On-chain volume is not a payment

The funnel from tens of trillions to a few hundred billion

Gross on-chain stablecoin volume in 2025 was about 33–35 tn USD (BIS, Hernández de Cos, April 2026; Visa/Artemis, 2026). The number is true. It is almost unused as a measure of whether a token replaced a payment, because in 99 cases out of 100 it is counting something else.

Three independent methods filter trading, exchange traffic, bots and internal rebalancing and land in the same order of magnitude, not on the same point. BIS and a joint McKinsey/Artemis Analytics paper estimate real-economy stablecoin payments in 2025 at about 390 bn USD, 0.02% of global payments (McKinsey/Artemis, February 2026; BIS, April 2026). BCG and Allium, for Ripple, count more widely and get 350–550 bn USD (BCG/Allium, January 2026). Artemis, counting only declared volume from named payment firms in a survey, gets a lower bound of 122 bn USD run-rate (Artemis "Ground Up", October 2025). Those three figures are three methods. They are not averaged. The gap from 122 bn USD to 550 bn USD is the ordinary spread of a young category in which each aggregator defines real-economy volume differently: whether to count card-linked spend, prefunding, or PSP-to-PSP.

Visa/Artemis, 2026, put 36% of 2025 adjusted volume in deposits and withdrawals at centralised exchanges. Subtract that channel and about 21 tn USD remains. DeFi (7.8 tn USD), centralised exchanges (4.3 tn USD) and MEV (1.9 tn USD) are trading and protocol mechanics, not a payment (Artemis, 2025). Automatic rebalancing and looping inside DeFi are counted many times in raw transfer volume. After those cuts, the Polish edition puts real payments between economically distinct parties at 1.4–1.7% of the original volume.

Inside the 390 bn USD McKinsey/Artemis figure, B2B is 226 bn USD, about 60% of the total, up 733% year on year. Asia is 245 bn USD, also about 60%. Those are two cuts of one sum, not a contradiction. Card-linked stablecoin spend grew to 4.5 bn USD, up 673% year on year, in the same paper. Coinbase's public response is that the growth rate matters more than the point estimate (Coinbase blog, June 2026). That is a participant with a book to defend, labelled as such.

Where they win, in the corridors that still hop

Correspondent banking is the chain of banks that pass a cross-border payment because the sender's bank has no account at the recipient's bank. SWIFT is the messaging network those banks use. It does not hold the funds and it does not clear them. Each surviving hop charges 10–30 USD, and each FX conversion takes a 0.5–2% spread (Polish rail table; knowledge base module 11). Corridors with two to four hops, unbanked recipients who have a smartphone, large one-off B2B transfers, and anything sent outside banking hours are the cases the knowledge base names as wins.

The World Bank's global average cost to send 200 USD was 6.4% (November 2025). Sub-Saharan Africa averaged 8.78% against a 3% Sustainable Development Goal target (World Bank Remittance Prices Worldwide, Q1 2025). The knowledge base records SWIFT on a 200 USD transfer at 5.8–14.55%, source-dependent, and 1–5 business days. A stablecoin transfer is 0.01–5 USD in network fees and settles in seconds to about a minute on the chain. It is irreversible by design. Those two sentences are not a complete payment. The ramps at each end still have to exist.

Four cost lines sit on a traditional cross-border payment, and a token only removes two of them.

1. The FX spread, the gap between the mid-market rate and the rate the customer is actually given. Most of the margin hides here, because the fee line on the receipt is small.
2. Prefunded liquidity: the provider parks local currency in every country in advance. That cost grows with the number of countries, not with volume, so a tiny corridor costs as much to open as a large one.
3. Correspondent-hop fees, the 10–30 USD per bank, with fewer paths left after large banks dropped high-risk regions.
4. Compliance: sanctions and AML screening. The user does not see a line item. The user sees money that does not arrive on time.

A chain hop removes (2) and (3). Prefunding is unnecessary for the middle segment, and one on-chain transfer replaces the hop chain. The FX spread does not vanish. Fiat-to-token and token-to-fiat still need a rate, so the spread moves to the on-ramp and the off-ramp. Compliance does not vanish either. It moves to the ramp provider, which has to run Travel Rule and screening before it pays out fiat. The token wins on the middle of the route. On the two edges, the cost changes vendor.

