Section 01

The landscape

Who competes in cross-border payments matters less than where the money goes on its way from the customer to the payment system. Read as a supply chain rather than as a field of rivals, this market answers two questions a list of companies cannot: which part of it we should buy from, and what owning infrastructure actually buys you.

Start with the payment, not the companies

A competitive landscape is usually drawn as a field of rivals. That turns out to be the wrong picture here, and it is worth seeing why before looking at any firm at all.

Consider a single payment: a European business owes a supplier in Mexico €100,000. Four separate things have to happen, and each is a different job requiring a different capability.

ONE PAYMENT, FOUR SEPARATE JOBS — AND FOUR SEPARATE MARKETSA firm can be excellent at one of these and buy the other three. Most are.COLLECTTake euros from the payerneeds:a euro account and access to the euro rails€0.002 per paymentCONVERTTurn euros into pesosneeds:a counterparty willing to take the other sideunder 1 bp to manufactureTRANSMITMove value across the borderneeds:a correspondent bank, a chain or a partnervaries enormouslyPAY OUTPay the recipient in pesosneeds:access to the destination country’s railfractions of a centThe conversion is highlighted because it is where almost all of the money turns out to be — see the next two figures.SOURCED · C2 C3 C6 · 2026-08-11
The four jobs inside one cross-border payment. They are separable, they are bought and sold separately, and very few firms do all four themselves.

This matters because the four jobs have wildly different economics. Taking euros from the payer and putting pesos into the recipient's account are both nearly free — the payment systems that do it charge fractions of a cent. Moving value across the border varies enormously depending on the route chosen. And converting euros into pesos costs, at wholesale, less than one hundredth of one per cent.

Yet a European business paying that Mexican supplier is typically charged around 1.6% all in. So the interesting question is not who competes — it is where the difference goes.

A note on basis points

Almost everything below is quoted in basis points, because the numbers are small and percentages become unreadable. One basis point is one hundredth of one per cent. On a €100,000 payment, one basis point is €10. The 1.6% figure above is 160 basis points, or €1,600.

Who sits between the customer and the rail

The four jobs are rarely performed by one firm. They are performed by a chain of firms, each buying from the one closer to the payment system and reselling to the one further away. Understanding that chain is the most useful thing in this section, because it explains almost everything else — including why the firms you have heard of charge what they charge.

The research found the chain has five distinguishable steps, and that the one step with no regulatory gate at all is exactly the step that sells routing.

BETWEEN THE CUSTOMER AND THE RAIL, IN ORDEREach step buys from the one below it and adds a margin. Read bottom-up: that is the direction the money travels.The rail itselfBanxico SPEI · Brazil PIX · TIPS · UKFaster PaymentsA national payment system. Operated by a central bank or a scheme.Not a company you can buy from.takes 20–60 bpA licensed principalSTP · Banking Circle · ClearBankHolds its own licence and its own connection to the rail. One hopfrom the payment system itself.takes 20–60 bpTHE STEP TO BUY ATclosest to the rail that will sell to usA local aggregatordLocal · EBANX · PagBrasilLicensed in the destination country, but reaches most of it bycontracting with other local firms.takes 20–60 bpA network aggregatorThunes · TerraPay · Nium · Paysend · PPROContracts with aggregators in many countries and resells thecombination. No licence gate to occupy this step.takes 20–60 bpNO LICENCE NEEDEDanyone can occupy this stepAn orchestratorthe layer the original proposal describedBuys from the step above and chooses between the options it isoffered.dLocal is placed at the local-aggregator step on the evidence of its own annual report: it names more than 100third-party entities, and its ten largest counterparties are only 40% of the volume it processes.SOURCED · E1 C6 · 2026-08-11
Read from the bottom up — that is the direction money travels. Each step buys access from the step below and resells it with a margin. Only two of the five require a licence to occupy.

Two things there are worth dwelling on, because neither is obvious and both took real work to establish.

First, firms are not where you would assume. dLocal is widely described as the local partner for Latin America. On the evidence of its own annual report it sits at the local-aggregator step: it names more than a hundred third-party entities it routes through, and its ten largest counterparties account for only 40% of the volume it processes. It is not close to the rail. It is an aggregator of firms that are.

Second, one step requires nothing at all. Thunes, TerraPay, Nium, Paysend and PPRO occupy the network-aggregator step, which has no licensing requirement, no capital requirement and no direct connection to any payment system. What they sell is the contract-gathering itself — the fact that they have already signed the agreements you would otherwise have to sign. That is a real service. The question this research had to answer is what it is worth, and the answer turns out to be uncomfortable for anyone proposing to sell the same thing.

The question every counterparty conversation should open with

Before leaving the chain, one practical finding. The unbundling question cannot be answered from public sources for a single counterparty that matters — not one of them publishes a payout rate card or a prefunding minimum. The reason for that silence is simple once stated: a firm selling a single leg has a per-transaction price it can publish, whereas a firm whose product is a blended conversion-plus-fee bundle cannot publish one without revealing how the bundle splits.

Which turns into a single question, worth more than any rate card:

Will you quote the payout leg as a flat fee per transaction in the destination currency, with no conversion attached?

A firm that will is selling a leg. A firm that will not is selling a bundle — and the bundle is where its margin hides.

Where the money actually goes

Now the chain and the four jobs can be put together. Here is the €100,000 payment again, with what the customer pays set against what the payment costs to produce.

