Artificial intelligence concentrated computing power in a handful of corporations and revived an old question in new clothing: do we still live under capitalism, or have we entered something else? Yanis Varoufakis named that unknown territory technofeudalism. His critics reply that the system has merely evolved into its most monopolistic version. The concept seduces and provokes in equal measure, while the numbers of the digital economy feed both sides of the debate.
The Return of the Feudal Lords
By: Gabriel E. Levy B.
The French economist Cédric Durand saw the phenomenon coming before anyone else. In 2020 he published Technoféodalisme, where he argued that the digital giants stopped producing goods and now charge a toll for access to their platforms, what economists call rents. Three years later, former Greek minister Yanis Varoufakis popularized the idea with Technofeudalism: What Killed Capitalism, where he contends that capital mutated into cloud capital: the infrastructure of servers and algorithms that governs digital life.
According to Varoufakis, platforms like Amazon stopped functioning as markets and now operate as digital fiefdoms, where the owner of the algorithm decides who appears before the buyer and how much tribute each seller pays to exist there. Companies that depend on them to sell act as vassal capitalists. Drivers and couriers directed by apps form a cloud proletariat. And users, by feeding the systems with free data and content, work as digital serfs without salary or contract.
The genealogy runs further back. Shoshana Zuboff described surveillance capitalism, the business of turning our behavior into data in order to sell predictions. McKenzie Wark argued that whoever controls information rules today. Jodi Dean speaks of neofeudalism. These voices share one diagnosis: in the digital economy, charging a toll pays better than producing and selling.
AI Turned the Metaphor into Infrastructure
Generative AI turned the hypothesis into measurable reality. Training and operating systems like ChatGPT or Gemini demands colossal computing power that belongs to very few hands. Jon Peddie Research calculates that Nvidia sells 92 percent of graphics cards, the chips that train AI. According to Synergy Research Group, three companies concentrate 63 percent of global server rental, the cloud: Amazon in the lead, with Microsoft and Google behind. The Magnificent Seven, the seven biggest listed tech companies, weigh more than a third of the S&P 500, the U.S. benchmark index.
A system of rents with medieval logic operates on that foundation. AI startups pay permanent tribute to the owners of the cloud for every query, and anyone who aspires to compete ends up renting the castle of a giant. Media outlets negotiate licenses under conditions they rarely control, while their archives feed other people’s models. An invisible army works behind every system: data labelers, the people who classify data to teach the machines, earn between one and three dollars an hour in countries like Kenya or Venezuela.
The fiefdom also captures attention: search engines now answer with AI-written summaries that keep the user on the page and, according to measurement firm Similarweb, almost 69 percent of news searches ended without a single click to media sites in May 2025. The traffic the open web once redistributed now stays home with the lord.
The Skeptics: Capitalism Is Alive and Well
Not everyone accepts the label. From the journal New Left Review, Evgeny Morozov replied that feudal language reveals more intellectual weakness than analytical precision, since Big Tech keeps making its profits under the rules of capitalism, in its most monopolistic version. Nick Srnicek prefers to speak of platform capitalism, where data works as raw material without the system changing its nature.
In January 2025, the Chinese company DeepSeek trained a competitive model for a fraction of what Silicon Valley had budgeted and erased close to a trillion dollars from stock valuations in days. Varoufakis responded with a distinction: AI services may be a bubble, but cloud capital, the infrastructure of chips and data centers, keeps its structural character. Software leaves room for competition. Concrete and silicon, far less.
The Lords of Compute Preach Openness
On July 24, 2026, twenty-five companies and organizations published the open letter Open Weights and American AI Leadership, a plea for open-weight models, AI systems that anyone can download and adapt without asking permission. Nvidia, Microsoft, Meta, and Mozilla sign alongside Hugging Face. Jensen Huang, chief executive of Nvidia, debuted his account on X, formerly Twitter, to share it. The text asks Washington to avoid sweeping restrictions, amid the debate over banning Chinese models like Kimi K3, and warns that concentrating the technology in a few closed systems creates single points of failure: if those systems go down, everything that depends on them goes down too.
Does the letter refute the feudal thesis or confirm it by another route? One reading sees a crack: if the giants defend models anyone can copy, competition remains alive. The opposite reading notices two details: neither OpenAI nor Anthropic signed, and every open model needs chips to train and clouds to run, so openness in software multiplies demand for chips and servers, where Nvidia has no rivals. The letter also invokes sovereignty, but American sovereignty, a nuance the Latin American reader should not forget.
Latin America: Between Vassalage and Sovereignty
For our region, the discussion stops being academic. Latin America consumes artificial intelligence in bulk and barely produces it. ECLAC, the UN’s economic arm in the region, warns that without decisive public policy it will end up buying an AI made by others. The figures confirm it: the region generates 14 percent of global visits to AI tools, but its companies do not reach 3 percent of the world total. Brazil concentrates 90 percent of regional supercomputing capacity, and more than half the countries lack critical infrastructure.
Telecommunications operators know the asymmetry: they deploy the networks that carry the traffic, while the giants capture the value in applications and now in models. Nick Couldry and Ulises Mejias named this pattern data colonialism: platforms appropriate everyday life the way historical colonialism appropriated territories and resources. Mexican scholar Paola Ricaurte adds that this extraction reproduces old colonial hierarchies, since those who decide which data matters operate far from the region.
Responses are beginning to emerge. Chile leads Latam-GPT, a ChatGPT-style model, open and trained on regional data by specialists from 15 countries, conceived as a public good. Brazil launched a national AI plan of some 4 billion dollars with its own supercomputing and a government cloud. They are modest efforts against the global fiefdoms, yet they point to the only path: building homegrown capacity before digital vassalage becomes irreversible.
In summary, the concentration of computing turned technofeudalism into something more than a provocative metaphor. A few corporations control the infrastructure of artificial intelligence and charge rent to businesses, governments, media, and citizens. The academic debate remains open, though the empirical evidence is overwhelming. For Latin America the dilemma is concrete: develop capacities of its own, as Latam-GPT attempts, or accept a role as serfs in the global digital fiefdom.
References
Couldry, N., & Mejias, U. A. (2019). The costs of connection: How data is colonizing human life and appropriating it for capitalism. Stanford University Press.
Durand, C. (2020). Technoféodalisme: Critique de l’économie numérique [Technofeudalism: A critique of the digital economy]. Éditions Zones.
ECLAC & CENIA. (2025). Índice Latinoamericano de Inteligencia Artificial, ILIA 2025 [Latin American Artificial Intelligence Index]. Economic Commission for Latin America and the Caribbean.
Morozov, E. (2022). Critique of techno-feudal reason. New Left Review, 133/134.
Nvidia, Microsoft, Meta, Hugging Face, et al. (2026, July 24). Open weights and American AI leadership [Open letter]. https://www.microsoft.com/en-us/corporate-responsibility/topics/open-weight/
Ricaurte, P. (2019). Data epistemologies, the coloniality of power, and resistance. Television & New Media, 20(4), 350-365.
Srnicek, N. (2016). Platform capitalism. Polity Press.
Varoufakis, Y. (2023). Technofeudalism: What killed capitalism. Bodley Head.
