A roughly $15 billion AI data centre project in India faces local opposition over water consumption and impact on wildlife. After months of announcements where the only limits mentioned were electricity and silicon, here is an obstacle of a different kind: the people who live where the construction is planned.
We covered this summer the race for compute, multi-billion-dollar leases, and even plans for orbital computing. All those analyses shared a blind spot: they reasoned in terms of technical resources, not inhabited territories. This dossier corrects that.
Why water, and not just electricity
The electricity consumption of data centres is well documented. Water is far less so, even though it is often the sharpest point of local friction.
A data centre produces a considerable amount of heat that must be dissipated continuously. Several technologies exist, and the most energy-efficient rely on water evaporation. That water is not recycled: it goes into the atmosphere. Depending on the facility and the climate, volumes run into millions of litres per day.
In a region where the resource is already contested between agriculture, domestic use, and industry, the arrival of a new consumer of this size changes a local equation. The debate then stops being abstract: it is no longer a question of global carbon footprint, but the question of who will have water in April.
We touched on these mechanisms in our article on the environmental cost of AI. This dossier shows their concrete political translation.
The opposition also concerns wildlife, and this is an aspect rarely mentioned. A data centre is not just a building: it is a significant land footprint, new power lines, access roads, and continuous activity with its lighting and noise. On a territory hosting sensitive species, the effect goes well beyond the built surface. This is the kind of argument that, in European law as in several jurisdictions, can delay or block a project regardless of its economic merit.
What this dossier reveals
Three lessons emerge, and they go beyond the Indian case.
AI's timeline meets that of the real world. A model deploys in a few weeks. A public inquiry, an appeal, an environmental procedure take months or years. This difference in tempo is structural, and it explains why industry players seek jurisdictions where they can build fast. Which raises another question: the choice of territories least equipped to say no.
Global South countries are not just labour suppliers. We described on Monday the annotation work carried out in low-income countries. This dossier shows another side: these territories also host the physical infrastructure, with its nuisances. The economic benefit is real, the jobs created exist, but the distribution between what stays local and what goes elsewhere deserves to be examined.
Local opposition becomes a financial risk factor. For an investor, a project blocked for two years costs more than a slightly less optimal site chosen elsewhere. This reality is already pushing major players to factor in social acceptability very early on, which is probably the only mechanism that actually works.
The honest debate
It would be easy to conclude by siding with one camp. That would be dishonest, because both positions are defensible.
On the project's side: investments of this scale create jobs, tax revenue, and digital infrastructure a country may need to avoid depending entirely on foreign servers. That is a sovereignty argument we have ourselves developed regarding open models.
On the opponents' side: a water resource is not expandable, the employment benefits of a data centre are real but modest relative to the investment, and the nuisances are concentrated on those living nearby while the benefits are spread elsewhere.
What is certain is that this type of conflict will multiply. AI's infrastructure needs are such that they will seek land everywhere, and everywhere they will find people who already lived there. The question of who arbitrates, and by what criteria, is probably one of the most concrete political debates this technology will impose in the years to come.