100% of UPI transactions, all legs
-
01
India runs 38.7 billion UPI transactions a quarter.
More real-time payments than anywhere on earth, and still compounding. 2026Q2, PhonePe's own share of them.
-
02
63.9% of them are merchant payments.
Someone buying something. This is the leg a merchant discount rate could ever be charged on, and it is the busy one.
-
03
But merchant payments are only 23.0% of the rupees.
Volume and value have separated. Person-to-person transfers are 30.8% of transactions but 71.2% of the money: fewer, far larger. The busy half is the small-ticket half, averaging ₹425.
-
04
That busy half earns ₹0.
Under zero-MDR the merchant leg is free to accept. India built the rails, then priced the only monetisable side of them at nothing. Who captures the value, and is there a business here?
The answer
India's instant rail moved the transactions and not the money, and the price it is not permitted to charge is larger than the industry that runs it. The merchant leg carries 64% of transactions and 23% of the value, earns nothing, and would be worth Rs 15,450 crore a year at the rate NPCI itself permits. That is not a gap any operator can close, so the investable business is the relationship the payment creates, not the payment.
- 1 64% / 23% The busy leg is the cheap leg, and that is structural rather than Indian. Brazil's Pix reaches the same shape, 45% of transactions and 11% of value, under a central bank that never had a discount rate to remove. Two regulators, one design, one outcome.
- 2 Rs 15,450 crore What the statute forgoes is larger than what the industry earns. Priced at the 30bps NPCI permits, the merchant leg would be worth 1.9 times One97 (Paytm)'s entire revenue. The state replaces 24% of it with an appropriation that has been thinning by arithmetic, from 8.68bps to 7.05bps.
- 3 6.8% vs 39.8% Distribution and infrastructure earn; the transaction does not. Filed accounts, same year: One97 (Paytm) converts 6.8% of the largest revenue in the panel, Angel One 21.2% selling distribution, and CDSL, the one piece of plumbing permitted to charge a toll, 39.8%.
The answer, first
India's payments rails have separated volume from value and priced the busy half at zero to whoever accepts the payment. Merchant payments are 64% of transactions but 23% of the rupees, and merchant is the only leg an MDR could touch. Zero is not free: the state pays acquiring banks 7.0 basis points to hold it there, and that rate is falling. The market is concentrated at the same time as it is unmonetised, which is why the share cap cannot bind, and why the investable business here is distribution, not transactions.
- 64% / 23% Merchant payments carry the transactions; P2P carries the rupees. The busy leg earns nothing.
- 78% Two apps intermediate four in five transactions, and both breach the 30% cap due this December.
- 4.3bn Transactions a month that would have to change app for the cap to bind. It will not bind.
- 7.0bps What zero MDR actually costs, paid by the taxpayer to acquiring banks, and falling every year without a decision.
- 114bps The private-bank margin advantage, and it splits almost evenly between pricing and funding.
- ₹0 50.7M merchants, ₹2,06,685 of GMV each per quarter, no payment revenue.
- 4.2× Wrappers per real strategy on India’s fund shelf. Most of the product count is packaging.
A Payments landscape and the monetisation gap
Exhibit 1 · Volume against value
Merchant payments carry 64% of transactions but 23% of the rupees
P2P is the mirror image: 31% of transactions, 71% of value. Volume and value have separated.
PhonePe Pulse, 2026Q2. PhonePe's own transactions, not all of UPI. Accessed 2026-08-20. Values as published at the 2026-08-21 pull; PhonePe has since withdrawn them from its open data. Download this exhibit's data (CSV)
Exhibit 2 · The revenue gap
Every rate other than the one in force would produce real revenue
Quarterly revenue the merchant leg would generate at each MDR, on 10.48 lakh crore of merchant GMV. The rate in force is 0bps.
PhonePe Pulse, 2026Q2. PhonePe's own transactions, not all of UPI. Accessed 2026-08-20. Values as published at the 2026-08-21 pull; PhonePe has since withdrawn them from its open data. Download this exhibit's data (CSV)
Exhibit 3 · Ten years of volume
Volume compounds without pause, and the series shows exactly where its data came from
The open mirror of NPCI's series stops in August 2023. Later months were transcribed from NPCI's own table, and are coloured differently rather than silently merged.
