By Nandan Nilekani, Co-Founder and Non-executive Chairman, Infosys
Using a basic feature phone, a Marathi-speaking farmer can now consult a chatbot for advice on soil, seeds, and irrigation in his own language. In government schools across several states, an artificial intelligence (AI) tutor helps children learn to read in their mother tongue. The potential is clear. So is the challenge: how do you move from these promising examples to systems that work reliably, every day, across thousands of institutions with vastly different capacities?
India has answered this question before. Over two decades, we built shared digital rails through public-private collaboration that now reach more than a billion people. Aadhaar has authenticated identities over 164 billion times. The Unified Payments Interface (UPI) processes around 20 billion transactions every month. These systems succeeded because they created trusted, reusable building blocks. A small bank could offer digital payments because the UPI existed. A rural hospital could verify identity because Aadhaar worked. We didn’t ask every institution to reinvent the wheel. We gave them the road.
AI demands the same thinking, but the problem is structurally harder. The UPI moves money. The protocol is stable; the outcome is binary: the payment went through or it didn’t. AI systems give advice, make predictions, and generate content. They require ongoing evaluation and adaptation in ways traditional software does not. Procurement frameworks, audit processes, and accountability structures built for fixed-price software purchases assume you can specify requirements upfront and verify compliance once. AI adoption requires different institutional mechanisms. Every institution discovers this mismatch. The question is whether each must learn these lessons independently or whether we can build shared infrastructure that compresses the learning curve.
Citizens trust institutions rather than algorithms. When the monsoon is delayed, a farmer wants to know who stands behind a recommendation. If AI is to be adopted at population scale, it must carry the credibility of recognised institutions—agricultural universities, government departments, co-operatives—with traceable sources and advice that can be questioned or corrected. Safety, in this context, means technical reliability coupled with institutional accountability.
This is where diffusion infrastructure matters. It provides reusable components for recurring adoption challenges so that institutions can focus on their mission rather than reinventing implementation from scratch. Maharashtra’s MahaVISTAAR provides agricultural advice in Marathi for more than 15 million farmers, powered by Bhashini, the government’s AI language platform serving more than 300 million users. The PM-Kisan chatbot lets farmers check eligibility and payment status by voice in their own language. Over 500,000 farmers used it on launch day. Researchers from AI4Bharat at IIT Madras collected over 12,000 hours of speech and 783 million translated sentences across all 22 official languages, creating open datasets that give India’s AI ecosystem an unmatched linguistic base. Nobody writes headlines about procurement templates or evaluation frameworks. These determine whether AI moves from isolated pilots to systems improving millions of lives.
India’s earlier digital successes worked because services and information could move freely across open rails. AI needs to work the same way: allowing models to be swapped, data to be verified, and services to be seamlessly combined. Interoperability, not just sheer scale, converts isolated success stories into a national ecosystem.
This requires building the infrastructure that makes adoption possible. Standards that let institutions trust AI outputs. Benchmarks that measure what actually matters for real-world deployment. Governance frameworks that work across sectors. Procurement models that accommodate systems that improve continuously. AI systems that are frugal and affordable to implement at population scale. Countries that build these
adoption enablers will see AI spread broadly across institutions. Countries that don’t will find AI is something that happens to them, not something they shape.
More than 30 countries have adopted elements of India’s Digital Public Infrastructure. If India builds the infrastructure that makes AI adoption work at scale, it becomes a playbook for every developing country facing similar challenges. India has become one of the largest markets for the world’s leading AI firms and the place where the technology’s possibilities and limits are most acutely tested. I invite global companies, academics and innovators to test their systems here, at national scale, with safety and accountability built in.
Trust drives adoption, and adoption provides feedback that improves systems. That is the virtuous cycle India must build. Our greatest contribution will be proving that AI works when it solves real problems at scale. Not in controlled pilots, but in how a district hospital actually operates, how a teacher teaches, how a farmer gets timely advice. The places where AI is hardest to deploy are also where it matters most. If AI can work in India’s classrooms, clinics, and farms, it can work anywhere. AI will transform India. And India will help shape how AI transforms the world.