Amul's Lesson for the AI Age: What if the Workers Owned the Platform?

The Amul cooperative model began with a small, stubborn refusal. In the 1940s, dairy farmers around Anand, in Gujarat, were being squeezed by a private contractor that held a monopoly on collecting and selling their milk to the city. The farmers did the hard part — kept the cattle, milked at dawn, walked the cans to the collection point — and a middleman set the price, kept the margin, and left them with barely enough to keep going. In 1946, backed by the freedom-movement leaders Sardar Vallabhbhai Patel and Tribhuvandas Patel, the farmers of the Kaira district did something that still feels radical: they stopped selling to the middleman and built their own cooperative to do the collecting, the chilling, the processing, and the selling themselves. That decision is the whole story, and I think it is the most useful story we have as we walk into the age of platforms and artificial intelligence.

I keep coming back to Amul because it answers a question most technology conversations avoid. Not what can the machine do, but who owns the machine — and therefore who keeps the money it makes. That is the question of our decade, and milk farmers in Gujarat answered it seventy years before the rest of us thought to ask.

What the farmers actually built

The genius of what became the Anand pattern was not a new way to milk a cow. It was a new way to own the parts of the business that sit above the cow. A single farmer with two buffaloes has no leverage. She cannot build a chilling plant, afford a laboratory to test fat content, advertise a brand, or negotiate with a city dairy. So each of those layers — the thing that aggregates, tests, brands, and distributes — had always been owned by someone else, and that someone else took the cream, literally and financially.

The cooperative flipped the ownership. Farmers in each village formed a village society. The village societies federated into a district union. The union owned the processing plant, the testing labs, the cold chain, the trucks, and eventually the brand — Amul — and the marketing that made it a household name. The federation that became the Gujarat Cooperative Milk Marketing Federation (GCMMF) sat on top, marketing for all of them. Crucially, the farmer who supplied the milk was also, through her society, a part-owner of every layer above her. The margin that a private processor would have pocketed flowed back down to the people who produced the milk.

Verghese Kurien, rightly remembered as the architect of this, understood something the technologists of his day mostly did not: the point of the institution was not efficiency for its own sake. It was to keep control, and the surplus, in the hands of the producers. Under his leadership the model was carried nationwide as Operation Flood — the White Revolution — which over the 1970s and 80s turned India from a milk-scarce country into the largest milk producer in the world, built on tens of thousands of village cooperatives and millions of small farmers, many of them women.

The people who supply the value can own the machine that aggregates it — and keep the margin that the aggregator would otherwise take.

The lesson underneath the milk

Strip away the dairy specifics and here is the transferable idea: in almost every market, the real power sits with whoever owns the layer that aggregates. The individual producer is replaceable. The aggregator — the one who pools everyone's output, sets the standards, controls distribution, and owns the brand — is where the leverage and the margin concentrate. The default arrangement is that this aggregating layer is owned by outside capital, and the producers are treated as suppliers to be paid as little as competition allows.

Amul's lesson is that this default is a choice, not a law of nature. The producers can own the aggregator. When they do, the same efficiency, scale, and brand power still get built — but the surplus is distributed to the people who created it rather than extracted from them. That is not charity or a subsidy. It is the ordinary economics of aggregation, pointed in a different direction by a different ownership structure.

I have written before about how technology gets captured — how a tool that could have widened opportunity ends up narrowing it, because whoever owns the infrastructure sets the terms for everyone who has to use it. Amul is the same pattern run in reverse. The technology of the modern dairy — the chilling plant, the testing lab, the logistics network — was exactly the kind of infrastructure that gets captured. The cooperative simply refused to let it be owned by anyone other than the people whose labour fed it.

Why this maps almost too neatly onto the platform and AI age

Now look at the businesses that define our moment, and notice how familiar the shape is.

A ride-hailing app is an aggregator. The drivers own the cars, do the driving, take the risk, absorb the fuel and the wear. The platform owns the layer above the driver — the matching algorithm, the pricing engine, the brand, the customer relationship — and sets the commission. The driver is the Kaira farmer of 1945: essential, replaceable, paid whatever the intermediary decides. A food-delivery platform, a freelance marketplace, a short-video app that sells creators' attention to advertisers — same structure every time. The people who supply the value do not own the machine that aggregates it.

Artificial intelligence sharpens this to a fine point, because the raw material of AI is us. The models reshaping white-collar work were trained on text, images, and code produced by millions of people who were never asked, never paid, and own no share of what was built from their output. The value chain is the oldest arrangement there is — many producers, one aggregator — except the producers are now everyone who has ever written or drawn or photographed anything online, and the aggregator is a handful of firms with enough capital to train frontier models. The margin flows up. The people whose work became the training data are, once again, suppliers to a machine they do not own.

