‘We Can't Drink Data': India's Villages Take On the Data Centres

The phrase that stuck with me came off a hand-painted banner in Visakhapatnam: “We cannot drink data.” Children and activists carried it through the streets this summer, some of them painting handcuffs onto a Google logo, and it captures the whole argument in four words. The data centre protests spreading across India are not really about servers or software. They are about water, land and electricity — the physical inputs that the cloud quietly consumes — and about a simple question of who pays for them. As artificial intelligence scales up, the compute has to live somewhere, and in India that somewhere is increasingly farmland, coastline and already water-stressed cities.

We talk about “the cloud” as if it were weightless, a place your photos and prompts float off to. It is not. A data centre is a warehouse of machines that run hot, drink water to stay cool, and pull enormous, steady loads off the grid. When you route a chatbot query or train a model, that request lands in a building made of concrete, copper and coolant, on a piece of ground that used to be something else. I have written before that the cloud is physical, and nowhere is that becoming clearer than in the towns now organising against it.

Where the cloud touches the ground

The flashpoint that has drawn the most attention is a very large project reported near Visakhapatnam, on Andhra Pradesh’s coast — a data-centre campus said to run into the tens of billions of dollars, built with big global and domestic backers. By some estimates it would be one of the largest such investments anywhere. Local opposition has coalesced around two worries: the water it would draw in a city that, by reported accounts, already delivers less than it needs and rations supply for its residents; and its proximity to a wildlife sanctuary. There are legal challenges in progress. The state, for its part, points to jobs — very large employment figures have been floated — and to the prestige of anchoring a global company on Indian soil.

Visakhapatnam is the loudest case, not the only one. In agricultural belts of Andhra Pradesh and Telangana, farmers have reported pressure to sell paddy fields and mango orchards to make way for data-centre development. Around industrial corridors near cities such as Thane and in other clusters, residents and activists have raised similar objections about water draw and grid load. The specifics differ from place to place, and I would treat any single figure — litres per day, megawatts, acres — as illustrative rather than exact, because reliable public numbers are exactly what these communities say they are not being given. But the pattern is consistent enough to name.

India is a sharp place to watch this collision because the arithmetic is unforgiving. The country holds a large share of the world’s people and a small share of its fresh water, and many of the sites attractive to data-centre developers — good connectivity, cheap land, willing state governments — sit in regions that are already water-stressed. Drop a thirsty, always-on facility into that map and you have not created a shortage so much as sharpened one that was already there.

Water you can’t drink, power you can’t use

Cooling is the part people underestimate. Servers throw off heat, and one common way to manage it is evaporative cooling, which consumes fresh water — often the same treated, drinking-quality water a city supplies to homes. That is the bite behind the banner. A resident who is rationed in summer, watching a new facility connect to the municipal supply, does not need a hydrology degree to see the trade. I have gone deeper into that specific tension in a piece on AI’s water use in India, and the short version is that the water cost of computation is real, local and, so far, badly disclosed.

A data centre is not a shortage machine, but drop a thirsty, always-on facility into an already water-stressed map and you sharpen a scarcity that was already there.

Then there is the grid. These facilities want continuous, high-quality power — not the intermittent supply many Indian consumers still live with, but firm baseload, the kind that gets prioritised. When a state courts a marquee investor, it is often implicitly promising reliable electricity and, sometimes, favourable tariffs. Every unit of firm power committed to a server hall is a unit not going somewhere else, and the strain on generation and transmission is a cost borne by the wider system. The scale of that appetite is the subject of my longer look at AI’s energy consumption; the number that matters here is not any single plant’s draw but the aggregate, and the aggregate is climbing fast.

Water and power together produce a strange local outcome. The facility sits in your district, uses your aquifer and your grid, and yet the thing it produces — compute, and the AI services built on it — is sold into a global market. The model being trained might serve users on another continent. The revenue books somewhere with a lower tax rate. What stays behind is the draw on shared resources and, if things go wrong, the environmental cleanup.

Privatised gains, socialised costs

That asymmetry is the heart of it, and it is an old shape wearing new clothes. The gains from these projects — the profits, the share price, the strategic advantage in the AI race — are concentrated and private. The costs — water drawn from a stressed system, load added to a strained grid, farmland converted, ecosystems disturbed — are diffuse and public. Economists have a dry phrase for this: externalities. On the ground it feels less abstract. It feels like being asked to subsidise someone else’s business with your rainfall.

I want to be fair to the other side of the ledger, because the case for these investments is not empty. Data centres do bring construction work, some permanent skilled jobs, tax base and infrastructure that can spill over into a region. Digital sovereignty is a genuine argument: a country may reasonably want its data and its compute hosted domestically rather than renting everything from servers abroad. And a well-sited, well-cooled facility — using recycled or non-potable water, drawing on dedicated renewable generation, sitting on genuinely marginal land — can be far less harmful than the worst cases. None of that is imaginary.

But those benefits have to be weighed honestly, and honesty is exactly where the process tends to break. The jobs figures quoted by promoters are often peak or lifetime estimates, heavy on temporary construction and light on the permanent headcount a highly automated warehouse of machines actually needs. The water and power commitments are frequently negotiated behind closed doors, in memoranda of understanding the public never sees. When the people who will live with the consequences cannot get straight answers about how many litres, how many megawatts, drawn from where, the deal is not being weighed — it is being waved through.

When the people who will live with the consequences cannot get straight answers about how many litres and how many megawatts, drawn from where, the deal is not being weighed — it is being waved through.

Whose future is being built

This is the part that connects data centres to a larger pattern I keep returning to — the way powerful technologies get framed as inevitable, and how that framing quietly hands the benefits to a few while spreading the costs to everyone else. I have written about how technology gets captured: not seized in some dramatic coup, but steered, through the ordinary machinery of incentives, subsidies and closed rooms, until it serves the interests of whoever is already powerful. A tax holiday stretching decades into the future, offered to global cloud providers, is capture in its most bureaucratic and effective form. It is a decision, dressed as a foregone conclusion, that the public purse should absorb the risk while private balance sheets keep the upside.

The villagers with their banners are not anti-technology, whatever the convenient caricature says. I don’t read “we cannot drink data” as a rejection of the internet or of progress. I read it as a demand to be counted in the arithmetic — to have the water and the power and the land they depend on show up as real costs in the ledger, rather than as free inputs assumed away in a pitch deck. That is a reasonable demand. It is, in fact, the demand any of us would make if the warehouse were going up next to our fields.

What would it look like to take that seriously? A few things, none of them radical. Publish the water and energy commitments before a project is approved, not after it is built. Require closed-loop or non-potable cooling in water-stressed districts as a condition, not a courtesy. Tie the tax breaks and land deals to real, verifiable local benefit — permanent jobs that materialise, grid capacity that gets added rather than merely reallocated, water infrastructure that leaves a district better supplied than it found it. Let the communities who host the compute have a seat at the table where the trade-offs are actually decided.

The AI boom is going to need an enormous amount of physical infrastructure, and a lot of it will, and probably should, be built in countries like India. That is not the fight. The fight is over the terms — whether the ground the cloud touches is treated as a partner to be dealt with fairly or a cost to be quietly externalised. Right now, in city after city, the people standing on that ground are telling us which one it feels like. It is worth listening before the concrete sets, because once it does, the water is already spoken for and the bargain is already struck. You cannot renegotiate a reservoir after it has been drawn down. And you still, whatever the banner says, cannot drink data.