NPCI monthly product statistics: open CKAN mirror to 2023-08, then transcribed from npci.org.in on 2026-08-20. Four months NPCI does not publish in its ten-row table are left as gaps, not interpolated. Download this exhibit's data (CSV)
Exhibit 4 · Growth bridge
Retail contributed 64% of all volume growth between 2018Q1 and 2026Q2.
Where the extra transactions came from. The merchant leg is the one a discount rate could be charged on.
PhonePe Pulse, 2026Q2. PhonePe's own transactions, not all of UPI. Accessed 2026-08-20. Values as published at the 2026-08-21 pull; PhonePe has since withdrawn them from its open data. Download this exhibit's data (CSV)
B What the rail costs, and who pays
Exhibit 5 · The subsidy
Zero MDR costs 7.0 basis points, paid by the taxpayer, and the rate is falling
Bars are the incentive the government actually paid acquiring banks. The line is that payment over ALL national merchant value, so both sides are measured. The final year is an approved outlay, not a payout, and carries no rate: its appropriation covers twelve months while the published value base covers ten.
Press Information Bureau, Ministry of Finance, 24 March 2025. Transcribed in a browser: PIB serves HTTP 403 to scripted requests and publishes these figures as chart images. Computed by analysis/08_rail_cost.py. Download this exhibit's data (CSV)
Why this changes the question: the merchant leg is not unpriced, it is priced at 7.0bps and billed to the exchequer. The appropriation is set in rupees while the base it covers compounds in percent, so the effective rate falls every year without anyone deciding that it should. Read the memo and what it cannot know →
C The prize, in rupees
Exhibit 6 · The pool that does not exist
Priced at the 30bps NPCI itself permits, the merchant leg would be worth 1.9x the entire revenue of India's largest listed payments company
FY2023-24 national merchant value of Rs 51.5 lakh crore, priced at each rate. Dashed bars are money that does not exist; solid bars exist. Not a forecast: an MDR actually introduced would change behaviour rather than simply multiply out.
PIB national UPI value split, and filed accounts via Yahoo Finance. Every bar is a published rate times a published base. Computed by analysis/11_value_pool.py. Download this exhibit's data (CSV)
Exhibit 7 · The ask nobody printed
Holding the subsidy rate steady would cost between 2.7k and 3.5k crore more than was appropriated
Section B recommends indexing the outlay to merchant volume and never sizes it. This is that number, as a band rather than a figure: the published merchant base for FY2024-25 covers 10 months, so the low end treats it as a year and the high end scales it pro rata.
PIB, Ministry of Finance. A range because a ten-month base annualised is an assumption, not a measurement, which is rule 12. Download this exhibit's data (CSV)
Why this reframes everything above it: the merchant leg is not a small business earning nothing. It is a large business that is not permitted to charge, and the state buys back 24% of the price it forgoes. Every recommendation in this report should be sized against what operators actually earn, not against this pool. Read the memo and what a pool is not →
D The rail that is allowed to charge
Exhibit 8 · The control group
Cards hold 2.9% of the transactions but 7.6% of the value: a price does not kill a rail, it segments it
2026-07, same country, same merchants, same regulator, and only cards may charge a merchant discount rate. Average ticket ₹3,482 on cards against ₹1,263 on UPI, 2.76x. Cash withdrawal excluded: it is not a merchant payment.
RBI Bank-wise ATM/PoS/Card Statistics, national Total row, transcribed by hand because RBI document links return an interstitial to a script. That Total reconciled against its 63 bank rows at 0.00%. Computed by analysis/13_cards_control.py. Download this exhibit's data (CSV)
Exhibit 9 · Where a merchant can be paid
There are 80 UPI QR codes for every card terminal in the country
Millions of acceptance points at month end. A terminal is hardware that costs money and carries a fee on every swipe; a QR code is a printed sticker that carries neither. The axis is linear on purpose.
