India’s Data-Centre Expansion Needs a Water-Justice Framework

In June 2026, Amazon announced that it had become “water positive” in India, returning 120% of the water used across its offices, fulfilment centres and data centres. The claim has the reassuring precision of an audited figure. Yet Amazon’s own methodology measures the India target at the country level.

It prioritises projects near operations and in areas of high water stress, but it also permits replenishment in other basins. A national surplus, therefore, does not necessarily tell a community living near a facility whether its own river, reservoir or aquifer will remain secure through a dry season. This distinction is familiar in environmental governance.

Loss or extraction in one place is often balanced against restoration elsewhere. But water is encountered through particular watersheds, seasons and social relations. It matters not only how many litres are returned, but where and when they are returned, in what form, and whether the people most affected have participated in deciding what counts as replenishment. “Water positive” may be a useful corporate indicator; on its own, it cannot establish water justice.

The issue is becoming more consequential as the sector expands. A 2026 report by the Council on Energy, Environment and Water and Systemiq counted 271 data centres in India in January, roughly a quarter of them in Mumbai. It cited market forecasts projecting India’s installed data-centre capacity to rise from about 1.5 gigawatts in 2025 to between 4.5 and 6.5 gigawatts by 2030. India nevertheless has no binding national data-centre policy.

A Union draft released in 2020 was not notified. Fifteen states have introduced policies or related frameworks, but explicit sustainability provisions appeared in only five. Even then, the distribution of costs among nearby communities receives limited attention.

Also read: The Great AI Gamble: Who Pays When Computation Becomes Cheaper?

Public discussion usually begins with promises of development and employment. Increasingly, these promises come with a new vocabulary: a city will become an “AI hub” or an “AI City”. The language recalls the earlier enthusiasm for “smart cities” when a technological label itself began to stand in for a much wider understanding of urban development. But these cities – Mumbai, Hyderabad, Chennai, Noida, Lucknow, and Visakhapatnam – are not empty spaces awaiting transformation.

Workers, fishers, farmers, small businesses, and public institutions already sustain their economies. The relevant question is how this new infrastructure alters those economies and how its benefits and burdens are distributed. This distributional question also concerns the terms on which land is acquired. For communities historically excluded from landownership, land is not merely a source of income but a basis of security, dignity and intergenerational belonging – none of which can be readily restored through a one-time monetary payment.

The widely varying employment figures invoked in these debates call for closer analysis. For the Visakhapatnam AI hub, one economic model projected 1.88 lakh direct and indirect jobs a year across construction, operations, engineering, information technology and supply chains.

Reporting on two project-clearance records, however, cited estimates of 575 and 650 permanent jobs. These figures measure different things, but public discussion seldom keeps the categories apart. Evidence from the United States cannot simply be transferred to India, but it nevertheless warrants caution.

Illustration: Rose Willis and Kathryn Conrad, betterimagesofai.org.

A Brookings study of nearly 1,500 facilities estimated roughly 100 to 200 additional jobs when a typical county received its first large data centre, without an increase in wages. A Virginia legislative audit found about 1,500 workers at peak construction but only around 50 full-time operational workers at a typical facility. Data centres create work, but credible claims must distinguish construction work, permanent posts, contract labour, local recruitment and modelled effects.

Publicly available information on project-level resource requirements is often fragmented and difficult to reconcile. In Thane, Maharashtra, a municipal official was reported as saying that 12 million litres a day of treated sewage water would be supplied to a proposed Amazon data centre. Amazon later stated that the development would use no water for cooling and that its approved supply of 0.01 million litres a day was only for staff needs.

The statements may refer to different arrangements or project stages, which is precisely why a single audited water balance should explain the technology, source, quantity and timing involved. Similar questions about water, land and wildlife have been raised in Visakhapatnam, where Google and the state government have offered assurances concerning audits, air cooling and drinking-water sources.

Although aspects of the project’s resource use have been addressed in legislative replies and public statements, they need to be brought together in a consolidated, publicly accessible and independently verifiable assessment.

Andhra Pradesh illustrates how investment commitments can move rapidly beyond the scale contemplated in an existing policy. The state’s Data Center Policy 4.0 (2024–29) set a target of adding a few hundred megawatts – up to 1 GW – of data-centre capacity during the policy period. In August 2026, the state’s IT minister told the Assembly that Andhra Pradesh had secured allocations for a proposed 6.5-GW data-centre ecosystem, a figure referring to projects in the pipeline rather than capacity already installed and operational.

The substantial difference between the policy’s stated target and the scale of subsequent allocations calls for a cumulative assessment of water and power requirements, as well as renewed public consultation. The same principle should apply in Maharashtra, Telangana and other states where data-centre projects are becoming geographically concentrated.

A further concern is whether the existing environmental-clearance framework is equipped to assess the specific impacts of data centres. In an August 2026 Rajya Sabha reply, the Union Environment Ministry stated that AI data centres do not, by themselves, require environmental clearance. They enter the process when they cross thresholds applicable to building-and-construction or township projects.

Yet their characteristic impacts arise from power demand, cooling technology, water sources, backup generation and the cumulative concentration of facilities. A clearance framework organised substantially around built-up area may therefore be too narrow to assess the distinctive and cumulative environmental demands of data centres.

Also read: The Broken Equation of ‘Atmanirbhar’ AI

Other jurisdictions offer partial lessons. Singapore paused new data-centre approvals between 2019 and 2022 and reopened them with tighter efficiency conditions. The European Union now requires larger facilities to report energy and environmental performance, while Germany has legislated on efficiency, renewable electricity and waste-heat reuse. In the United States, legislatures in 14 states had considered or were considering moratoria by June 2026. These measures show that governments can slow, condition or reassess expansion. They also remain heavily concerned with technical performance; the people living around the infrastructure can still disappear from policy.

India needs a national framework, but not a centralised command that ignores ecological and institutional differences. The Union should establish common definitions, minimum disclosure, cumulative-impact assessment, and rights to consultation and remedy. States should be able to impose stronger conditions according to river basins, aquifers, coasts, climate risk, biodiversity and grid capacity.

Municipalities, panchayats and, wherever legally applicable, Gram Sabhas need a meaningful role before land, incentives and utility commitments are finalised. The Thane Municipal Commissioner’s reported decision to forward residents’ objections to the state administration because the matter lay beyond the corporation’s authority indicates the present weakness of local participation.

Nor should approval be treated as the end of assessment. Operators should publish project-level figures for absolute water and electricity use, seasonal water sources, water usage effectiveness, power Usage effectiveness, backup-diesel use, noise and heat. Employment reporting should separately identify construction job-years, permanent posts, contractual work and local recruitment, alongside livelihoods lost or altered. Independent verification and a public registry would allow policy to respond to drought, technological change, expansion and non-compliance rather than relying on assurances made at the beginning of a project.

India will almost certainly build more data centres. It would be unhelpful to reduce the discussion to a choice between technological growth and public protest. The protests have brought into view questions that investment announcements leave aside: whose water is used, whose livelihoods may change, which institutions are answerable, and who participates before a decision is made. The task is to connect national digital ambitions with the ecological and social conditions of the places in which infrastructure is built. The adequacy of India’s policy will ultimately depend on whether it can hold these two scales together.

Asifa Zunaidha F teaches Political Science at GITAM University, Visakhapatnam. 

The Wire is now on WhatsApp. Follow our channel for sharp analysis and opinions on the latest developments.