NPCI wants AI to push UPI past a billion daily transactions. Dilip Asbe said on stage at Mumbai Tech Week that the National Payments Corporation of India, working with the central bank and government, expects artificial intelligence to be central as it tries to bring another half a billion people onto digital payments. NPCI already runs pilots, including a voice assistant and a small model called FIMI that has been used to resolve user disputes. The aim is to move from point solutions to a payments-native AI layer for onboarding, fraud detection and credit decisions as UPI volumes climb.

Backstage after his session at Mumbai Tech Week, Asbe turned concrete about what that ambition looks like in practice. NPCI runs multiple AI-led pilots today, he said, and intends to stitch those use-case agents into a foundational, sovereign layer that banks, fintechs and payments apps can call in real time.

From pilots to a payments-native AI layer

The organisation already operates two headline systems. NPCI launched a voice assistant aimed at simplifying common tasks across languages, and introduced a small language model called FIMI to resolve user disputes. FIMI handles tasks such as cancelling mandates and processing complaints, cutting manual casework and offering quicker, automated grievance resolution.

Asbe described FIMI and the voice work as the starting point rather than the end state. NPCI demonstrated agentic payments and commerce use cases with Razorpay last year, and on 17 February 2026 it announced a collaboration with NVIDIA to develop a sovereign, payments-native AI foundation built on NVIDIA’s open Nemotron models and accelerated computing platforms. Executives at NPCI framed the effort as an evolution from one-off helpers toward a low-latency, shared stack that preserves India’s data sovereignty and security requirements.

The technical plan the organisation is exploring is pragmatic. NPCI is looking at architectures such as Mixture of Experts to manage very high volumes while keeping latency low and to broaden multilingual coverage. The idea isn't to sell a closed product. Instead, NPCI intends to offer infrastructure primitives: standardised, sovereign models that a bank or fintech can call for fraud scoring, voice-enabled onboarding across multiple Indian languages, or deterministic credit decisions based on richer digital footprints.

Why the market thinks it will pay

Commercial logic is a key driver. Paytm founder and chief executive Vijay Shekhar Sharma told media that AI acts as a force multiplier for scale, efficiency and customer engagement, and several fintech executives have told industry outlets that AI-driven merchant demand and payments processing are growing rapidly.

Mastheads

Razorpay’s chief operating officer Rahul Kothari told reporters he expects AI-driven sectors to expand faster than other segments and that companies are preparing for substantially higher volumes.

That dynamic matters to NPCI because the operator’s role is both coordinator and platform provider. Asbe made clear that NPCI still wants healthy rivalry among UPI apps even while it builds shared infrastructure and models to serve the whole ecosystem. The trade-off is familiar: build common rails and risk complaints about centralisation, or leave every player to their own and forfeit the efficiency gains a shared stack can deliver.

For users, the promise is both operational and experiential. FIMI has already shortened complaint cycles and reduced manual handling. A broader foundational layer would let third parties rely on standardised services for multilingual onboarding, agentic commerce interactions and near real-time fraud detection. NPCI emphasises these services will be primitives for the ecosystem, not closed single-vendor offerings, so banks and fintechs can integrate them into customer journeys without rebuilding models from scratch.

Adoption isn't automatic. Asbe warned that voice interfaces still need higher accuracy before they can scale across India’s linguistic diversity.

He also stressed the need for regulatory guardrails and consent frameworks so agentic systems can be audited for instructions and for user permissions in the event of disputes. NPCI plans to prioritise trust, resilience and security as it scales the models underpinning these services.

Technically and commercially, NPCI’s roadmap is iterative. The organisation will continue to pilot and scale agents while working toward an underlying platform that meets sovereignty and latency requirements. That approach is calculated: prove value in narrow, high-impact workflows and then generalise the tooling into shared primitives for fraud, onboarding and credit decisioning.

For the payments ecosystem the central question isn't whether AI will play a role. It already does. The question is how fast the industry can move from demonstrations and pilots to a stable, auditable foundation that serves a billion daily UPI transactions without compromising security or competition.

Related Articles

One concrete number anchors the project: UPI already processes north of 750 million daily transactions. In February, NPCI announced a collaboration with NVIDIA to build a sovereign payments AI foundation, a step it says will help push volumes past a billion daily transactions.

This article was created with AI assistance.