RBI Bank-wise ATM/PoS/Card Statistics, national Total row. Download this exhibit's data (CSV)
Exhibit 10 · Standing, not incidental
The split holds: cards move within 0.23 of a percentage point of value across every month transcribed, while QR acceptance pulls further away
2026-05 to 2026-07, three separate workbooks. The red line is the priced rail share of value between the two rails; the teal line is UPI QR codes per card terminal, which went from 68 to 80 in three months. Three points show the split is standing; they do not make a trend.
RBI Bank-wise ATM/PoS/Card Statistics, one hand-transcribed workbook per month, each reconciled against its own bank rows. UPI from the NPCI series. Download this exhibit's data (CSV)
This is the cleanest test in the report: everything except the price is held constant, and the priced rail did not die. It kept the large ticket and lost the volume. That gives the merchant leg volume-heavy, value-light shape a mechanism rather than only a description, and it bounds what re-pricing UPI could ever move. Read the memo and what it will not price →
E Market structure and the share cap
- PhonePe45.9%
- Google Pay32.3%
- Paytm8.1%
- Navi4.0%
- super.money1.8%
- BHIM1.0%
- FamApp0.9%
- WhatsApp0.7%
- Axis Bank apps0.6%
- CRED0.6%
Exhibit 11 · Who captures the value
Two apps intermediate 78% of India's transactions, and both sit above the 30% cap
2026-07. Column width is share of transactions, height is average ticket, so area is roughly rupees moved.
above the 30% cap within the cap everyone else Area ≈ rupees moved. Hover a column for detail.
NPCI UPI Ecosystem Statistics, per-application volume and value, transcribed 2026-08-21. Shares computed against the NPCI national monthly total. Download this exhibit's data (CSV)
Exhibit 12 · The cap gap
Complying would move 4.3 billion transactions a month between apps
Share of national UPI volume by app. The dashed line is the 30% cap; panels that breach it are marked.
2023-12 → 2026-07
2023-12 → 2026-07
2023-12 → 2026-07
2024-12 → 2026-07 partial window
2024-12 → 2026-07 partial window
2025-06 → 2026-07 partial window
NPCI UPI Ecosystem Statistics and NPCI monthly product statistics. Download this exhibit's data (CSV)
Exhibit 13 · Share shift
The leaders have barely moved; the challengers started from nothing
Share of national UPI volume, first period against latest.
NPCI UPI Ecosystem Statistics. Download this exhibit's data (CSV)
Exhibit 14 · Concentration
Concentration is falling, nowhere near fast enough to matter by December
Herfindahl-Hirschman Index on true national volume shares. Above 2,500 is 'highly concentrated' by most competition authorities.
Computed from NPCI per-application data. Download this exhibit's data (CSV)
F Banking health scan
- Axis BankPrivate
- Bank of BarodaPublic
- Canara BankPublic
- HDFC BankPrivate
- ICICI BankPrivate
- IndusInd BankPrivate
- Kotak Mahindra BankPrivate
- Punjab National BankPublic
- State Bank of IndiaPublic
- Union Bank of IndiaPublic
Exhibit 15 · Cohort margins
Private banks hold a 114bps margin advantage that has not closed
NIM proxy: net interest income over average total assets, computed from filed statements. The shaded band is the gap.
Filed annual income statements and balance sheets, retrieved via yfinance. A proxy, not reported NIM. Banks use average earning assets, a smaller denominator. Download this exhibit's data (CSV)
Exhibit 16 · Margin decomposition
The advantage splits almost evenly between what banks earn and what they pay
FY2026. Neither lever alone closes the gap, which is why it persists.
Filed annual income statements and balance sheets, retrieved via yfinance. Download this exhibit's data (CSV)
Exhibit 17 · Financial inclusion
By 2024, each banked adult was running 14.9 UPI transactions in an average month
Up from 4 in 2021. Access stopped being the constraint; intensity is the story, and the cost base.
World Bank Findex account ownership (% of population 15+); the 15+ base is total population minus ages 0 to 14. UPI volume is the mean month of each year. Download this exhibit's data (CSV)
Exhibit 18 · The market's verdict
The thinner-margin cohort massively outperformed: the franchise was already in the price
Median price return by cohort, 2021-11-08 to 2026-11-05.
