The Possibilities of DPI & AI
In 2023, Nandan Nilekani introduced the phrase “DPI to the power of AI” to describe the next phase in India’s digital transformation, where digital public infrastructure and artificial intelligence converge to unlock new forms of public value and amplify human potential.
Since then, AI adoption has grown rapidly. Both the technology itself and the ways people use it have evolved across sectors. From improving access to services and information to enabling new models of inclusion and productivity, AI is now demonstrating tangible societal value. Yet this acceleration also brings new challenges: how to ensure safety, fairness, and trust as adoption expands.
Almost every technology creator, funder, governments, and users agree that real benefits of any technology happen when there are benefits on the ground and to the people of this planet. Impact happens only when people use, discover issues, challenges, and overcome them to emerge with tangible value. DPIs have seen this journey and the same principles of DPI can be applied to AI with use cases as the anchor.
Our understanding of this convergence comes from lived experiences, from observing how AI adoption takes shape across different contexts and sectors. The approach outlined here builds on those experiences and the lessons they offer.
This blog brings together what we have learned so far. As AI use cases multiply, so do the possibilities. The following sections explore how AI and DPI not only strengthen each other, but how they can continue to do so safely, inclusively, and at scale.
Every DPI Layer Generates Data That Can Power New AI Systems
Every digital public infrastructure layer creates structured, high-quality data trails. Whether it's payments, identity verification, or logistics coordination, these systems leave behind the raw material that powers AI. When used responsibly and with consent, this data can unlock powerful new capabilities: better targeting, faster detection, more adaptive service delivery.
Consider UPI: Millions of transactions across small merchants and individuals create a real-time pulse of economic activity. AI models built on aggregated and anonymized versions of this data could predict credit risk, detect fraud, or forecast local economic trends. The data exists. The infrastructure exists. What remains is building the analytical layer that transforms transactional records into economic intelligence.
Aadhaar-enabled verification has already accelerated onboarding and reduced fraud in welfare and banking. AI models can now build on these verifiable trails to flag anomalies in real time. Duplicate claims in subsidy transfers. Suspicious patterns in benefit distribution. These aren't hypothetical use cases. They're extensions of systems already in operation, made sharper by pattern recognition at scale.
The same principle applies in health, insurance, and social protection. Structured data streams from DPI systems can train AI models that anticipate public needs. A state-level health exchange, for instance, could power early detection of claim fraud or identify population-level disease risks before they become crises.
Each DPI layer becomes a training ground, producing safe, high-quality data that helps AI learn from real-world use. This is where AI for public value begins: grounded in actual digital trails, accountable to the people those trails represent.
DPI Thinking Will Shape How AI Systems Are Designed and Governed
The most transformative possibility lies in applying the principles of Digital Public Infrastructure to artificial intelligence itself. DPI is not a single technology but a way of designing systems that are open, interoperable, and inclusive. When these principles are applied to AI, they change how intelligent systems are built and governed. The focus shifts from isolated products to shared digital infrastructure that is trustworthy, adaptable, and designed for public purpose.
This convergence between AI and DPI is unfolding in three main ways.
1. New AI-DPIs are emerging within sectors
These are vertical applications of AI that strengthen existing digital ecosystems inside a sector and make them easier to use, safer, and more inclusive.
Agriculture is a leading example. Platforms like OpenAgriNet connect data, advisories, and AI tools across institutions so farmers can access localized, reliable insights. MahaVISTAAR-AI in Maharashtra builds on this foundation: an AI-powered companion developed by the Department of Agriculture with partners such as EkStep, COSS, and OpenAgriNet. Using generative AI, it brings together weather data, soil health records, market prices, and government schemes to deliver timely, personalized guidance in local languages. The result is an agricultural DPI that is intelligent, multilingual, and trusted.
Health is a parallel example. Platforms like the Ayushman Bharat Digital Mission (ABDM) connect health records across institutions through ABHA IDs and open standards, enabling interoperable, longitudinal digital health records. Building on this foundation, the National TB Elimination Programme (NTEP) has deployed AI chest-X-ray (computer-aided detection, or CAD) tools in public-health workflows - aligning with WHO guidance that recommends CAD for TB screening. Evidence from India shows such tools can support screening and triage in real-world settings (e.g., qXR deployments and evaluations).
Together with Nikshay, NTEP’s digital patient-management system, these components let care teams tie symptoms, investigations, prescriptions, and follow-ups into timely, reliable guidance for patients and providers - creating a health DPI that is intelligent and trusted.
2. AI horizontals enabling across sectors
These are public-good layers that enable many use cases at once. They include language and voice technologies, AI-ready data, and AI safety and evaluation frameworks - each operating as an AI horizontal. AI needs a shared digital spine to scale responsibly. Just as open rails like UPI unlocked digital payments, AI now requires shared rails that make language and voice technologies, data, and safety infrastructure accessible as public goods.
Language as DPI. India’s Bhashini initiative demonstrates how to operationalize this spine: multilingual models, datasets, and APIs as public goods that any developer or institution can use. In such a model, verified providers offer affordable compute to researchers and startups; open datasets are curated for local relevance and built with consent; and shared model repositories enable reproducible evaluation and safety checks across use cases.