Wise reaches a similar all-in cost without a chain. A sender in the UK pays pounds into a local Wise account; a recipient in Brazil is paid reais from a local Wise account over Pix. Two domestic payments net one cross-border transfer on Wise's books. World Bank Remittance Prices Worldwide 2025, as cited in September 2026, puts bank transfers above 13%, digital-wallet operators around 3.5%, and Wise in the 0.5–1% band. Wise built that with eighty-odd local licences. A stablecoin corridor has to buy an on-ramp and an off-ramp in each country, which is the same licensing problem under another name.

A remittance corridor is a named pair of countries along which wages travel, measured as a volume. Addressability is low on G10-to-G10 flows, where SEPA Instant already settles in ten seconds at a regulated cost no higher than a standard SEPA credit. It is highest on non-G20-to-non-G20 and emerging-market-to-emerging-market flows, where the correspondent chain is longest. That ranking is a knowledge-base item via the brief.

Where they lose, including to a rail that is already free

The usual slide puts a stablecoin next to a card and next to SWIFT, the dearest and slowest parts of the global system. For most domestic retail payments the alternative that already exists is a local instant rail. UPI, Pix and, in Poland, BLIK won that fight against the card years ago.

UPI, built by the National Payments Corporation of India, a bank-owned non-profit, processed 228.3 billion transactions worth 299.7 tn INR, about 3.4 tn USD, in 2025 (NPCI, July 2026). Pix, built and mandated by Banco Central do Brasil, processed about 80 billion transactions worth more than 35 tn BRL, about 6.5 tn USD (Banco Central do Brasil, October 2026). Each of those two rails, in one country, is larger than the entire global real-economy stablecoin payment estimate: UPI about 9×, Pix about 17×, against the 390 bn USD McKinsey/Artemis point. They are domestic instant rails, not cross-border competitors. The comparison is about which use cases are still addressable, not a head-to-head loss.

Visa processed 71.7 billion transactions in one quarter of fiscal 2026, with payment volume above 4 tn USD in that quarter (Visa earnings, 28 July 2026), more than 16 tn USD annualised, again a multiple of the whole stablecoin real-economy band.

UPI's merchant discount rate on person-to-merchant transactions has been zero since 2020. The bill is paid by the state budget (subsidies to issuing banks) and by NPCI, which runs as a non-profit. Pix never charges the consumer; the merchant pays 0.22–0.33%, or 0.30–0.90 BRL per transaction, and large merchants subsidise micro-merchants. Calling UPI free at the point of payment means the cost was moved onto the taxpayer and the bank. The rail still has a cost.

SEPA Instant is, by regulation, no more expensive than a standard SEPA credit and final in ten seconds, 24/7. FedNow and RTP in the US clear in under 20 seconds; the sending bank pays 0.045 USD per transaction plus 25 USD a month. A stablecoin still carries a network fee of a fraction of a cent to 5 USD plus a 0.1–1% ramp spread at each end. On a developed-market retail payment that already has a free or near-free instant rail, the token adds a conversion the local infrastructure does not need.

A chargeback is a card network's reversal of a settled payment, open to the cardholder for 120–540 days on the Polish rail table. Card chargebacks cost merchants 33.79 bn USD globally in 2025. A stablecoin transfer is a push payment: once the chain has finalised it, nobody in the protocol can reverse it. That is a feature for a B2B settlement that should stay settled. It is a defect at a retail checkout that still wants chargeback protection. The merchant who accepts the token takes the dispute risk the card network used to carry. Small developed-market retail, G10-to-G10 corridors, and anything below the on/off-ramp cost threshold are the losses the knowledge base names.

Remittances already running at scale

The documented case is a corridor with no working local instant rail and no Wise-style licence stack, or a country where dollars are scarce and inflation is high.

Felix Pago runs a USDC-to-SPEI path over WhatsApp: dollars in, USDC across a chain, pesos out. By September 2026 it had processed more than 8 bn USD of remittances across 11 Latin American countries and served more than 6 million people, at under 1% against 5–7% on traditional channels (Mizuho, cited September 2026). The main corridor is US to Mexico; Colombia, El Salvador and Nicaragua are named in the Polish table.

