> ## Content Index
> Fetch the complete content index at: https://www.dailyferment.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Stablecoin FX went below interbank. The cost moved into routing.
- URL: https://www.dailyferment.com/stablecoin-fx-routing-cost-provider-churn/
- Published: 2026-09-21T04:00:00.000Z
- Updated: 2026-09-21T04:00:00.000Z
- Description: The median stablecoin payment now prices below interbank. The cost did not disappear, it became a routing decision, and most payments systems cannot see it.
- Author: Rana Bilal Zafar
- Tags: Fintech, Engineering, #long-argument

On 13 July 2026, a payments firm called Borderless published its quarterly benchmark of stablecoin foreign exchange rates and reported something that should not happen in a cross-border payments market. Across 260 corridors in 108 countries, the median stablecoin payment in the second quarter priced at [negative 3.2 basis points against interbank](https://www.theblock.co/post/408006/stablecoin-fx-priced-below-interbank-rates-in-q2-with-routing-now-the-biggest-cost-lever-borderless?ref=dailyferment.com), reaching negative 5.9 in June. Not close to the rate the banks quote each other. Below it.

I spent a week trying to work out where the cost went, because costs in payments don't evaporate. They move. And this one moved somewhere I've never thought to look, in systems I built myself.

## What a negative spread is actually telling you

A year ago the interesting question about stablecoin FX was whether it could get close to interbank at all. In the first quarter of 2026 the answer arrived: Borderless found LATAM corridors trading [within roughly 22 basis points of interbank](https://www.theblock.co/news/markets/2026-04-09-stablecoin-fx-nears-institutional-grade-parity-with-bank-rails-in-latam-and-east-africa-report-396866?ref=dailyferment.com), Brazil at zero across multiple providers, and provider gaps in Kenya, Tanzania and Rwanda narrowing by 60 to 80% over the quarter. Fourteen of 21 tracked currencies sat inside 100 basis points of interbank by March, from more than 1.1 million rate observations.

Then the spread went negative, which sounds like a rounding artefact and is not. It happens because the two sides of the comparison are not the same trade. Interbank is a price between banks with balance sheet costs, capital charges and settlement windows. The stablecoin quote is a price between competing liquidity providers fighting for volume on a rail with no settlement window at all. When enough of them fight on the same corridor, the quote can sit under the benchmark, and for a while it did.

The obvious reading is that cross-border FX has been solved and the fee went to zero. That reading is wrong in a specific way, and Borderless, to its credit, says so in the same report.

## Stablecoin FX routing is where the cost went

![A red street sign with white lettering reading money exchange](https://storage.ghost.io/c/d7/b7/d7b77195-48c5-45ab-8688-51750879d1d6/content/images/2026/09/df5-exchange.jpg)

The visible price of changing money has almost stopped mattering. Photo by Youssef Mubarak on Unsplash.

The same benchmark found that provider selection is now the dominant remaining cost. A business routing everything through a single provider paid about **$2,330 more per $1 million** than a business taking the best available price at the moment of payment. Borderless calls this the Routing Tax, and since Borderless sells routing, that framing is a product pitch as much as a finding. The underlying number is still the number.

What stopped me was the churn rather than the size of the gap. On Brazilian real, **the cheapest USDT provider changed 34 times in 88 days.**

Read that again, because I did. Not the cheapest rate changed 34 times. The cheapest counterparty did, on one corridor, in under three months. Every two and a half days, somebody else was winning.

Regional dispersion tells the same story from another angle. In the second quarter, Asian corridors sat at 6.1 basis points of spread while Latin America compressed to 89.0 and Africa widened by 166 basis points to 512.8\. The median across everything was negative. Almost nobody pays the median, which Borderless states plainly: no single payer pays it, because what you pay depends on your corridor, your size and the provider you happened to pick.

The global median went below interbank. Almost nobody pays the median.

Stablecoin FX spread against interbank, second quarter 2026, in basis points

Global median, all 260 corridors

minus 3.2

Asia

6.1

Latin America

89.0

Africa, after widening 166 points in the quarter

512.8

One hundred basis points is one percent. The headline number and the African number describe the same market in the same quarter. Source: the Borderless Benchmark, second quarter 2026, as reported by The Block.

## Meanwhile the flow is moving the rate underneath you

There is a second thing happening that has nothing to do with provider pricing, and it is the part with real academic weight behind it.

In March 2026 Iñaki Aldasoro, Paula Beltrán and Federico Grinberg published [Stablecoin flows and spillovers to FX markets](https://www.bis.org/publ/work1340.htm?ref=dailyferment.com), released jointly as BIS Working Paper 1340 and [IMF Working Paper 2026/056](https://www.imf.org/en/publications/wp/issues/2026/03/27/stablecoin-inflows-and-spillovers-to-fx-markets-575046?ref=dailyferment.com). One paper, two mastheads. I'm labouring this because I've seen both citations stacked in the same paragraph as though they corroborated each other, and they don't.

