Before the first server hums: India needs an impact test for the AI age

In Visakhapatnam, India’s digital future is beginning to acquire a distinctly physical form. A proposed artificial intelligence (AI) and data-centre hub promises billions of dollars in investment, enormous computing capacity, new infrastructure and a stronger position for India in the global AI economy. Yet beneath the vocabulary of clouds, models and compute lies a more terrestrial set of questions. How much electricity will such infrastructure require? How much water will it consume? What will it mean for land, local communities and public resources? And, crucially, were these consequences adequately assessed before the project gathered momentum?

That last question deserves far greater policy attention.

India routinely evaluates the implications of highways, ports, mines and power plants before they are built. Large AI infrastructure is increasingly acquiring comparable characteristics. High capital intensity, concentrated electricity demand, substantial cooling requirements and long-lived consequences for local resources. Yet data centres continue to be treated primarily as digital assets.


That distinction is becoming difficult to sustain.

In an earlier article, “When the Cloud Gets Thirsty: Should India Make AI Pay Back Its Water Debt?”, we examined the growing water footprint of India’s AI and data-centre ecosystem and argued for greater efficiency, replenishment and carefully designed water-credit mechanisms. The present argument moves one stage upstream. Rather than merely compensating for externalities after they arise, India must begin assessing their environmental, social and economic consequences before large AI infrastructure is approved, incentivised or built.

India therefore needs a robust ex-ante Environmental, Social and Economic Impact Assessment (ESEIA) framework for major AI and data-centre projects. This is not an argument for technological caution. It is an argument for greater economic and institutional discipline in deciding where, how and at what resource cost infrastructure should be developed.

The economic case for AI investment is formidable. Google’s proposed Visakhapatnam AI hub represents an announced investment of roughly $15 billion between 2026 and 2030. India’s data-centre capacity has expanded from approximately 375 MW in 2020 to 1,500 MW in 2025, while more than 38,000 GPUs have been onboarded under the national AI compute initiative. Government estimates suggest that AI could contribute as much as $1.7 trillion to India’s economy by 2035.

But scale changes the nature of responsibility.

A data centre may resemble an office complex from the outside while behaving more like an industrial installation in its demand for electricity, cooling and supporting infrastructure. The International Energy Agency estimates that data centres consumed about 415 terawatt-hours of electricity in 2024, roughly 1.5 per cent of global electricity demand, and projects consumption to rise to around 945 TWh by 2030. At such a trajectory, assessing impacts after construction is rather like inspecting a building’s foundations after the upper floors have already been completed.

That is why the emphasis must be ex ante.

Also read | India’s AI data-centre push could drive 5% of global chip demand by 2030

Conventional sustainability reporting tells us what happened. Impact assessment should help determine what ought to happen before capital, land and infrastructure become irreversibly committed.

The environmental component should examine projected electricity requirements, peak-load implications, energy sources, freshwater withdrawal, basin-level water stress, cooling technologies, recycled-water availability, land-use change, biodiversity, waste generation and climate resilience. Carbon alone is an inadequate proxy for environmental sustainability.

But environmental arithmetic tells only part of the story.

The social assessment must identify who bears the costs and who receives the benefits. A facility may consume only a small fraction of a state’s aggregate water availability yet place disproportionate pressure on a particular watershed shared by households, farmers and urban users. Aggregate statistics can therefore obscure highly localised consequences.

Employment quality, livelihood effects, land acquisition, community access to water and electricity, public-health implications, infrastructure pressures and stakeholder consultation should all form part of the assessment. The relevant question is not merely, “How much resource will the project consume?” It is, “Whose resource, in which location, during which season, and at whose opportunity cost?”

The economic pillar is equally important. Impact assessment must not become a euphemism for environmental restriction. A credible framework should quantify investment, direct and indirect employment, local procurement, tax revenues, digital connectivity, productivity gains, innovation spillovers and strategic access to domestic compute.

Yet every balance sheet has two sides.

States may provide subsidised land, fiscal incentives, transmission infrastructure, water connections and other forms of public support to attract large technology investments. These may be legitimate development choices, but they should be evaluated against the value ultimately created. The question is not simply how large the investment is, but what its net development value becomes once subsidies, infrastructure costs, environmental externalities and social consequences are entered on the same ledger.

ESEIA should therefore function not as a compliance ritual, but as a decision-making instrument.

The framework should be proportionate and risk-based. Small facilities should undergo simplified screening. Larger projects should face progressively stronger requirements based on electricity load, freshwater consumption, land footprint, location in water-stressed regions and cumulative data-centre concentration. Low-risk projects could receive expedited approval; medium-risk projects could proceed with clear mitigation obligations; high-impact projects should require independent assessment and alternatives analysis before substantial public incentives or final clearance are granted.

Alternatives matter. A cheaper site in a water-stressed basin with a constrained grid may appear attractive in a financial model. A slightly more expensive location offering treated wastewater, renewable-energy access and stronger transmission infrastructure may prove substantially cheaper to society over the project’s lifetime.

Impact assessment, in this sense, is not the brake pedal. It is the navigation system.

India must also assess cumulative impacts. Five individually manageable data centres can collectively overwhelm the same watershed or electricity network. Assessment should therefore extend beyond individual buildings to data-centre clusters, power systems and water basins.

India does not need less AI infrastructure. It needs better-sited, better-designed and better-assessed infrastructure.

Sustainable development is not achieved by placing growth and ecology on opposite sides of a scale. It is achieved by making trade-offs visible before they become irreversible.

Before the first server hums, before the first litre is drawn and before billions are committed, India should insist on seeing the complete balance sheet. Environmental, social and economic.

Because the most expensive impact is often the one discovered too late.


Pradeep S. Mehta is the secretary-general of CUTS International. Sohom Banerjee is associated with Jaipuria Institute of Management, Noida.