NPCI's CEO Lays Out the Roadmap to One Billion Daily UPI Transactions. AI Does the Heavy Lifting.

With UPI having grown to over 750 million daily transactions, Dilip Asbe has filled in the operational picture of what getting to one billion actually requires.

The AI Thesis, Specifically

Asbe laid out four distinct jobs for AI in UPI's next phase: acquiring new users, detecting fraud and money mules, extending credit to merchants and individuals who have digital transaction histories, and building voice and multilingual onboarding tools. He made these remarks to TechCrunch during Mumbai Tech Week 2026.

"AI will be used very effectively when we look at the next wave of UPI," Asbe said. "We must use AI effectively to protect our current citizens, to find fraud, and to find mules. AI must also be used to provide credit to all the users and merchants who have digital footprints."

Asbe is not describing AI as an enhancement to a working system. He is describing it as the mechanism that drives the next half a billion users — people who are harder to reach, less formally documented, and who transact in languages other than English. He envisions NPCI, India's central bank, and the government working together to achieve this.

Voice: Promising, But Not There Yet

NPCI launched a voice assistant-based interactive payment system in 2023. Asbe acknowledges adoption has not taken off. His explanation is direct: voice models need to be more accurate before they can be trusted in a payments context.

This is an honest admission from a regulator-operator who could easily have oversold the technology. The infrastructure exists. The use case hasn't crystallized yet.

Small Language Models Over Large Ones

Asbe's most concrete strategic bet is on domain-specific small language models rather than general-purpose large ones. The argument is straightforward: India's payments ecosystem has one of the richest transaction data sets in the world, and that data is a competitive asset.

"We believe that the models will differentiate from each other based on the data sets that are made available to them," he said. "I think there is a big opportunity for Indian companies — the banks, FinTechs, and the ecosystem — to create small language models which are sharp, specific, and as deterministic as possible."

NPCI already has one operational example. Its FIMI model, launched last year, handles user disputes, specifically canceling mandates and resolving payment issues. Asbe says FIMI is currently serving over one million users and scaling fast.

One million users against 750 million daily transactions is a small footprint. But it's real deployed infrastructure, not a demo.

Agentic Payments: Demo Stage, Not Deployed

NPCI showed demonstrations of agentic commerce, with AI agents executing payments on a user's behalf, with Razorpay last year. There has been no wider rollout. Asbe says the prerequisite is a regulatory framework that establishes user consent protocols and defines what happens when an AI agent makes a mistake.

The U.S. is further along here by private-sector metrics. Coinbase and Robinhood now allow AI agents to trade on behalf of users, and OpenAI lets users load personal account data into ChatGPT for financial advice, according to TechCrunch. India's approach is deliberately more cautious, with NPCI, the Reserve Bank of India, and the government expected to coordinate before any broad rollout.

Whether that caution is prudent risk management or bureaucratic drag is a legitimate question. The counterargument to India's measured pace is that the U.S. is running live experiments on real users with real money and accumulating the failure data that will shape the actual regulatory response. India may arrive at a better-designed framework, or it may arrive late to a market that has already moved.

The Market Concentration Problem

Asbe has consistently advocated for healthy competition among UPI apps. The data does not reflect it. Walmart-owned PhonePe and Google Pay together hold over 80% of UPI transaction market share, according to TechCrunch. NPCI's plan to cap an app's market share at 30% is set to take effect on December 31, 2026, unless it defers the deadline again.

Asbe noted that PhonePe and Google have poured millions into their apps to attain their market position, and that UPI apps have very low switching costs with most core features shared. He argued that if new apps find viable business models within the fintech ecosystem, their share will rise.

"I believe that there are multiple issues why we see this concentration risk exist, and one of the important reasons is the availability of a viable commercial model. The moment we see the commercial model being available to the ecosystem, I believe newer players will start investing very heavily," Asbe said.

Concentration in payment rails is a genuine systemic risk. If one dominant private app experiences an outage, fraud event, or policy dispute, it affects the majority of India's digital payment volume. That risk does not disappear because the underlying UPI infrastructure is state-operated.

In 2024, NPCI spun off its BHIM UPI app to make it more competitive, though its overall market share remains around 1%. Asbe said there is no particular target market share NPCI is eyeing for BHIM, but it wants to make it a sovereign and secure alternative to other apps.

The Open Question

NPCI's FIMI dispute-resolution model is the clearest test case for whether AI can operate reliably inside India's payments infrastructure at scale. It is currently handling over one million user interactions. If FIMI's accuracy and coverage expand materially, it becomes the proof of concept that accelerates NPCI's broader AI rollout. If it plateaus or produces high error rates in edge cases, Asbe's roadmap faces its first real credibility test.