The study covers four dollar-pegged stablecoins, 27 fiat currencies and 64 exchanges from 2021 to 2025\. Its headline result is that a 1% increase in net stablecoin inflows raises parity deviations by 40 basis points. The same shock depreciates the local currency in ordinary spot markets and widens the dollar premium in synthetic funding markets. Effects concentrate in emerging markets, where arbitrage between the stablecoin venue and the bank venue is weakest. The authors estimate that halving the frictions between the two markets would cut the funding spillover by about half and the exchange rate effect by nearly a third.

So the corridor you are paying through is not a passive pipe. Aggregate flow through it moves the very rate your price is measured against, and the effect is largest exactly where the spreads are widest and the arbitrage thinnest, which is also where most of this volume actually goes.

## What this looks like from inside an engineering team

![Yellow and green fibre optic cables neatly connected in a network cabinet](https://storage.ghost.io/c/d7/b7/d7b77195-48c5-45ab-8688-51750879d1d6/content/images/2026/09/df5-cables.jpg)

Routing is a solved problem one layer down and an unasked question one layer up. Photo by Albert Stoynov on Unsplash.

I've built settlement and payment monitoring across chains, and the design error I kept running into was never throughput. It was treating a counterparty as a constant.

Think about how a payout integration gets written. Someone in commercial picks a provider. Someone in engineering writes an adapter against that provider's API. We are all disciplined about the things we were taught to make configurable, so the base URL goes in config, the API key goes in a secret store, the retry policy gets a backoff, the webhook gets a signature check. Everything that might vary by environment becomes a variable.

The provider does not. The provider is the thing the adapter is named after.

That was a correct decision for most of the history of the craft, because vendor pricing moved on contract cycles. You renegotiated annually and the difference between providers was a procurement question, settled once a year by people with a spreadsheet. A build-time constant is the right way to model something that changes annually.

Thirty-four changes in 88 days is not a procurement cycle. It is a market. And when the best counterparty rotates every two and a half days, an integration written against one of them stops being a neutral choice. It decays, quietly, in a direction nobody picked.

The second half of the problem is worse, and it is the part I want engineers to actually take away. **Nothing in the system can report the loss, because the ledger has no field for the price you did not get.**

Open any payments schema and look at what a payout row stores. Amount sent. Currency. Provider reference. Chain and transaction hash. Status. Fee, if the provider itemises one. Settled amount in local currency, as reported back by the provider. Every one of those is a fact about what happened. Not one of them is a fact about what was available at the same moment somewhere else, because at the time the schema was designed there was no meaningful somewhere else. The rate was the rate.

Which means the Routing Tax is invisible by construction. It does not show up as a fee line, because it is not a fee. It does not show up as a variance, because there is no benchmark stored to vary from. If your CFO asked tomorrow what single-provider routing cost the company last quarter, most of the payments systems I've worked on couldn't answer. Not because the answer is bad. Because nobody ever wrote the number down.

## The claim I will defend

**A choice made once against a market that keeps repricing is a position you are still holding, and a system built to record outcomes cannot tell you what that position costs.**

The two halves matter together. Plenty of people have noticed that vendor prices move. The engineering point is that nobody blundered here. Recording outcomes rather than alternatives is the default shape of every ledger, and it's the correct shape right up until alternatives become a live question. The moment they become one, the system keeps working perfectly and stops being informative, and nothing about it looks broken.

This travels well beyond payments. Instance types, reserved capacity, CDN, egress, the region you deployed in six years ago. All picked once, all repriced constantly since, all recorded as what you spent rather than what you could have spent. The version with the fastest clock right now is model routing, where the price and capability ranking of the frontier models has changed several times a quarter and most applications still have exactly one provider name compiled into them.

I would separate this from the argument I made about [redemption queues](https://www.dailyferment.com/staked-ether-etf-exit-queue-redemption/), because the two are easy to confuse and they are not the same failure. That one was about capacity that's genuinely adequate right up until the day you need it. This one is about a decision that was correct when it was made and has been quietly sliding ever since. Different failure, different fix.

## Prior art, because someone got here first

In April 2026 Lisk published [a piece arguing that stablecoins relocate FX exposure rather than remove it](https://lisk.com/blog/posts/stablecoins-dont-solve-your-fx-problem-they-move-it/?ref=dailyferment.com), identifying conversion into stablecoins, reconversion out, and pricing mismatches as the three points where exposure survives, and citing FXC Intelligence's estimate that 70 to 80% of cross-border cost sits in spread, routing and timing rather than headline fees. That relocation argument is theirs and I am not going to restate it as a discovery.

PYMNTS made the adjacent macro case in 2026, calling stablecoin conversion [a shadow FX market that has become a corporate treasury issue](https://www.pymnts.com/news/b2b-payments/2026/stablecoins-shadow-fx-market-is-becoming-a-corporate-treasury-issue/?ref=dailyferment.com), noting that about 70% of stablecoin demand originates outside the United States and therefore involves a currency conversion by definition.