Daily closes via yfinance. Price return only, and it excludes dividends, which understates the higher-yielding public cohort. Download this exhibit's data (CSV)
G Private equity diligence simulation
Exhibit 19 · Unit economics
50.7 million merchants, ₹2,06,685 of GMV each per quarter, ₹0 of payment revenue
Each merchant runs 487 transactions a quarter. The bars show what one merchant would be worth per year if an MDR existed.
PhonePe Pulse, 2026Q2. PhonePe's own transactions, not all of UPI. Accessed 2026-08-20. Values as published at the 2026-08-21 pull; PhonePe has since withdrawn them from its open data. Download this exhibit's data (CSV)
The recommendation: conditional no on a payments-fee thesis, conditional yes on merchant financing. The payment is the acquisition channel; the loan is the P&L. Read the diligence memo and its ambiguity register →
H Survey analytics
Exhibit 20 · Episode-level NPS
One brand contains both the best and the worst episode in the market
Scored per customer episode rather than per brand. A blended brand NPS averages these opposites into a number that hides the problem.
SYNTHETIC panel of 2,000 respondents generated from a documented, seeded model (analysis/04_survey_nps.py). The method is the artefact; the figures illustrate it and are not market fact. Download this exhibit's data (CSV)
I Geographic gap
Exhibit 21 · One spends, one sends
Delhi runs 69% of its transactions through merchants; West Bengal runs 56%
Merchant share is measured from each state's own category split, not inferred from ticket size. Red is above the national 64%; bubble size is share of national transactions; states under 1% are omitted.
PhonePe Pulse, 2026Q2. PhonePe's own transactions, not all of UPI. Accessed 2026-08-20. Values as published at the 2026-08-21 pull; PhonePe has since withdrawn them from its open data. Per-state category files, reconciled against the country file at a ratio of 1.000000. A composition measure, not per-capita: no population denominator is invented. Download this exhibit's data (CSV)
J Wealth and asset management
Exhibit 22 · The fund shelf
14,430 listed schemes are really 3,398 strategies
Strip the Direct/Regular and Growth/IDCW wrappers and roughly 76% of the product count disappears. Counts schemes, not rupees.
AMFI daily NAV file, 04-Nov-2026. Scheme counts, not AUM. AMFI's NAV file carries no assets under management. Download this exhibit's data (CSV)
K The only comparable rail
Exhibit 23 · Same shape, twice
Both rails put the transactions in the merchant leg and the money somewhere else
Each bar is a leg's share of its own country's total, so nothing is converted between currencies. India is 2026Q2 on PhonePe's book; Brazil is 2026-10, national.
Banco Central do Brasil, Pix open data, and PhonePe Pulse. Computed by analysis/09_pix_comparator.py. Download this exhibit's data (CSV)
Exhibit 24 · The migration
Brazil's merchant leg went from 5.3% to 45.3% of the rail with no price to attract it
Monthly since Pix launched, on a time axis. The dashed line is India's merchant share of transactions in 2026Q2. The value line barely moves, which is the point.
Banco Central do Brasil, Pix open data. Merchant is BCB's own P2B label, not derived. Download this exhibit's data (CSV)
Exhibit 25 · What was built on top
Six years on, everything layered on the free rail is 0.36% of merchant transactions
Recurring debits, open-finance initiation and contactless, each as a share of Brazil's merchant-leg volume. A layer reads null before it launched, never zero. Dynamic QR, which is distribution rather than revenue, is 84% and off this scale.
Banco Central do Brasil, Pix open data, initiation method within the P2B leg. Download this exhibit's data (CSV)
Exhibit 26 · The disclosure gap
Brazil publishes instant-payment fraud every month; India publishes no machine-readable equivalent
Accepted fraud disputes per 100,000 transactions, and the share of disputed value returned. There is no Indian line to draw. The series ends 2026-09, four months behind the transaction table.
