India’s digital infrastructure has reached its next test

India’s most important AI problem may have nothing to do with AI. It may be that its institutions still do not speak the same digital language. That may sound like an administrative inconvenience but it’s becoming a technological constraint.

Over the past decade, India has built some of the world’s most ambitious digital public infrastructure (DPI). Aadhaar created a digital identity layer. UPI changed payments. DigiLocker made credentials portable. New systems are emerging for health, agriculture, commerce and education. The first task was to put services online. The next task is much more: making those systems work together.

And once AI begins making decisions across them, interoperability stops being a technical nicety. It becomes a national capability.

The newly released State of DPI in India 2026, from the Center for Digital Public Goods at IIM Bangalore, describes this transition clearly. India’s DPI is moving beyond simply implementing platforms toward connecting them across sectors. The report’s proposed next stage, sometimes described as DPI 3.0, involves adding an intelligence layer capable of anticipating needs and coordinating action across institutions.

This is usually presented as a story about what AI will allow India to do.
The more important question is what India’s institutions will allow AI to do.

Consider a farmer applying for a loan. An intelligent system might need to know who the farmer is, whether the farmer owns or leases the land, what crops are being grown, what the weather has been like, whether previous loans were repaid, whether the farmer has insurance and what prices are being received in local markets.

None of those facts necessarily live in the same place. Some may sit with a bank. Others with a state Government. Others in land records, agricultural 
databases, insurance systems or market platforms. An AI model can process all of this information in milliseconds.

But it cannot solve a more fundamental problem: What happens when the databases disagree?
That is the coming Indian AI problem.

We have spent years asking whether machines can become intelligent enough to understand the world. We are now approaching a different question: can the institutions feeding those machines agree on what the world looks like? India’s answer will matter enormously.

The ICRIER State of India’s Digital Economy 2026 report ranks India fifth among 71 countries on its digital-economy index and fourth on its AI index, behind the United States, China and Singapore. India’s scale is extraordinary: it has become one of the world’s largest markets for digital services and AI use.

But those numbers conceal a distinction that deserves much more attention. There is a difference between using intelligence and organising an economy around intelligence. India is already very good at the first.

The harder prize is the second. A country can have millions of AI users and still fail to capture the full economic value of AI if its institutions cannot exchange reliable information, establish common standards or give machines clear authority to act. This is why India’s next digital bottleneck may not be compute. It may be coordination.

The Government appears to recognise the problem. Its National Data Governance framework is designed around policies, standards, platforms and governance, with the stated aim of reducing data silos and improving interoperability across Government databases. State Data Governance Committees are also being established across the country. That is more important than it sounds. For years, India’s digital question was: Can we digitise this service?

The question is now becoming: Can this service safely communicate with another one? Those are fundamentally different problems.

Digitising a department can be done within the department. Interoperability requires agreement between departments. And AI makes that agreement much more consequential.
Imagine an AI system that can identify a citizen, check eligibility for a Government programme, examine financial information, verify a credential and initiate a transaction. Technically, none of these tasks is particularly mysterious.

The difficult questions are institutional. Who gave the AI permission? Which database is authoritative? How recent must the information be? Can the system explain why it trusted one record over another? Who is responsible when the information is wrong? Can the citizen challenge the decision? And what happens when an AI system can move from reading information to changing the world? That last transition is crucial. A chatbot that gives incorrect information is one problem. An AI system that changes a land record, approves a loan, alters an insurance claim or initiates a Government payment is another.

The moment software acquires the ability to act across institutional boundaries, permissions become as important as intelligence. This is where the Indian experiment becomes unusually important. India has already built digital rails at population scale. The question is whether it can build a trustworthy layer of intelligence on top of them.
But interoperability should not mean creating one giant database. That would be the wrong lesson.

The objective should be for institutions to cooperate without surrendering all their boundaries.

A hospital should not need to hand an entire medical record to an unrelated institution simply because an AI system wants one piece of information. A bank should not need unrestricted access to a citizen’s data to verify one fact. A Government department should not become the owner of information merely because it can technically access it.
India therefore needs something more sophisticated than data sharing.

It needs controlled interoperability. The distinction matters because data is not simply a collection of facts.

Every database contains assumptions. A land registry has a definition of ownership. A bank has a definition of creditworthiness. A Government agency has a definition of eligibility. A hospital has a definition of a medical record.
When those definitions differ, connecting databases does not eliminate the disagreement.

It merely passes the disagreement to the machine. This is where a basic lesson from physics becomes surprisingly relevant.
Different instruments can measure the same physical system only if we understand their units, calibration and uncertainty. Connecting two instruments does not make their readings automatically comparable.

Digital systems have a similar problem.
A database is not reality.It is a measurement of reality, produced according to particular rules.If two systems measure the same person, company, property or transaction differently, an AI model does not possess a magical ability to discover which one is true.It must choose.And the moment it chooses, governance has entered the algorithm.This is why India’s AI debate needs to move beyond models.

India should certainly build computing capacity. It should develop talent. It should support startups. It should invest in research.
But the country should also build something less visible: common definitions, reliable registries, data provenance, machine-readable rules, interoperable standards and clear permissions.
The IIM Bangalore report makes precisely this broader point about the evolution of digital public infrastructure. The value of DPI increasingly comes not from individual platforms but from what happens when shared infrastructure allows new ecosystems, services and third-party innovation to emerge around it.

That suggests a different way of thinking about India’s competitive advantage.
India does not necessarily need to reproduce the enormous capital expenditure of the United States or China to become an AI power.
It has another asset: scale combined with digital public infrastructure.

If India can make its institutional systems interoperable, an Indian AI company could potentially build on infrastructure that already reaches hundreds of millions of people.
That is potentially more powerful than simply producing another chatbot.

But it requires discipline. The Government should resist measuring progress by the number of AI models announced or the number of AI applications launched. Those are visible metrics. The harder metrics are invisible:

Can two Government databases reconcile their records? Can a citizen see where an AI decision came from? Can a machine verify the provenance of a piece of information? Can an institution revoke an AI’s permission? Can an AI system act across departments without creating a new concentration of power?

Those are the questions that will determine whether India’s digital infrastructure becomes a genuine economic advantage or simply an impressive collection of platforms.
The first phase of India’s digital transformation was about access. The second was about transactions. The next phase will be about coordination. And AI will accelerate all three.
Now comes the harder task: making the rails connect, making the information trustworthy and deciding where the machines are allowed to go.

The country that solves that problem will not merely have more AI. It will have something more valuable: an economy in which intelligence can actually move.