AI-ready data as a horizontal. When datasets are standardized, consented, anonymized where necessary, and discoverable through common APIs, they become an AI-ready substrate that any model or application can safely build on. This horizontal cuts across sectors—health, climate, agriculture, education - so innovations aren’t trapped in vertical silos. AI Agents added to data which has been locked till now suddenly gives more meaning, wider access, and with trust and guardrails can unleash several opportunities.
An example - If Goods and Services Tax Network (GSTN) information is responsibly anonymized and analyzed through AI, it can surface actionable insights on trade flows, regional growth, and supply-chain bottlenecks. These insights, exposed through open, well-governed interfaces, can power everything from small-business credit models to district-level planning - illustrating how AI-ready data functions as a horizontal public good rather than a closed asset.
Similar infrastructure can extend to other sectors: open AI-ready data for health, climate, agriculture, and education. Each becomes a sector-specific DPI that democratizes access and accelerates innovation.
Safety and evaluation as horizontal rails. Shared safety and evaluation frameworks applied to several sectors, and use cases, can become DPIs ensuring systems remain fair, reliable, and transparent by default.
The goal is distribution rather than concentration: ensuring that the benefits of AI don’t belong to a few institutions or geographies but remain accessible through an equitable, transparent commons.
3. New AI-DPI Solutions unlock exponential value
This is where the most powerful outcomes appear. When a horizontal like language infrastructure meets a vertical like agriculture, entirely new solutions emerge.
Consider a service like Kisan e-Mitra: a conversational AI with local-language interfaces that helps farmers access schemes, training, and market information via voice or chat. With user consent, Aadhaar/eKYC can streamline eligibility checks and pre-fill applications; DBT can route approved benefits directly to the farmer’s bank account; and the assistant can confirm submission and payment status back to the farmer in their own language.
In practice, this brings together linguistic DPIs, agricultural DPIs, and DBT rails, so disconnected services behave like a single, responsive system - minimizing paperwork, reducing errors, and making entitlements easier to claim, understand, and track.
Together, these three pathways show how AI and DPI can evolve side by side. Verticals make AI useful within sectors, horizontals make it inclusive across them, and their convergence creates new forms of public infrastructure that are intelligent, trusted, and designed for everyone. This is where exponential value will be unlocked and the true power of AI will have real impact on the ground.
But the question remains, how can we unlock this exponential value and ensure more AIxDPIs are created?
The Path to AI-DPIs
The journey begins with use cases: real, grounded AI applications that solve meaningful problems. A use case refers to a real-world application of AI that delivers measurable societal value. It is not a pilot, prototype, or research experiment, it is a repeatable model that connects a defined user need to a practical solution.
DPI Thinking enabled exponential transformation not just through technology but through a balanced amalgamation of ideas to create a unified way for people. In AI, use cases will be the anchor, to drive us towards that north star of a unified approach.
Nothing proves the benefit of AI more than an actual use case moving from proof of concept to population scale. More AI-DPIs will be discovered, the possibilities are endless: AI agents conversing with citizens, after being authenticated through existing DPIs, serving them on intelligence and insights across multiple applications, private and public. The magic has just begun.
The Way Forward
The convergence of AI and DPI marks a new phase in digital development. This phase will be shaped as much by trust as by technology. Building systems that people and institutions can rely on will matter as much as building powerful algorithms. AI and DPI are co-evolving ecosystems. AI makes DPI more intelligent and adaptive. DPI makes AI more trusted, inclusive, and scalable.
In the coming decade, AI will : strengthen existing DPIs, spawn new AI-enabled sectoral DPIs, and run on a shared digital AI spine that makes compute, data, and safety universally accessible. Existing rails, identity and payments will become smarter and more usable as AI adds retrieval, reasoning, and multilingual interaction. Alongside them, sectoral AI-DPIs will crystallize in agriculture, health, education, climate, and beyond - tying together AI-ready data, vetted knowledge, and local interfaces to deliver timely, personalized public value.
Powering both shifts is the spine: open standards, data flows, affordable compute, language and voice technologies and safety/evaluation frameworks baked in from the start. This architecture favors distribution over concentration: small enterprises and public institutions can build on common rails, citizens can act with confidence, and trust is reinforced by auditability and clear accountability.
If we commit to openness, AI won’t sit atop DPI as another siloed layer - it will become the living infrastructure that continuously improves these rails and enables new ones, ensuring innovation is inclusive by design and durable at population scale.
As Nandan Nilekani says, “AI adoption will not only improve public services but also enrich the science of AI itself, through feedback from hundreds of millions of users”. The task before us now is to explore how purpose, intent, and use cases can come together to build this next chapter of the future.
If you're working on use cases in this space or want to help build AI-DPIs in your sector, reach out: info@peopleplus.ai
____________
With additional inputs from Shankar Maruwada, Jagadish Babu, Alok Gupta, and Kameswararao.
Editor: Sonia Rebecca Menezes
