22 Billion UPI Payments a Month Move Through One Central Switch

Seven parties, and you only ever see one of them

The piece, published by the open-source data project Time Series of India, opens with the moment everyone in India recognizes: scan a QR code, watch a name pop up, tap a PIN pad, see a green tick. What the reader never sees is the choreography behind that tick. A single payment touches seven parties: the app on your phone (a "TPAP," or third-party application provider, like PhonePe or Google Pay), your bank's sponsor (the PSP that actually issues your UPI address), your bank itself, the NPCI switch — the one non-profit, government-linked hub that every payment in the country passes through — and then the mirror image on the payee's side: their bank, their sponsor, their app.

The article's account of what happens in those two or three seconds is precise about the ordering: your bank is debited first, and only once that debit is confirmed does NPCI instruct the payee's bank to credit them. "The money always leaves before it arrives, never the other way round," as the piece puts it. Your PIN itself never reaches the app that collects it: it's encrypted by a certified component on the phone before the sponsor bank ever signs the request. Address resolution runs through the suffix of your UPI ID, so @ybl routes to Yes Bank and @okaxis to Axis. When payer and payee share the same sponsor bank, the whole thing resolves internally as an "on-us" transaction, saving NPCI's resolution fee of roughly one paisa.

The failure rate that used to be the story, and now isn't

By the article's numbers, NPCI's switch carried 2,272 crore payments in June 2026 alone, up from a few million a month when UPI launched in 2016, and past the point in August 2022 where merchant payments overtook transfers between people. About one payment in eleven still gets declined, but the article's real point is where those declines now come from: technical failures at the infrastructure level have fallen from more than one in a hundred payments to fewer than one in four hundred. What's left is mostly business logic, an insufficient balance, a wrong PIN, a daily limit hit, not the plumbing. A payment that lands in limbo (an unconfirmed "deemed" status) gets auto-reconciled by NPCI within about ninety seconds, and if a reversal is still owed, the money is guaranteed back within a day for a transfer or a few days for a merchant payment, under a 2019 RBI penalty rule.

That reliability sits on top of a lopsided market. PhonePe and Google Pay together carry roughly four-fifths of every UPI payment, with Flipkart's super.money climbing from outside the top fifty apps into the top five in about a year. On the receiving side, Yes Bank takes in a share of incoming payments no other bank comes close to, one that has roughly doubled in two years, a concentration the article flags without dwelling on.

The thread's gut check: is 22 billion payments a month actually a lot?

The Hacker News discussion opened not with awe but with arithmetic. Working from an annual volume figure, an engineer who ran the switch's average load against Nasdaq's market-open feed put NPCI's throughput at roughly 700 queries per second, a number he didn't find alarming:

"…that still doesn't sound that bad - for comparison, a quick Google tells me Nasdaq TotalView ITCH feed peaks at 100k+ QPS at market open."

The scale claim held up; the novelty claim didn't survive as cleanly. In the same thread, one reader pointed out that China's Alipay had already solved this problem years earlier, popular by 2010 and ubiquitous by 2015, and asked what UPI does technically that a Chinese QR-code wallet hadn't already shipped a decade before. Nobody disputed the timeline; what followed was a different argument for why India's version still counts as a feat: not the QR code itself, but building it as neutral rail that any bank or app can plug into, rather than one company's walled wallet.

A government ledger, not peer-to-peer

The loudest dissent wasn't about performance at all. It was about what the system actually is. A critic in the thread rejected the framing that UPI is peer-to-peer in any meaningful sense:

"…it's a terrible system for privacy and autonomy. It has so many intermediaries, requires phone number and linked to person's identity. It should not be called a peer-to-peer payment system in anyway. It is controlled by the government rather than Visa or Mastercard. That's the only the difference."

Nobody rebutted the identity-linkage point directly. The closest thing to a counter came from how deeply the system has embedded itself in daily life regardless of the objection. A software engineer of 17 years who splits his time between Kolkata and a Himalayan village described a level of use that makes the privacy trade-off almost beside the point for most Indians now:

"On an average day in a city like Kolkata, me and my partner make up to 20-30 transactions. I live in a small Himalayan village most of the year and I still make roughly 6-8 transactions a day."

What the architecture piece leaves out

Not every reader was won over by the write-up itself. A skeptical reader argued the article stops exactly where the harder engineering starts:

"Nothing about settlements, payment card messaging flow (you see a simplified logic in the article), nothing about other payment rails."

He has a point the rest of the thread's enthusiasm skips past. The piece is a beautifully documented map of the message flow between apps, banks and NPCI, but it never opens up how money actually moves between banks afterward, or how that clearing compares to the ISO 8583 messaging underneath Visa and Mastercard elsewhere. UPI's front end is the part every Indian reader touches half a dozen times a day; its settlement layer is still the part nobody outside NPCI has written up in public.