Banco Central do Brasil, EstatisticasFraudesPix. India's losses are answered in Parliament on perimeters that do not line up with each other, which is why no comparable series is attempted. Download this exhibit's data (CSV)
Why the comparator settles it: the split this report opens with is not an artefact of India's zero-MDR statute. Brazil never had a discount rate to remove and arrived at the same shape, on a rail its central bank has had six years to monetise. What Brazil built on top is real and still 0.36% of merchant transactions, which is an option rather than a business. Read the memo and what one comparator cannot show →
L What each model actually earns
Exhibit 27 · The margin ladder
The transaction business carries the most revenue and converts the least of it
Filed annual accounts, FY2026. Bars are revenue, the line is net margin. Named companies rather than cohort averages, because two of the three groups hold a single company.
Filed annual income statements via Yahoo Finance. Not like-for-like companies: different revenue recognition, perimeters and regulators, so only the shape of the margin each model produces is compared. Computed by analysis/10_business_models.py. Download this exhibit's data (CSV)
Exhibit 28 · How they got there
The rail took four years to reach a margin the others started above
Net margin by fiscal year. A company reads null in a year it had not listed or did not report, never zero, so the lines break rather than bridge.
Filed annual income statements via Yahoo Finance. Download this exhibit's data (CSV)
This is the recommendation, audited: the report has argued since section A that the business is distribution rather than transactions. The only listed transaction business earns the thinnest margin in the panel, and the depository, which is plumbing everyone must cross and is allowed to charge for it, earns the widest. The difference between those two is not scale. It is price. Read the memo and what four companies cannot show →
M Analyst workbench
Interactive · Client-ready dashboard
Every state's payment mix, three linked ways, without leaving the page
A tile cartogram, a sortable table and a detail panel, all linked: select in one and the others follow. Metric switch, live filter, no BI backend.
Loading the map…
44% 71% Merchant share of a state’s own transactions
Select a state for its full breakdown. Boundaries follow India's official depiction.
Select a state on the map or in the table to break out its transaction mix.
Merchant P2P Utility
- Merchant share of value
- Average ticket
- Merchant average ticket
- Share of national transactions
- Share of national value
| State / UT | Merchant share | Share of txns | Share of value | Avg ticket |
|---|---|---|---|---|
| Chandigarh | 70.96% | 0.07% | 0.06% | ₹1,004 |
| Dadra & Nagar Haveli & Daman & Diu | 69.29% | 0.09% | 0.06% | ₹804 |
| Puducherry | 69.05% | 0.05% | 0.05% | ₹1,157 |
| Delhi | 68.51% | 3.03% | 2.59% | ₹1,004 |
| Haryana | 67.94% | 2.92% | 2.78% | ₹1,117 |
| Goa | 67.90% | 0.14% | 0.15% | ₹1,251 |
| Karnataka | 67.53% | 11.62% | 11.63% | ₹1,178 |
| Maharashtra | 67.39% | 13.10% | 12.01% | ₹1,079 |
| Lakshadweep | 67.00% | 0.00% | 0.00% | ₹998 |
| Uttarakhand | 66.38% | 0.64% | 0.56% | ₹1,033 |
| Telangana | 66.03% | 10.03% | 11.27% | ₹1,322 |
| Gujarat | 65.39% | 2.99% | 2.85% | ₹1,121 |
| Kerala | 65.14% | 0.76% | 0.74% | ₹1,139 |
| Rajasthan | 64.20% | 6.60% | 7.01% | ₹1,249 |
| Himachal Pradesh | 64.10% | 0.19% | 0.18% | ₹1,096 |
| Madhya Pradesh | 63.78% | 5.91% | 5.65% | ₹1,125 |
| Tamil Nadu | 63.63% | 2.71% | 2.89% | ₹1,255 |
| Andaman & Nicobar Islands | 63.40% | 0.02% | 0.02% | ₹1,347 |
| Punjab | 62.67% | 0.73% | 0.81% | ₹1,309 |
| Jammu & Kashmir | 62.57% | 0.49% | 0.45% | ₹1,081 |
| Uttar Pradesh | 62.56% | 9.48% | 8.66% | ₹1,075 |
| Chhattisgarh | 62.29% | 1.67% | 1.49% | ₹1,050 |
| Odisha | 61.24% | 3.94% | 3.66% | ₹1,094 |
| Jharkhand | 61.16% | 2.11% | 2.00% | ₹1,115 |
| Andhra Pradesh | 60.96% | 8.40% | 10.16% | ₹1,424 |
| Arunachal Pradesh | 60.77% | 0.11% | 0.10% | ₹1,169 |
