Watts, Water, and Waste: The Environmental Price of Artificial Intelligence and the Law’s Failure to Collect It

Artificial intelligence is often presented as something clean, invisible and almost weightless. We type a prompt, receive an answer and move on. What we rarely see is the physical infrastructure behind that simple interaction.

Every AI query depends on data centres filled with power-hungry processors, industrial cooling systems, backup generators and enormous quantities of hardware. Behind the digital experience lies a physical economy of electricity, water, minerals and electronic waste.

The question, therefore, is no longer simply how artificial intelligence should be regulated for privacy, safety or misinformation.

We also need to ask: who is responsible for the environmental cost of AI?

The Hidden Physical Cost of AI

The environmental footprint of AI can broadly be understood through three resources: energy, water and hardware.

AI systems require enormous amounts of electricity, particularly during the training of large models and the continuous inference that occurs every time users interact with them. The carbon footprint associated with AI systems has been projected at between 32.6 and 79.7 million tonnes of carbon dioxide in 2025.

The problem becomes particularly significant in India, where electricity generation continues to rely substantially on thermal sources. Every additional watt consumed by a data centre can therefore carry a significant carbon cost.

Then comes water.

Data centres need sophisticated cooling systems, and some facilities can consume more than a million litres of water every day. India is particularly vulnerable because it has a large population but a comparatively small share of the world's freshwater resources. At the same time, data-centre development is expanding in cities such as Mumbai, Navi Mumbai, Hyderabad, Chennai and Pune, where water availability is already an important concern.

The third problem is electronic waste.

AI hardware evolves rapidly. GPUs and other computing equipment can become obsolete as newer and more powerful generations arrive. The environmental impact therefore begins long before a data centre starts operating and continues long after its hardware is discarded. Mining, manufacturing, transportation, operation and disposal all form part of the environmental footprint of AI.

This creates an important legal problem.

Who pays for this environmental cost?

The Law Has Not Caught Up

India already has a substantial body of environmental law. Article 21 of the Constitution has been interpreted to include the right to pollution-free water and air. Article 48A directs the State to protect and improve the environment, while Article 51A(g) places a duty on citizens to protect the natural environment.

India also recognises important environmental principles such as the polluter pays principle, precautionary principle, sustainable development and public trust doctrine.

At first glance, this may suggest that India already has enough legal tools to address AI's environmental footprint.

But there is a problem.

Traditional environmental law was largely designed around identifiable pollutants, identifiable sources and identifiable instances of harm. AI-related environmental damage does not always fit that model.

A data centre may not discharge a conventional pollutant into a river. Instead, it may consume enormous quantities of electricity and water over years, contributing to cumulative environmental stress.

The harm is therefore distributed, continuous and often difficult to attribute to a single activity.

That makes AI different from the traditional industrial pollution for which much of environmental law was developed.

Can the Polluter Pays Principle Apply to AI?

The polluter pays principle provides an interesting starting point.

Indian courts have repeatedly recognised that those responsible for environmental degradation should bear the cost of addressing it. The Supreme Court has also developed the principle of absolute liability in cases involving inherently dangerous activities.

AI infrastructure does not necessarily resemble a traditional hazardous industry. A data centre does not usually create the kind of sudden catastrophic accident associated with chemical or industrial facilities.

Its danger is different.

It lies in scale and accumulation.

Large AI infrastructure operators control activities that consume substantial quantities of electricity, water and computing hardware while simultaneously benefiting economically from that activity.

This raises a legitimate question: should enterprises generating foreseeable and substantial environmental externalities be required to bear those costs even when the harm is gradual rather than catastrophic?

The answer may require existing environmental principles to evolve rather than simply be discarded.

Water May Be the Hardest Question

The environmental debate around AI often focuses on carbon emissions.

Water could prove even more politically and socially difficult.

The public may never see the water consumed by a data centre, but the consequences can be very real when private technological infrastructure competes with households, agriculture and ecosystems for limited freshwater resources.

This is where the public trust doctrine becomes relevant.

Indian environmental jurisprudence recognises that resources such as rivers, forests and other natural resources are held by the State in trust for the public. If a water-stressed region is required to allocate substantial freshwater resources to privately operated AI infrastructure, the State may eventually have to answer a deeper constitutional question:

How much of a public resource can be allocated to private technological infrastructure before public interest is compromised?

