India's UPI Moved 22 Billion Transactions in March. ONDC Moved Under 4 Million.

The number that doesn't make it into the press release

UPI cleared roughly 22 billion transactions in March 2026 and has expanded to nine countries. That figure gets repeated so often it barely registers as remarkable anymore. What gets repeated less often is what the rest of India's digital public infrastructure stack looks like next to it.

ONDC's retail network moved under 4 million transactions a month by early 2026, down from about 6.5 million in October 2024. The Unified Lending Interface, live since August 2024, had 64 lenders on board and had disbursed close to ₹27,000 crore across 600,000 loans by December 2025. The older Open Credit Enablement Network disbursed roughly ₹1,100 crore across about 50,000 loans in the first eight months of FY26, with November 2025 marking the first month in six years that monthly disbursal crossed ₹200 crore.

None of these numbers are failures in the sense of a product that never shipped. They're failures in the sense of a product that shipped, launched with the same government backing and press cycle as UPI, and then plateaued well short of the volume that would make it structurally important. The interesting question isn't whether ONDC, ULI or OCEN are 'working'. Narrowly, they are. It's why the same playbook that turned UPI into critical national infrastructure hasn't repeated itself, three separate times, despite comparable political will behind each launch.

What UPI actually got right, and it isn't the protocol

UPI is routinely described as a triumph of protocol design: an interoperable, bank-agnostic rail that let any app move money through any other app. That's true, but it's not what made adoption happen. Two other things did.

First, UPI was mandated at zero merchant discount rate. Banks and payment apps couldn't extract a transaction fee from merchants, which sounds like it should have killed the incentive to build on it. Instead it removed the only reason an incumbent would have fought adoption. There was no card-network-style toll for UPI to threaten, so there was no lobby against it, and no boardroom debate about whether to prioritise it.

Second, and more important, UPI launched into genuine white space. Most Indians in 2016 did not have a fast, cheap, digital way to move money peer to peer or to a small merchant. Cash and cards were the alternative, and UPI beat both on cost and convenience without taking revenue away from an entrenched player. It was additive. That combination, mandated free access plus a market with no defended incumbent, is unusually hard to reproduce, and it's the piece every subsequent DPI layer has been missing. Every rail that came after UPI has had to compete with a business model that already existed, run by a company that already had customers.

ONDC's problem is retail infrastructure, not the retail protocol

ONDC decouples discovery, ordering, payment and fulfilment into an open network, which is a genuinely elegant piece of protocol design. But a protocol that decouples fulfilment from discovery still needs someone to do the fulfilment. Amazon, Flipkart, Zomato and Swiggy spent years before ONDC existed building warehouses, delivery rider networks, return logistics and seller-rating systems. A seller who plugs into ONDC gets discoverability. They don't get any of that underlying infrastructure, because the protocol was never designed to provide it. It assumes network participants will supply it themselves, and in practice too few have.

The mobility vertical shows the gap clearly. ONDC-linked ride-hailing does roughly 11 million rides a month, against 10 to 12 million rides a day across Uber, Ola and Rapido combined, roughly a thirtieth of the scale. Retail has followed a similar path with worse optics: Flipkart announced it would bring food delivery onto ONDC in May 2024, then made essentially the same announcement again in February 2026, which is itself a signal that the first attempt didn't stick. Ola and Paytm both launched ONDC-based food delivery in 2024 and withdrew within a year, after service-quality problems surfaced faster than a loosely coupled network of independent restaurants and delivery partners could absorb them. In each case, the network could route an order. It couldn't guarantee the order arrived the way a customer expected, because no single party in the chain owned that outcome.

ULI and OCEN ask lenders for something UPI never asked banks for

UPI didn't cost banks anything they were already earning: debit rail margins were already thin, and UPI mostly won them transaction volume and customer stickiness instead. Lending-side DPI is a different proposition. OCEN and ULI ask lenders and fintechs to expose underwriting data and originate credit through a shared interface, which touches the part of the business that actually makes money: interest income and the cross-sell that comes from proprietary borrower data.

A number of fintechs built their economics around exactly that data advantage, in part because UPI's zero-margin model pushed them to look for revenue elsewhere. Lending was the obvious place to look. Standardising origination through OCEN or ULI reduces the informational edge those companies compete on, which gives them a direct commercial reason to integrate slowly, or only at the margins of their business, rather than route core volume through it. None of this shows up as public resistance. It shows up as integration that stalls at a pilot cohort of borrowers instead of scaling to a lender's full book, quarter after quarter, without anyone on record saying no.

The regulatory signal that's easy to miss

The RBI's proposed market-share cap on UPI apps, a limit on how much of total volume any single payment app can hold, has been floated and deferred repeatedly since around 2021, and is now pushed out to December 2026. On its own that reads as a minor regulatory delay. In context, it's a live example of how hard it is to change the rules for a rail after volume has already concentrated on it, and it matters for ONDC, ULI and OCEN because it shapes how much a company is willing to bet on rails that don't yet have that kind of settled status. A founder deciding how much engineering time to put behind an OCEN integration is, implicitly, also pricing in how the UPI cap saga eventually resolves.

Read across the row, and the pattern is consistent: the rail that grew fastest is the one where regulation removed the incentive to resist, and where the market it entered had nothing to lose. Every other rail is asking an existing, profitable business to route volume through infrastructure it doesn't control, on a timeline set by a regulator whose own rules for the first rail are still being renegotiated five years in.

What would actually move the needle

The fix isn't a better app or a clearer onboarding flow. ONDC's constraint is that sellers without their own fulfilment capacity need a shared logistics layer to compete with Amazon and Flipkart on delivery speed, and building that is a multi-year infrastructure problem, not a protocol one. A handful of network participants are already trying to build pooled logistics for ONDC sellers; whether that catches on will matter more to the network's trajectory than any change to the app itself.

ULI and OCEN's constraint is closer to a negotiation than an engineering task. Origination needs to become additive to a lender's revenue rather than a threat to the data moat that currently substitutes for it, which likely means the interfaces have to prove they bring in borrowers a lender wouldn't otherwise reach, not just standardise the ones it already has. That's a harder sell than 'plug in and your costs go down', which is roughly what UPI offered banks in 2016.

India has signed digital public infrastructure cooperation agreements with 24 countries as of early 2026, and the DPI-as-export narrative is likely to keep growing. It's worth exporting the part that's actually proven: a mandated, free, additive rail into genuine white space, rather than the assumption that any open protocol will replicate UPI's growth curve just because it shares an acronym family. The stack has one rail that won outright. The other three are still waiting on the specific economic conditions that made the first one possible, and no amount of protocol polish substitutes for getting those conditions right.