| Sikkim | 58.30% | 0.04% | 0.04% | ₹1,221 |
| Assam | 57.31% | 1.42% | 1.14% | ₹946 |
| Ladakh | 56.93% | 0.02% | 0.03% | ₹1,670 |
| Tripura | 56.68% | 0.06% | 0.06% | ₹1,170 |
| Bihar | 56.58% | 5.92% | 5.89% | ₹1,170 |
| West Bengal | 56.43% | 4.61% | 4.86% | ₹1,241 |
| Nagaland | 49.22% | 0.03% | 0.04% | ₹1,325 |
| Meghalaya | 48.37% | 0.06% | 0.05% | ₹1,086 |
| Mizoram | 48.20% | 0.01% | 0.01% | ₹1,313 |
| Manipur | 43.52% | 0.03% | 0.04% | ₹1,623 |
PhonePe Pulse, 2026Q2. PhonePe's own transactions, not all of UPI. Accessed 2026-08-20. Values as published at the 2026-08-21 pull; PhonePe has since withdrawn them from its open data.
N Coverage and limits
Scope
The areas this report does not cover are the ones that would change its conclusion
A scan that only ever reports full coverage is not a scan. Every gap below says what is missing and why.
5 covered 7 partial 2 not covered of 14 areas
| What this report covers | Coverage | Artefact |
|---|---|---|
| Retail payments economics | The monetisation gap, four exhibits | |
| Payments market structure | Shares, concentration, the cap priced | |
| What the rail costs to run | The subsidy, measured against national merchant value The state subsidy is measured. Fraud is not: losses are answered in Parliament on perimeters that do not line up with each other or with the fiscal year, and stitching them into one series would be a false comparison rather than a finding. | |
| Banking margins and funding | Public against private, four fiscal years The margin gap and its split into pricing and funding are measured. The mechanism usually offered for it, deposit mix, is not: no CASA or deposit-composition series is computed here, so the joint between the payments sections and this one is a hypothesis rather than a finding. | |
| Private equity diligence framing | A recommendation and an ambiguity register | |
| Geographic segmentation | All 36 states and union territories One quarter, and the merchant mix is a single operator’s. A second operator could invert the ranking, and nothing open publishes one. | |
| Wealth and asset management | The fund shelf, schemes against strategies Scheme counts, not assets. The public AMFI file carries no AUM, so none of this can yet be stated as a fee pool. | |
| Survey analytics | Episode-level NPS The method and the instrument are real; the panel is seeded and synthetic, because episode benchmarks are proprietary. The chart says so as well. | |
| Interactive analysis | Linked cartogram, table and detail panel | |
| Reproducible sourcing | Generated provenance, no figure typed by hand | |
| Consumer credit and lending | Not covered The diligence memo concludes that the loan is the P&L, and then cannot size it. No open, current, India-wide series does. This is the largest gap on the page. | |
| Insurance | Not covered IRDAI publishes to PDF only. There is no machine-readable series to fetch, so nothing is claimed. | |
| A priced rail to compare against | Cards, the domestic control group One month. India runs cards and UPI side by side under one regulator with a merchant discount rate permitted on only one of them, which is the cleanest test in this report. It is a single transcribed month, and it deliberately does not price the card rail: RBI publishes no blended effective MDR, so multiplying a statutory ceiling by total value would publish an estimate as a measurement. | |
| Comparison with other instant rails | Brazil’s Pix, measured against the same question One comparator, not a survey of rails. Pix is the only instant rail at national scale that publishes a merchant leg its own regulator labels, so the shape can be compared without deriving it; every other candidate would have to be inferred. Brazil’s fraud series also stops four months before its transaction series, so the two are not read against each other. |
Self-assessed against the question this report set out to answer. Coverage is judged on what is computed from data in this repository, never on what is asserted in prose.