How SWIFT actually works

When a customer in New York sends dollars to a supplier in Jakarta, no dollars leave New York. What travels is a standardised instruction, usually an MT103, telling a chain of banks to adjust their internal ledgers. The dollars themselves stay on the New York bank's books. The Jakarta bank credits its customer against a different balance sheet entry entirely.

Value moves by ledgers agreeing with each other, not by cash flowing across borders.

For those ledgers to agree, the two banks need a pre-existing account relationship. A nostro is an account you hold at another bank, in their currency. A vostro is their account at you. These are credit relationships. Every time a nostro is drawn to fund a payment, one bank is extending the other a line, intraday or overnight.

Most banks cannot economically maintain nostros in every currency they trade, so payments hop through larger correspondent banks that do.

Each hop lifts a fee. Somewhere in the chain, FX conversion happens at whatever rate the correspondent offers.

Somewhere in the chain, currency conversion happens at whatever rate the correspondent offers, with a spread that neither the sender nor the receiver sees until the payment lands. Settlement between correspondents happens on scheduled cycles, such as CHIPS, Fedwire and TARGET2, rather than atomically. Which means the SWIFT message arrives hours or days before the balances between correspondents actually clear.

What you are actually paying for

The result is the cost stack every cross-border payment inherits, all of it bundled into a single opaque number the customer sees at the end.

• Lifting fees: Charged by each correspondent in the chain for handling the payment.

• FX spread: Applied at whichever hop converts, at a rate the sender does not see in advance.

• Credit cost: Someone is extending a line on a nostro to fund the payment before it settles.

• Float: The period between the instruction arriving and the balances clearing.

The banks are not being inefficient. The architecture bundles four different services and prices them together.

Correspondents are contracts, not connections

A correspondent relationship is not a technical link. It is a bilateral contract between two banks, specifying which currencies each will clear for the other, the daylight and overnight credit limits on the nostro, the pricing schedule for lifting and FX, and the sanctions and AML terms each side commits to.

Setting one up takes months, sometimes over a year, and is signed off at treasury and compliance level on both sides.

Illustrative terms. Because these are contracts rather than connections, banks maintain a finite number of them.

A large global bank might have hundreds of these relationships. A mid-size emerging market bank might have five, covering the currencies its customers actually need. Anything outside that list has to be routed through one of those five, which is why chains form in the first place.

The two flows

Every cross-border payment is actually two parallel flows that have to stay in sync.

The information flow is the SWIFT message traversing the chain node by node, from sender bank to correspondent A to correspondent B to receiver bank. Each hop parses the MT103, runs its sanctions screen, and re-emits an MT202 to the next correspondent or an MT103 to the ultimate recipient.

The funds flow is the ledger adjustment at each hop against the standing correspondent balance. The sender bank's nostro at correspondent A is debited, correspondent A's nostro at correspondent B is debited, and so on down the chain.

The two flows are not atomic. A message may arrive at the receiver before the correspondent balance has cleared, or a nostro may be debited before the next message hop lands.

Every bank runs a reconciliation function whose entire job is to paper over the mismatch between what its correspondents have said and what has actually settled.

Chains grow longer, not shorter

The four-node example above is the simple case. Real chains routinely run to five or six hops in emerging market corridors. A Luanda bank routes dollars through a Lisbon correspondent, which uses its New York correspondent for the clearing leg, which hands off to a Singapore bank with a rupiah relationship, which finally settles into the Jakarta receiver.

A payment that stays in the Southern Hemisphere crosses three continents to get there.

Real chains in emerging market corridors routinely run to five or six hops.

The trend since 2015 has run the other way from what you might expect. Global banks have de-risked, closing correspondent relationships where the compliance cost per dollar of revenue could not be justified, especially in Africa, the Caribbean and Central Asia. That pushes payments in those corridors through longer chains, not shorter ones.

What changes

Two things are collapsing parts of this architecture.

Local payment service providers collapse the last-mile leg, absorbing the licensing, the rails and the KYC inside a destination country so a global business does not have to hold a relationship with every receiving bank.

Atomic stablecoin settlement collapses the two flows entirely. When the instruction and the value are the same object moving on the same ledger, there is no mismatch window, and therefore nothing for a reconciliation function to paper over.

What remains in both cases is the on-ramp and the off-ramp, where value still has to enter and leave the banking system. That is a narrower problem than the one described above, and it is where the current market is concentrating its effort.