The first week of October, I will be in Singapore for Token2049 where I will also be hosting a side-event with Fintech Guild and M0 (sign up here). If you are around and want to talk about any of this, ping me.
The stablecoin sandwich – which I presented thoroughly in my post “The Price of Trust” – is a new model that is using stablecoins to re-architect the nature of FX transactions.
The promise of that architecture is that it kills the frictions of correspondent banking but, in its current iteration, only some frictions are solved, while others are simply pushed to the edges of the sandwich.
My last post was about one of these frictions: integrations with banking rails. This post wants to cover another friction vector: local liquidity.
The problem is always the same: a payment company has to have capital sitting in the destination corridor before, or at the very moment of, settlement – and capital sitting there is capital it cannot deploy anywhere else. The traditional answer is to pre-fund: park your own money in a handful of strategic accounts covering the corridors that matter most, size it to peak demand rather than average flow, and accept the yield you give up as the cost of being able to settle at all. As a consequence of the stablecoin sandwich model, new solutions are emerging.
The goal of this post is to analyse the two families of new solutions emerging – the hybrid one and the stablecoin-native ones – and to draw a conclusion on the most likely evolution of this domain.
Hybrid solutions
The first family of solutions starts from a simple idea: if the problem is that capital has to sit in the corridor, then make sure it is not your capital. Every player in this group does the same thing: it replaces the money a payment company would have parked in advance with money somebody else advances at the moment it is needed, and it charges for the difference. The stablecoin speeds up the delivery, but it doesn’t radically change the model. It is still based on a credit relationship, and that is why I call these solutions hybrid: onchain settlement relying on a traditional lender-borrower structure underneath.
Arf is the cleanest expression of this solution. It is a Swiss entity providing unsecured, short-term working capital in USDC to licensed financial institutions so they can settle same-day without pre-funding the corridor. There are two products: a fixed credit facility, with an origination date, a repayment date and a negotiated interest rate, and an on-demand liquidity line. The capital behind them comes from Huma – an infrastructure protocol which merged with Arf in 2024. Huma is essentially a venue where lenders deposit in polls mostly stablecoins and earn a yield, and Arf’s assets are pledged against those pools.
Arf has originated so far over $17b in loans with a solid acceleration from January 2026.
Mansa arrives at the same place from a slightly different direction. A borrower submits a request specifying amount, corridor and tenor; Mansa validates it against credit limits, compliance rules and past utilisation; if approved, the request becomes a loan and the funds are disbursed in USDT into a custodial wallet Mansa controls in the borrower’s name. The borrower then draws down into its own wallet, and repays on a waterfall where interest is settled before principal, which restores the available credit. The qualifying borrowers are exactly the counterparties you would expect at the edge of the sandwich: licensed money transmitters, banks with cross-border services, fintechs holding payment licences, established remittance providers.
Where Mansa genuinely differs from Arf is not the mechanism but underwriting logic. Arf lends unsecured, underwriting the institution. Mansa seems to underwrite the transaction: every drawdown has to be justified with a utilisation intent, then evidenced with proof of utilisation – a PSP settlement record showing the money was applied to a real payment – and with proof of payment, meaning evidence of the incoming fiat from the borrower’s own customers that will fund the repayment. In other words, Mansa is lending against the same receivable that the payment company was going to receive anyway, and it wants to see it. That is a very old form of credit, wearing new clothes.
Zynk never uses the word credit, but Transporter, its liquidity product, is a borrowing facility: stablecoins delivered on demand, repaid later with a fee, and charged until the position is closed. Neither the pricing nor the source of the capital is public, so I cannot really say whose balance sheet is at risk.
None of these solutions removes the need for somebody to hold capital in the corridor. They change who holds it, and turn the cost of holding it from foregone yield into an interest rate.
Stablecoin-native solutions
The second family does something different: instead of financing the capital inventory, it removes it.
The most interesting solution is Avenia, a Brazilian company facilitating FX transactions with LATAM. Avenia also issues BRLA, a real-denominated stablecoin backed 1:1 by audited reserves held in Brazil, and uses it to settle the local leg into Brazilian rails directly.
The mechanism is visible in their own API. Every operation is a quote followed by a ticket: the quote fixes price and fees, the ticket executes against it. Take the off-ramp direction – BRLA on Polygon in, reais out by PIX. The ticket carries a signed permit letting Avenia pull the tokens from the wallet the moment the quote is accepted, and a beneficiary Brazilian account for the payout. With one single call, tokens onchain at one end, reais landing in a local bank account at the other. There is no separate off-ramp step, because the redemption is the PIX.
The ramp is essentially collapsed into a single instruction, since the same quote can take USDC in and deliver reais out. And the local leg is cheap, because minting and redeeming is not an FX trade: on Avenia’s published example, converting between 100 reais and BRLA costs about 25 centavos all in, with the conversion fee at zero. The FX exposure shrinks accordingly, from a holding period (that could last weeks in exotic corridors) to a quote window of roughly ten minutes. Essentially, a payment company reaching Brazil does not need to source reais and hold them, nor borrow them from anyone. It mints against reserves and redeems at par. There is no tenor, no rate, no repayment and nothing to underwrite, because there is no loan.
This is the model I have been pointing at for a while, and in theory it is the most efficient. But it is really early, and the numbers say so: BRLA circulates around 136 million tokens, roughly $27 million, a rounding error next to the dollar stablecoins it is meant to complement. More telling than the size is the routing: Avenia’s own documentation, describing what happens when a transaction fails halfway, gives the internal path for a PIX-to-dollar payment as PIX to USDC to USD – with no BRLA leg in the middle. So even at Avenia, the local stablecoin is not always what closes the FX loop.
Conclusions
Pre-funding an account was always a sort of credit in disguise: either your own equity immobilised in a nostro, or a bank line funding it. What most of these players have done is make it explicit, short-dated and recyclable, so the same pool turns over in days instead of sitting idle against peak demand. That is a real gain in capital efficiency but the friction was not eliminated: the existing model was made more efficient, not replaced by a new one.
And what the hybrid solutions finance is generally the dollar leg. None of them puts reais, naira or pesos into the corridor: that last step still largely depends on somebody at the far end holding local currency and being willing to exchange it. The credit refinances the sender’s side of the trade and pushes the local holding one party further down the chain.
As stated multiple times before, a well adopted local stablecoin would remove the need for credit altogether. But a local stablecoin is slow to build, because it has to be issued, audited, licensed and, more importantly, you have to start building organic liquidity through real use cases and acceptance. That’s not an overnight task.
For this reason, what I expect is sequencing rather than competition between the two models. The hybrids will carry the volume for as long as local stablecoins stay thin, and that will not be a short wait.
The need for local liquidity in FX transactions will not disappear instantaneously: it will disappear at the speed it takes to build a local stablecoin from scratch, one jurisdiction at a time.