I have also argued before that the headline stablecoin volume figure banks size their builds against is [measuring something other than what they think](https://www.dailyferment.com/stablecoin-volume-wrong-number-side-core/). That one was about volume and this one is about price, and I notice I keep arriving in the same place from different directions. Either that is a property of this market or it is a habit of mine, and I have not worked out which.

What I have not found anyone writing is the schema argument. The treasury framing asks whether finance teams are tracking this. The engineering framing asks whether the systems could tell them if they tried, and in most cases the answer is no.

## How the corridor repriced, in order

**2021 to 2025.** The period covered by BIS 1340\. Stablecoin flows are already large enough to move spot rates and funding spreads in emerging markets, well before the payments industry treats them as infrastructure.

**Q1 2026.** The Borderless Benchmark launches as an open reference for stablecoin to fiat rates, described by its authors as v0.1 in open beta. LATAM corridors come within about 22 basis points of interbank, Brazil hits zero, East Africa provider gaps compress 60 to 80%.

**27 March 2026.** The BIS and the IMF publish the spillover paper, putting a number on the transmission channel: 1% net inflow, 40 basis points of parity deviation.

**April 2026.** Lisk publishes the relocation argument from the treasury side.

**Q2 2026.** The median goes below interbank at negative 3.2 basis points, negative 5.9 in June. Regional dispersion pulls apart: Asia 6.1, LATAM 89.0, Africa widening 166 points to 512.8\. Provider identity on Brazilian real changes 34 times in 88 days.

**18 August 2026.** Treasury publishes its [Notice of Proposed Rulemaking under the GENIUS Act](https://www.federalregister.gov/documents/2026/08/18/2026-16796/genius-act-regulations-on-payment-stablecoin-issuance-offer-and-sale?ref=dailyferment.com), at 91 FR 53368, defining who may issue, offer and sell payment stablecoins in the United States, with a safe harbour route for foreign issuers who register with the OCC and control for user location.

## Where this stands this month

The GENIUS Act comment period closes on 19 October 2026, so the rules deciding which issuers can legally serve US-connected flow are being written right now, and the provider set on a given corridor is one of the things they will reshape. A routing layer that treats its provider list as fixed is about to be wrong for a second, entirely separate reason.

Now the part that cuts against me. Africa's spreads widened by 166 basis points in the quarter to 512.8\. If dispersion is widening rather than compressing in the region with the most need, then the story that competition has driven the visible price to zero is at best regional, and the corridors where routing matters most may simply be corridors where nothing has converged yet. A negative global median can coexist with a five percent spread in the place you actually pay.

I also want to be careful about the source. The Borderless Benchmark is a beta dataset published by a company whose product is provider routing, and the finding that provider routing is the biggest remaining cost lever is a finding that sells the product. The rate observations look serious and [the methodology is public](https://borderless.xyz/insights/introducing-the-borderless-benchmark-bringing-clarity-to-stablecoin-fx-rates?ref=dailyferment.com), but I am not treating it as neutral, and there is no second independent benchmark of stablecoin FX rates I could check it against. That is a real gap in this piece.

## Marking my confidence

**Established.** The Q1 and Q2 Borderless figures, as reported: 22 basis points in LATAM, zero in Brazil, 60 to 80% compression in East Africa, negative 3.2 median in Q2, negative 5.9 in June, 260 corridors, 108 countries, the $2,330 per million single-provider gap, 34 provider changes in 88 days, and the regional numbers. The BIS and IMF results, from the paper itself: 40 basis points per 1% net inflow, four stablecoins, 27 currencies, 64 exchanges, 2021 to 2025, and the friction counterfactual. The Treasury rulemaking date, citation and comment deadline.

**Not original.** That stablecoins move FX exposure rather than eliminate it was argued by Lisk in April 2026\. The shadow FX framing is PYMNTS. The term Routing Tax is Borderless's, and so is the observation that provider selection is now the dominant cost.

**Inferred.** That most production payout integrations treat the provider as a build-time constant, and that most payments schemas store no counterfactual rate. This comes from the systems I have built and reviewed, which is a real sample and not a representative one. I would expect the largest payment companies to be exceptions, since best execution logic is exactly what they sell.

**Guess.** That within two years, the rate available at the moment of payment against a published benchmark becomes a standard stored field in payout schemas, the way we eventually started storing the FX rate on a multi-currency invoice instead of only the converted total. I would not put much weight on the timing.

What would change my mind: a benchmark independent of any routing vendor showing that provider churn is much lower than 34 changes a quarter on major corridors, or evidence that dispersion on the corridors carrying real volume is small enough that the routing decision is not worth instrumenting.

## What I would actually check

Here is the test, and I am running it on my own systems first. Pick one corridor. Take last quarter's payouts. For each one, ask what rate was available somewhere else at that timestamp.

If that's a query, you're fine. If it's a project, then your provider may have been excellent or terrible for three months and your systems were never going to mention it either way.

I don't have a tidy closing line for the version of this that isn't about payments. I've been turning it over for a week and mostly it has just made me uneasy about two or three choices I made in 2021 and have not looked at since.