When I trace who takes, who pays, and who could fight back, this is where it always lands: the fight is not really about whether the technology is good or bad. Gig platforms are genuinely convenient; AI models are genuinely astonishing. The fight is about the ownership of the aggregating layer — and Amul is proof that this layer does not have to be owned by outside capital to work at enormous scale.

The honest difficulties of copying it in tech

I want to be careful here, because it would be easy to wave the Amul flag and pretend the copy-paste is simple. It is not, and pretending otherwise does a disservice to anyone who actually tries to build a producer-owned platform.

Capital. A district dairy union needed money, but the amounts were within reach of state banks, cooperative finance, and eventually government-backed programmes like Operation Flood. Training a frontier AI model costs a level of capital that no cooperative of freelancers can casually assemble. The chilling plant of the AI age is a data centre full of scarce chips, and that is a genuinely harder mountain than a milk-processing plant.

Network effects. A dairy cooperative competes on the price and quality of milk, a physical good sold locally. A platform competes on network effects — riders go where the drivers are, drivers go where the riders are — and an incumbent with millions of users on both sides is far harder to dislodge than a milk contractor in one district. A worker-owned rival does not just have to be as good; it has to overcome the gravity of a network that is already full.

Engineering and iteration. A dairy's technology, once built, changes slowly. A platform's software must be rebuilt constantly, which means a permanent, expensive engineering capability rather than a one-time capital outlay. Cooperatives have been good at owning stable infrastructure and less practised at running a fast product organisation.

Governance at scale. Amul itself shows the strain here. A three-tier structure of village societies, district unions, and a state federation is a serious act of institution-building, and even Amul has had its share of political capture and internal disputes over the decades. Democracy inside a large economic organisation is genuinely hard. Scaling one-member-one-vote governance to millions of globally dispersed gig workers or data contributors is a problem no one has fully solved.

These are real obstacles, not rhetorical ones. Anyone who tells you a worker-owned AI lab is a weekend project is selling something.

What a producer-owned alternative would actually need

And yet. The difficulties are reasons to be clever, not reasons to give up — and there is already a live movement asking exactly the right question. I have written about platform cooperatives, where drivers and couriers own the app they work through, and the early examples suggest what it takes to make the Anand pattern run on software.

A few things seem to matter most.

• Start where the capital requirement is survivable. You do not have to begin by out-training the largest AI labs. You begin where a cooperative can plausibly own the aggregating layer today: a ride-hailing or delivery app in one city, a stock-photo or translation marketplace, a data trust that licenses its members' contributions collectively. Amul did not start national. It started in one district and federated outward.
• Federate rather than centralise. Amul's real invention was the three-tier structure — small local units that keep the human-scale democracy, federated into something big enough to have market power. The software equivalent is networks of small worker-owned platforms sharing open protocols, brand, and back-office scale, rather than one giant coop trying to be democratic across millions of strangers.
• Own the standards, not just the labour. The cooperative's leverage came from owning the testing lab and the brand — the parts that set the terms of trade. For platforms, the equivalents are the reputation system, the data, and the customer relationship. A worker-owned platform that lets a third party own the ratings and the payment rails has given away the very layer Amul insisted on keeping.
• Get the state to build the road. Operation Flood was not pure bootstrapping; public policy and public finance built the rails that let cooperatives scale. The platform-cooperative movement will likely need the same — procurement rules, cooperative finance, and data-rights law that treat producer ownership as something worth backing.

Amul did not start national. It started in one district and federated outward — and that is exactly how a worker-owned platform has to begin.

I do not think any of this is naive, and I say that as someone allergic to techno-utopianism. The reason I trust the Amul lesson is that it is not a theory. It is a working institution that has paid small farmers for three-quarters of a century, in a poor country, against entrenched private interests, at a scale of millions. It sits alongside other Indian examples that prove ordinary people can rewrite the terms of a system that was built to extract from them — Kerala's land reform, which put land in the hands of the people who worked it, and the way India's farmers won their fight against laws written over their heads. None of these were gifts. Each was organised, and each rested on the same insight: the people who supply the value can, if they organise, own the machine that aggregates it.

The question worth sitting with

So here is the question I would put to anyone building — or being reshaped by — the platforms and models of this decade. When the aggregating layer gets built, and it will, who is going to own it? That single decision determines almost everything else: who captures the surplus, who sets the terms, and whether the technology widens opportunity or narrows it into a few hands.

Amul did not wait to be told the answer. Dairy farmers with no capital and no leverage looked at the middleman taking the cream, and decided that the machine which aggregated their milk should belong to them. That decision built one of the most successful economic institutions the country has produced. The tools have changed — from milk cans to matching algorithms to trained models — but the question has not. The people who supply the value can own the machine. The only thing standing between us and that outcome is whether we organise for it, the way the farmers of Anand did, with nothing but the milk they refused to keep selling on someone else's terms.