A few years ago, I dedicated one of the first posts on this blog to correspondent banking and its implications in the global financial system.
That analysis covered the first layer: the logical foundation of how correspondent banking works, how the information involving financial transactions is processed, and how messages are exchanged among banks worldwide. But it didn’t cover how value moves. It didn’t explain how money is materially transferred from one bank account in one banking system to another bank account in another.
The goal of this post is to fill that vacuum, to explain how money really moves, what are its financial implications and whether the emerging stablecoin infrastructure is genuinely improving it or merely relocating its constraints to a different layer.
The current model
International payments do not move money across borders. They move instructions and this is what SWIFT messages are: financial instructions.
When a bank in Milan sends US dollars to a counterparty in New York, it doesn’t transmit value, it sends a message via SWIFT, which routes the instruction through a chain of intermediary banks until it reaches one with an account in the right jurisdiction and the right currency. The actual transfer of value happens later, through domestic settlement systems, governed by local rules and local hours.
This architecture exists because no universal settlement layer connects all banks globally. Instead, banks maintain bilateral relationships with foreign counterparties holding pre-funded balances in each other’s books across multiple jurisdictions and currencies: these pre-funded accounts are called Nostro and Vostro accounts.
- The account a Bank A holds abroad at Bank B, in a foreign currency, is its nostro (“ours” in Italian).
- The corresponding account a foreign bank (Bank B) holds with Bank A is its vostro (“yours” in Italian).
To see how this works in practice, let’s take a simple FX trade: a European bank – Unicredit – sells EUR and buys USD. What looks like a single transaction is actually two separate payments, running through two separate systems (I will use JPMorgan as the counterparty for this example).
Unicredit holds a USD nostro account at JPMorgan: pre-funded dollars sitting on JPMorgan’s books, ready to be debited when Unicredit needs to make a dollar payment. JPMorgan, symmetrically, holds a EUR vostro account at Unicredit.
When the trade settles, the two legs move independently.
Leg1 (EURO)
– Unicredit’s TARGET2 account is debited (Unicredit loses EUR)
– JPMorgan’s Vostro at Unicredit is credited (JPMorgan gains EUR exposure)
Leg2 (USD)
– JPMorgan’s Fedwire account is debited (JPMorgan loses USD)
– Unicredit’s Nostro at JPMorgan is credited (Unicredit gains USD)
Neither system waits for the other: TARGET2 closes at 18:00 CET, Fedwire closes at 21:00 ET which is midnight in Frankfurt. If the EUR leg settles at 10:00 CET and the USD leg doesn’t clear until the New York afternoon, Unicredit has been short EUR for several hours with no guarantee the dollar delivery is coming. The gap between obligation discharged and value received is the risk that pre-funded accounts exist to absorb.
Across the global correspondent banking network, approximately $3–4 trillion sits pre-funded in nostro accounts at any given moment, capital deployed not to generate returns, but simply to guarantee that the second leg of a trade can settle when the first already has.
Why Prefunding Exists
The Unicredit/JPMorgan example illustrates the operational reality, but the deeper question is why the system was built this way at all. There are two root causes.
The first is structural: every jurisdiction runs its own domestic settlement system, with its own hours, cut-offs, and rules. No shared rail connects them. Pre-funded nostro accounts are the virtual bridges across different settlement systems: capital deployed in advance to substitute for a single settlement layer that doesn’t exist.
The second is historical. In 1974, Bankhaus Herstatt received Deutsche Mark payments from American counterparties in Germany and then was shut down by regulators before paying out the corresponding USD leg in New York. The American counterparty banks had paid and received nothing back. This episode gave a name to the settlement risk – Herstatt risk – and made pre-funding not just operationally convenient but a financial necessity.
The answer offered by the system was CLS (Continuous Linked Settlement) Bank, established almost 30 years later in 2002, a payment-versus-payment mechanism that forces both legs of an FX trade to settle simultaneously, so neither side is ever exposed mid-trade. CLS works but only partially, as it covers roughly half of global FX volume. The other half – which mainly comprises emerging markets currencies – still settles bilaterally, leaving those corridors reliant on bilateral nostro pre-funding.
The cost of the current model
As seen above, pre-funding solves crucial trust issues, but it does so at a significant cost.
