India’s AI-Driven agent state needs a sustainable foundation

A new race is quietly reshaping the global technology landscape. It is no longer just about building the most powerful AI models or manufacturing the fastest chips. The next frontier is using AI to transform how Governments serve citizens. Intelligent systems can anticipate, learn, and meet citizens’ needs, and automate repetitive administrative processes to proactively provide public services rather than simply responding to requests.This is the promise of agentic AI. India enters this race with an advantage that few countries possess.

Agentic AI is not like traditional AI systems, which follow directives from humans; It is a smart system that understands its environment, can engage in reasoning in complex tasks, and can take action on behalf of a person. For instance, instead of informing a citizen that they qualify for a subsidy, an agent could verify eligibility, submit the application, monitor approvals, and notify the beneficiary when funds arrive. The citizen no longer navigates Government offices or multiple portals. The Government comes closer to the citizen.

This shift represents far more than technological progress. Governments have spent decades digitising services. Agentic AI offers the possibility of digitising governance itself.
India has been investing for over 10 years in setting up the digital foundation required for this. Aadhaar provides a trusted identity layer. UPI enables seamless payments. DigiLocker offers secure document access. ONDC demonstrates how open networks can connect multiple participants through common standards. Collectively, they have developed the rails on which intelligent AI agents can run. The United States has no nationwide digital identity platform. The eIDAS regime is still unevenly adopted in Europe. Unknowingly, India has built one of the most complete DPIs in the world.

Early deployments already demonstrate what this future looks like. In the last two years, the Pradhan Mantri Surya Ghar Muft Bijli Yojana has increased the number of rooftop solar systems from 7.3 lakh to almost 30 lakh households. The first one lakh installations required 118 days; today, the same number is completed in nine days. The Ministry of New and Renewable Energy is working on an interface through WhatsApp to register, monitor progress, and give the citizens a subsidy of up to `78,000 in one conversational platform. Chhattisgarh’s CHiPS platform connects departmental databases and scaled annual welfare applications from 30 lakh to nearly 70 lakh without proportionate administrative expansion. 

Its AI layer recognises common rejections and can warn applicants prior to submitting. These are not experimental pilots. They are functioning systems pointing toward an Agent State where entitlements find the eligible citizen rather than the other way around.
There is another side to this story that receives far less attention. Every Agentic AI system relies on a physical infrastructure. Operational servers, cooling systems, electricity, and data centres are required for digital interaction. The digital and physical infrastructure, therefore, works in tandem. 

According to PwC India, the country’s data centre value chain could generate an order book worth $280 billion by 2035. Installed capacity is likely to increase from 1.6 GW today to 13.8 GW in 10 years. Direct investment in core facilities alone may exceed $71 billion, while spending on chips, servers, and networking equipment could account for another $180 to $210 billion. 

However, for each digital capability that grows, so do the physical resource requirements. That’s the debate India now has to face. Modern AI infrastructure is rapidly moving beyond the limits of traditional cooling systems. Workloads approaching 150 kW, can be supported by digital cooling, while immersion cooling can exceed even that. These are not routine engineering upgrades. They represent an entirely different scale of energy and water consumption.

A single hyperscale data centre can consume as much water in a day as a small village. Over 60 per cent of the current data centre installed capacity is projected to be in water-stressed regions in India. The geography sharpens the problem further. Hyderabad, Bengaluru, and parts of Maharashtra have attracted major data centre investments because of connectivity and talent. These cities are already under stress and have shrinking groundwater reserves. Public opposition has emerged in Pune against the allocation of water in the new data centre projects. Unlike electricity, water is a local resource. Every litre diverted to industrial cooling competes with households, agriculture, and small businesses drawing from the same supply. India supports nearly 18 per cent of the world’s population but controls only about 4 per cent of global freshwater. As AI adoption accelerates, this imbalance will become harder to ignore.

Corporate commitments to water positivity are encouraging but insufficient as substitutes for public policy. Replenishing water in one basin does little to relieve stress in another. When it comes to voluntary sustainability reports, there is no uniform methodology, making it challenging to independently assess. Without standardised disclosures, policymakers cannot accurately compare water use across facilities or verify whether conservation targets are being met.

The strategic risks extend beyond water. PwC reporthighlights that India relies on imports of mechanical, electrical, and plumbing (MEP) equipment in data centres. Many components have long procurement cycles and remain vulnerable to geopolitical disruptions. 

A supply chain shock could slow AI infrastructure deployment precisely when demand is rising. Strengthening domestic manufacturing, therefore, becomes a critical part of India’s AI strategy. Production Linked Incentive scheme for critical data centre equipment should be seriously considered, and the manufacturing clusters in the Make in India initiative should be expanded. 
Location should become a central criterion in approving new AI infrastructure. Resource availability must carry the same weight as land prices, connectivity, or tax incentives. A mandatory reporting system of water and energy use by data centre operators would help create more transparency and make investments more precise.Public-private partnerships.