A €100,000 PAYMENT FROM EUROPE TO MEXICOEverything between what it costs to make and what the customer pays is distribution — margin taken by the firms in between.What the customer pays160 bp on €100,000 = €1,600What it costs to manufactureunder 1 bp — the conversion itself, at wholesaleWhat the rail costsfractions of a cent, both ends≈159 bp OF DISTRIBUTIONshared between every firm that touched the payment — roughly 20–60 bp per stepThis is the pool. The whole strategic question is which part of it we can hold.The 160 bp all-in figure is measured for business cross-border payments; the sub-1 bp manufacturing cost is for a major-currency conversion at wholesale. Both are sourced.MODELLED · A2 C3 C2 E1 · 2026-08-11
The same payment, priced. The gap between manufacturing cost and customer price is not a margin any single firm earns — it is shared out along the chain.

That gap is the whole prize, and it is worth being precise about what it is made of. It is not a technology cost and it is not a payment-system cost; both are close to zero. The conversion costs under one basis point to manufacture at wholesale while customers pay between 52 and 400 — so what sits in between is distribution: the cost of reaching the customer, and the fact that the customer does not know what the wholesale price was.

That carries an uncomfortable implication which recurs throughout this work. A business built on this gap is not built on producing something more cheaply than anyone else. It is built on being the firm the customer buys from. Those are different kinds of advantage, and they decay at very different rates.

The intuition about owning rails, and why it fails

Here is the reasoning almost everyone applies to this market, ourselves included at the outset. Every step in the chain takes a margin. So a firm that removes steps — by connecting directly to the payment systems itself — ought to keep more of what the customer pays. Owning rails should mean earning more.

It is a reasonable theory, and it is testable, because several firms disclose enough to place them on both axes: how many payment systems they connect to directly, and what share of each payment they keep as revenue. That second figure is conventionally called the take rate.

IF OWNING RAILS PAID, THIS CHART WOULD SLOPE. IT DOES NOT.Each dot is a firm. Right means more direct connections to payment systems. Up means a higher share of the payment kept as revenue.1101004000369PAYMENT SYSTEMS CONNECTED TO DIRECTLYKEPT FROM THE PAYMENT, bps (log scale)Wise’s own cost: 18.6 bpthis is what owning rails bought —a lower cost, not a higher priceBanking Circle1.5 bpNium22 bpCorpay22 bpAirwallex45 bpWise52 bpPayoneer115 bpRemitly211 bpdLocal238 bpZepz375 bpmost connected of anyone here —and it cut its price rather than raise its marginTake rates are not perfectly like-for-like across these firms — some are revenue over volume, some gross profit over volume. Thespread is far too large for that to explain it.SOURCED · B2 B1 C3 A1 · 2026-08-11
If the theory held, this would slope upward from left to right. Instead the most connected firm sits in the middle of the range, and the firm with the deepest direct access earns the least of anyone here.

Banking Circle has direct access to more payment systems than almost anyone in this market and keeps about 1.5 basis points of what passes through it. Zepz owns almost no infrastructure and keeps between 350 and 400. On the theory above, those two should be the other way round.

Wise is the case that explains why. It has more direct scheme connections than any other firm here, built up over eight years, and its owned-rail cost advantage is real, large and quantifiable — and it is being competed away as fast as it is earned. The research reconstructed its unit cost at 18.6 basis points. It charges around 52. It could have kept the difference; instead it cut its price, and its net income fell 9% in a year when its volume rose 31%.

So the correct statement is not that owning rails is worthless. It is that the take rate is a function of distribution and of the customer’s alternative, not of routing intelligence. Owning the rail sets the floor — how cheaply the payment can be produced. What is charged above that floor is set by who the customer is and what else they could do instead. The two move independently, and conflating them is how a business ends up built on the wrong thing.

Stated plainly for our own planning: the inverse relationship between rail ownership and take rate holds across the disclosed names. The risk of not owning a rail is a higher unit cost and more exposure when a partner has an outage. It is not lost margin.

What this leaves for a new entrant

Three consequences follow, and they shape everything in the later sections.

The incumbents each have a reason not to route honestly. The structural weakness is not routing — it is licensed last-mile depth without balance-sheet drag. Every established firm earns its money on one particular leg, which means recommending a route that bypasses that leg costs it revenue. A firm with no leg of its own does not carry that conflict. That is a genuine asymmetry, and it survives the obvious objection that aggregation already exists.

Most of the firms here are suppliers, not rivals. These are overwhelmingly counterparties rather than competitors, with two exceptions. Read as a supply chain rather than a battlefield, most of this landscape becomes a list of firms to buy from at the right step — which is a considerably more useful list.

And neutrality is not available to us. The asymmetry above works only for a firm with no leg of its own, and the later sections conclude we should hold one — the conversion — precisely because that is where the money is. The revised position forfeits the neutrality argument, and we should stop citing it. That is the correct trade, but it should be made deliberately rather than discovered in diligence.

What the buyers are doing

One last piece of context, because it bears on how this ends. Consolidation is running hard, and value is accruing to distribution and to licensed local rails — not to routing. Eight firms in this layer changed hands in the eighteen months to August 2026. Not one was bought for a routing layer. The largest of them was bought for orchestration plus more than twenty-five licences, 130 markets and live volume — which is a statement about what the acquirers believe is scarce.

The section on what we take from all this argues that acquisition is the realistic outcome, and the good one to plan for. It appears here only because it follows directly from the pattern above, and because a reader who disagrees should disagree at this point, against this evidence, rather than later.