Part 1 of 3: Techno-Feudalism and the AI Divide
There is a word for what is happening and most people have not heard it yet.
The word is techno-feudalism. It was coined by the economist Yanis Varoufakis, and while it sounds like the kind of term academics invent to sell books, it describes something real and already in motion. The argument is simple: the major technology platforms are no longer competing in markets. They own the markets. And when you own the market, you do not earn profit. You collect rent.
I am a software engineer. I build things on these platforms every day. I pay for compute, storage, deployment, search, and inference. I do not experience any of these transactions as feudal. They feel like services. But Varoufakis's point is precisely that they are designed to feel that way. The platform looks like a marketplace. It is not. It is a fiefdom with a payment terminal.
What Varoufakis actually argues
The thesis runs deeper than "big tech is too powerful." It is a structural claim about how value moves through the economy.
In traditional capitalism, the engine is profit. A factory owner employs workers, sells products on an open market, and the difference between what workers are paid and what the product sells for is surplus value. Markets are the medium. Profit is the engine. Competition keeps the system alive.
Varoufakis argues that the major cloud platforms have replaced this dynamic with something older. Amazon does not primarily make money by selling goods. It makes money by charging every vendor a cut — reportedly around 51 cents on every dollar — for access to its customer base. Google does not make money by answering your questions. It makes money by selling your attention to advertisers who have no alternative channel of equivalent reach. AWS does not compete in a market. For a growing number of businesses, it is the market. You build on it or you are invisible.
This is rent, not profit. Rent, in the economic sense, is income derived from ownership of an asset that others must access, not from productive activity. A landlord does not produce housing. They own it, and you pay to exist inside it. Varoufakis calls this new asset class cloud capital — the platforms, servers, algorithms, and data infrastructure that every other economic actor now depends on. The owners of cloud capital are not capitalists. They are cloudalists. And the rest of us — developers, vendors, content creators, users — are some combination of vassals and serfs, paying rent in money, data, or unpaid labour.
The thesis is contested. Critics point out that platform companies still compete with each other, still face market pressures, still earn what looks like profit. The Jacobin review of Varoufakis's book made a fair point: advertising, the core business of cloud capital, is itself a market. Twitter's financial collapse under Musk hardly looks like the unassailable power of a feudal lord. These are legitimate objections.
But even if you do not accept that capitalism is dead — and I am not sure I do — the structural observation stands. A small number of companies now own the infrastructure that everyone else rents. That infrastructure is not optional. And the terms of access are set unilaterally by the owner.
If you are a developer, you already know this. You do not negotiate with AWS. You accept the pricing page.
Why AI makes this worse
Every previous wave of digital technology eventually became accessible enough to spread. Mobile phones leapfrogged landlines. Social media gave anyone a publishing platform. Web hosting became cheap. The barriers to entry, while real, lowered over time.
AI is moving in the opposite direction.
Training a frontier model requires capital measured in billions. Running inference at scale requires specialised hardware that a handful of companies manufacture, allocated through supply chains that a handful of companies control. The datasets these models learn from are vast, proprietary, and increasingly locked behind licensing agreements.
This is not a market that new entrants will disrupt from a garage. The compute required to compete at the frontier is itself a form of cloud capital. You rent it from the same companies whose models you are trying to compete with. The landlord and the competitor are the same entity.
The result is a tightening of the dynamic Varoufakis describes. Cloud capital was already concentrating power. AI accelerates that concentration by raising the minimum viable investment to participate. If you cannot train your own models, you rent access to someone else's. If you cannot build your own data infrastructure, you feed your data into someone else's pipeline. At every level, the dependency deepens and the rent increases.
The EU has attempted to respond to this with the AI Act, which establishes risk categories and transparency requirements. It is a serious effort and, on paper, more thoughtful than anything coming out of Washington or Beijing. But regulation assumes you are governing a market. If Varoufakis is right that the market has already been replaced by something else, the regulatory toolkit may be aimed at the wrong target. You cannot antitrust a landlord. You can only change who owns the building.
The geopolitics of cloud rent
There is a geopolitical layer to this that matters.
The United States dominates cloud infrastructure. AWS, Azure, and Google Cloud between them control the majority of global compute capacity. China has built a parallel system — Alibaba Cloud, Tencent Cloud, Huawei Cloud — but access to it is shaped by political alignment and subject to state oversight. Europe has regulatory ambition but limited cloud sovereignty. It remains, in Varoufakis's framing, a vassal territory: it sets rules about how the landlord must behave, but it does not own the building.
Africa barely registers. The entire continent holds less than one percent of global data centre capacity. Of the roughly 211 data centres that exist in Africa, nearly half are concentrated in just four countries: South Africa, Kenya, Nigeria, and Egypt. AWS has a single African region, in Cape Town. For the rest of the continent, cloud services are accessed over long-distance connections to servers in Europe or the Middle East, with the latency, cost, and dependency that implies.
This is not a gap that will close on its own. Cloud infrastructure is capital-intensive. It requires stable power, fibre connectivity, skilled labour, and regulatory certainty. These are preconditions that many African nations do not yet have — not because of any lack of capability, but because of structural underinvestment that has compounded over decades.
The consequence is that as more of economic and civic life runs through AI, entire regions of the world will participate only as tenants. They will rent access to models trained on data that does not represent them, running on infrastructure they do not own, subject to pricing and terms they cannot influence.
Varoufakis calls this cloud rent. I would call it something more familiar. Extraction has a long history on this continent. The commodity has changed. The structure has not.
What I am watching for
I do not think the question is whether techno-feudalism is the right label. The academic debate matters, but the structural reality is already here regardless of what we name it.
The questions that matter to me are practical:
Who owns the compute? Who sets the terms of access? Whose data trains the models, and who benefits from the output? When a nation's businesses, schools, hospitals, and government services all depend on infrastructure controlled by a foreign company, what does sovereignty actually mean?
These are not abstract concerns. They are engineering and policy problems that will define the next two decades. And they are not being solved by the people most affected.
In Part 2, I will look at what this means on the ground — starting with electricity, infrastructure, and the extractive dynamics that keep the AI divide in place.
This is Part 1 of a three-part series on techno-feudalism and the AI divide. Part 2: The Infrastructure Trap → | Part 3: The Lion Learns to Write →