Bitso is an exchange-native remittance. It processed 6.5 bn USD of US–Mexico remittances in 2024, about 10% of that corridor, at under 1% including on/off-ramp fees.

Zepz, the parent of Sendwave and WorldRemit, put Sendwave Wallet on Solana with USDC from 10 October 2025. It transferred 17 bn USD for clients in 2025 on UK/US-to-Africa corridors (Ghana, Kenya, Tanzania). Traditional costs of 7–12% are recorded as compressed to 1–3%.

LemFi, a USDT-backed fintech on UK/US/Canada/Europe to Nigeria, Ghana, Kenya and 30-plus markets, processed more than 1 bn USD of monthly volume as of August 2026, with the FX margin compressed to 1.5–2.5% above mid-market. Tether is an investor. The firm is in the Polish table and not in the chapter brief's company list; the scale is kept because the Polish edition dated it.

Circle Payments Network, the infrastructure that connects licensed institutions for real-time settlement, had four active payment corridors and four financial institutions on 10 August 2025 (Circle Form 424B4 / S-1, SEC, August 2025). The number is hard because it sits in a filing. It is also small, which is the size this segment still is once the 33 tn USD headline is removed.

B2B, merchant checkout and payroll in production

B2B is the largest slice of the McKinsey/Artemis 390 bn USD, at 226 bn USD. PYMNTS puts 2025 B2B stablecoin volume at about 36 bn USD. Those two numbers use different definitions of B2B. They are not comparable without matching methodology, and they are not averaged.

PayPal's PYUSD checkout reaches more than 35 million merchants and a user base of more than 400 million (knowledge base via the brief). The Polish edition puts PayPal's active accounts at 439 million at year-end 2025. Those are two readings of the distribution surface, not two supplies. PYUSD supply on 4 September 2026 was 3.00 bn USD. The checkout can settle fiat automatically for a merchant who does not want to hold a token.

Stripe charges 1.5% for stablecoin payments against about 2.9% for cards. The chapter on chains recorded the card-standard as 2.9% + 0.30 USD; both phrasings sit in the sources and are not collapsed.

Payroll is in production at three named firms. Deel launched stablecoin payroll on Polygon on 20 May 2026. Mercury Rise passed 1 bn USD of cumulative payroll volume in November 2025, with 53% or more of users choosing a stablecoin payout. Bitwage has processed more than 400 mn USD with more than 90,000 registered workers. All three dates and scales are knowledge-base items via the brief.

A later chapter maps the GENIUS Act's payment-stablecoin definition onto which of these use cases can scale in the United States. Company-level profiles of Stripe, PayPal, Visa and Mastercard sit in a later chapter as well.

A cross-border payment, step by step

An on-ramp converts fiat into a token. An off-ramp converts the token back into fiat. Each end carries a 0.1–1% spread in the Polish rail table, so a transfer that costs nothing on-chain still has a ramp cost.

Take Felix Pago's US-to-Mexico path, because it is the one with a public scale.

1. The sender pays dollars, often from a US bank account or a cash agent, into Felix's on-ramp. The FX clock starts here. The spread that used to sit inside a correspondent quote sits inside this conversion.
2. The on-ramp mints or sources USDC. Travel Rule and sanctions screening happen at this vendor, not at a correspondent in the middle.
3. USDC moves on a chain. Network fee 0.01–5 USD, seconds. This is the hop that replaced two to four correspondent banks and the prefunded peso inventory those banks used to hold.
4. The Mexican off-ramp burns or sells the USDC and pays pesos. A second 0.1–1% spread.
5. The recipient is credited on SPEI, Mexico's instant rail. From the recipient's seat this is a domestic peso payment.

All-in, Felix publishes under 1% against 5–7% traditional. The chain did not make FX free. It made the middle free enough that the two ramps, plus WhatsApp as the front end, still undercut Western Union.

The same sandwich fails where Pix or UPI already runs end to end inside one country, because there is no correspondent middle to delete, and it fails at a US checkout that still wants a chargeback window, because the push payment has no third party that can reverse it. The previous chapter, on chains and distribution, is why so much of the P2P leg of this sandwich already sits on Tron rather than on Ethereum: the fee that the casual sender sees is the chain's fee, and Tron priced that fee for a 50 USD transfer.