
We are watching that divergence recur in real time. Europe is legislating presumption of employment and ex-ante gatekeeper obligations. The United States is loosening classification rules and losing a landmark monopolisation case on market definition. The International Labour Organization has, for the first time in its history, written a binding standard for platform work, and two of the countries with the largest platform workforces on earth declined to support it. Nepal has a platform economy worth roughly 7 percent of its GDP and a labour law that does not acknowledge its existence.
None of this is settled. That is the whole point.
A Note on the Argument
This essay has presented the technofeudal thesis in its strongest form and then argued that it is literally false and analytically valuable. Readers should know that this position is contested from both directions.
From the left, critics including Evgeny Morozov argue that the feudal framing mystifies rather than clarifies, that it substitutes a moral vocabulary of corruption and parasitism for an analysis of capital accumulation, and that it risks reconstructing a defeated enemy, the idle aristocrat, in place of the actual adversary, which is capital operating normally under conditions of extreme concentration. Some Marxist critics add that calling these firms "cloudalists" rather than capitalists obscures the class relation rather than sharpening it.
From the mainstream of economics, the objection is that everything described here has an existing name. Network effects, switching costs, two-sided markets, vertical integration and economic rent are well-developed concepts that explain platform dominance without requiring a new epoch. On this view the correct response is better antitrust and better regulation, not a new theory of history. Judge Boasberg's Meta ruling and the appellate skepticism toward structural remedies in the Google case both reflect a genuine and defensible judicial position: that these markets are more contestable than critics claim, and that courts should be cautious about restructuring fast-moving industries.
From the technology industry, the reply is that platforms created enormous consumer surplus, that most of the services in question are free or nearly free at the point of use, that the aggregation of demand is precisely what lets a small seller in a small country reach a global market at all, that commissions fund infrastructure and fraud prevention and logistics that the seller could not build, and that the counterfactual for most small firms is not independence but invisibility. This argument is stronger than critics usually admit, particularly for producers in small or landlocked economies.
And the empirical questions are genuinely open. Whether platform take rates are rising in real terms across the whole economy or only in specific segments; whether platform work raises or lowers earnings relative to the alternatives actually available to the workers who choose it; whether AI concentrates or diffuses market power over a ten year horizon; whether ex-ante regulation improves or reduces innovation. Serious researchers disagree on all four. Anyone who presents any of them as settled is arguing a position, not reporting a finding.
The honest conclusion is that the medieval analogy is a lens, not a proof, and that the questions it makes visible are more valuable than the label it supplies.
Sources and Further Reading
On the technofeudal thesis: Cédric Durand, Techno-féodalisme (2020); Yanis Varoufakis, Technofeudalism: What Killed Capitalism (2023); McKenzie Wark, Capital Is Dead (2019); Jodi Dean on neofeudalism; Evgeny Morozov, "Critique of Techno-Feudal Reason," New Left Review (2022), for the leading critique.
On medieval economic history: Susan Reynolds, Fiefs and Vassals (1994), on why the term itself is contested; Robert Brenner and the "Brenner debate" on the divergence between western emancipation and the eastern second serfdom; Karl Polanyi, The Great Transformation, on enclosure and the creation of labour markets; E. P. Thompson on customary rights.
On current enforcement: European Commission Digital Markets Act decisions of 23 July 2026 (Google, 890 million euros) and April 2025 (Apple, 500 million euros; Meta, 200 million euros); United States v. Google LLC (Mehta, liability 2024, remedies 2025, on appeal); the April 2025 adtech monopolisation ruling; FTC v. Meta (Boasberg, November 2025, on appeal); Epic Games v. Apple and Epic Games v. Google.
On platform labour: ILO Convention No. 193, Decent Work in the Platform Economy (adopted 12 June 2026); EU Platform Work Directive 2024/2831 (transposition deadline December 2026); Asian Productivity Organization data on Nepal's gig economy; reporting by the Kathmandu Post on Nepali platform workers and social protection.