The most visible cost is opportunity cost. Capital sitting in a nostro account is capital not deployed elsewhere. Banks pay to hold it, earn nothing on it, and size it not to actual transaction flow but to peak demand, which means the buffer is almost always larger than necessary. The settlement window that actually matters is roughly 6am to 11am GMT, the narrow band when most relevant domestic systems are online simultaneously. Outside those five hours, the pre-funded balance sits idle.
The distribution problem is as significant as the size problem. Liquidity is allocated per corridor: a separate pre-funded account for each currency pair, each jurisdiction, each correspondent relationship with no pooling mechanism, nor dynamic reallocation. Capital locked in a Tanzania shilling account cannot cover a shortfall in a Philippine peso account, even if both are held by the same institution. In emerging market corridors the cost structure is worse due to capital controls and higher probability of local currency devaluation.
In addition, there is a massive operational cost: fragmented standards, too many intermediaries, redundant compliance checks, mismatched operating hours. Even when liquidity is in the right place, the plumbing required to move it introduces settlement delays.
The net effect is a model that sizes liquidity need to monthly averages while real demand spikes within a three-hour window, distributes capital by corridor rather than dynamically allocate it by need, and charges the cost of all of it to the businesses and individuals at the end of the chain.
The emerging model
Connecting directly to domestic systems, settling in fewer currencies, reducing intermediary hops are the traditional mitigations that have helped at the margin but haven’t changed the underlying architecture. But a new model is being assembled right now.
The emerging model rests on a simple observation: USDC and USDT already function as a de facto global settlement layer. Unlike correspondent banking, both stablecoin networks operate 24/7, have no cut-off windows, and settle in seconds rather than days. A fintech in Lagos and a supplier in Manila can transact in the same asset, on the same rail, without a chain of intermediary banks in between.
But this is not a finished system yet. The settlement rail exists but the local interoperability layer is still relying on the old infrastructure.
At the moment the interoperability layer with the existing local banking systems is created by a number of on and off-ramps solutions that convert local currencies into global stablecoins (and vice versa) using traditional banking rails. These services provide a useful service, but they represent more a transitionary solution rather than a long term configuration of this new model.
My conviction is that the on and off-ramps of the future will be local stablecoins, which means onchain representations of local currencies. These local stablecoins will be a near instantaneous bridge and allow the minting directly from a local bank account, collapsing the on-ramp friction to a single step and almost eliminating the FX exposure that makes pre-funding so expensive in volatile corridors.
Several companies are building different parts of this stack. Mansa sits at the liquidity layer, providing capital at the point of settlement rather than in advance, with the same pool recycling in 6–48 hours. Hercle pairs a proprietary liquidity book with its own banking rails, supporting local and international accounts and payment networks, so institutional clients can move between fiat and stablecoins within one regulated platform. Noah operates as a stablecoin-native correspondent bank, bridging fiat and stablecoin rails for institutional flows. OpenFX focuses exclusively on the FX conversion layer and now moves $45 billion annually.
Conclusions
The correspondent banking system was not designed to be inefficient. It was designed to be stable and pre-funding is the price of that stability. Where no universal settlement layer exists, capital posted in advance is the only available substitute for trust.
Blockchain infrastructure makes a different claim: that atomic, trustless settlement can close the trust gap without requiring capital to be immobilised in advance. Whether the new generation of solutions will actually deliver this, or merely relocate the trust buffer to a different layer – the stablecoin issuer’s reserve, the liquidity provider’s balance sheet, the on-ramp operator’s capital – remains an open question.
The path towards this new model has proved to be slow and fragmented, but the terms of the debate are changing. Before the recent Hormuz crisis, efforts to modernise cross-border payments were largely framed around efficiency: faster settlement meant lower costs and better capital utilisation. The events of the past few weeks have positioned financial infrastructure as part of economic defence: the ability to move capital quickly is now directly linked to a country’s capacity to maintain trade flows, stabilise markets, and respond to external shocks.
That urgency is new and it may finally be enough to accelerate the migration that efficiency arguments alone never could.
Resources
- Marcello De Cecco — Money and Empire: The International Gold Standard 1890–1914
- R.G. Hawtrey — Currency and Credit (1919)
- Westermeier, Campbell-Verduyn, Brandl (eds.) — The Cambridge Global Handbook of Financial Infrastructure (2024), Chapter 16
- https://stripe.com/en-it/resources/more/global-settlement-systems
- Mansa founder podcast – Link
- https://www.thegccedge.com/gcc-liquidity-mobility-cbdc-27-trillion/