The National Green Tribunal Can Help — But It Cannot Solve Everything

The National Green Tribunal could become an important institution in this emerging debate.

Its broad jurisdiction over substantial environmental questions provides a possible route for challenging environmentally harmful data-centre practices. Litigation could potentially compel greater disclosure, impose operational conditions and establish principles for dealing with AI infrastructure.

But the Tribunal cannot create an entire regulatory architecture.

It cannot, by itself, establish standardised environmental metrics for the AI industry, create a comprehensive liability regime or permanently replace legislative action.

The NGT can be a bridge, but it cannot be the final destination.

The World Has Not Solved This Either

India is not alone in facing this problem.

The European Union has taken one of the world's most ambitious approaches to AI regulation through the AI Act. Yet its environmental provisions remain limited, relying substantially on standards, documentation and voluntary approaches rather than creating a comprehensive environmental liability regime.

The United States has considered legislation addressing AI's environmental impact, but the dedicated federal proposal discussed in the article did not progress beyond committee.

China provides a different model, with more prescriptive efficiency standards for data centres, including limits relating to power usage effectiveness and water consumption.

Singapore offers another interesting example. It temporarily imposed a moratorium on new data-centre construction because of resource constraints before reopening development under sustainability-oriented conditions.

The global lesson is clear:

No country has completely solved the environmental governance of AI.

That means India is not necessarily playing catch-up. It has an opportunity to build a framework before the problem becomes significantly larger.

What Should India Do?

A purpose-built framework for AI's environmental footprint should begin with transparency.

Large AI systems and data centres should be required to disclose measurable information about their environmental impact, including:

• Energy consumed during training and operation

• Water consumed for cooling

• Hardware turnover and electronic waste

• Lifecycle emissions

• Environmental impact of large-scale computational activity

The article proposes a mandatory AI Environmental Impact Assessment for large AI systems and data centres above specified computational thresholds. Such assessments would measure projected energy use, water consumption and hardware lifecycle impacts before deployment and compare them with actual consumption after deployment.

India could also consider creating a dedicated AI Environmental Division within the Ministry of Environment, Forest and Climate Change to coordinate environmental oversight of AI infrastructure.

Liability could be structured around measurable factors such as computational intensity, energy consumption, water consumption in stressed regions and hardware lifecycle.

Environmental responsibility should also extend across the supply chain rather than stopping at the doors of a data centre.

Innovation Should Not Mean Environmental Exemption

The argument is not that India should stop developing artificial intelligence.

That would be neither realistic nor desirable.

AI has enormous potential in healthcare, education, science, business and public services. The question is whether its environmental costs should simply be passed on to communities and ecosystems.

The principle of sustainable development offers a better approach.

Innovation and environmental protection do not have to be enemies. But technological progress should internalise the environmental costs that make that progress possible.

If AI companies benefit from consuming enormous quantities of electricity, water and hardware, the public should not be left to absorb the environmental consequences without transparency or accountability.

The Bigger Constitutional Question

Ultimately, this is not merely a question of technology policy.

It is a constitutional question.

If environmental degradation affects the right to life under Article 21, and if the State has a constitutional responsibility to protect the environment, then the environmental consequences of rapidly expanding AI infrastructure cannot remain outside the regulatory conversation.

The precautionary principle is equally important. Scientific uncertainty about the precise scale of future environmental damage should not become an excuse for doing nothing.

India has already developed many of the legal principles required to address environmental harm.

What it has not yet developed is a framework designed specifically for the environmental realities of artificial intelligence.

Conclusion

The most important misconception about artificial intelligence may be that it is weightless.

It is not.

Behind every model are data centres, electricity grids, cooling systems, water resources, minerals, manufacturing facilities and eventually mountains of discarded hardware.

The law now has to catch up with that physical reality.

India does not necessarily need to invent an entirely new philosophy of environmental protection. Its constitutional jurisprudence, environmental principles and institutions already provide a foundation.

What is required is adaptation.

The central challenge is therefore not whether India has the legal principles to regulate AI's environmental footprint.

It is whether India is willing to use them before the environmental cost of AI becomes irreversible.