The Bridge Between the Bridges

Six tokenized deposit networks are forming in the US, three of the same banks sit inside two of them, question is, who clears between them.

Jun 21, 2026

Some of the banks building the new big bank network call it the bridge. Others call it the chain. According to the Wall Street Journal, JPMorgan, Citigroup, Bank of America, and Wells Fargo are putting a shared tokenized deposit network inside The Clearing House, the bank owned operator of CHIPS and RTP, with a target launch in the first half of 2027. The technology provider has not been chosen. The access model is undecided. But the intent is clear enough, keep deposits inside the regulated banking system while giving large clients the 24/7, programmable, atomic settlement the stablecoin issuers have been selling for two years.

It is not the only network forming. The Cari Network is six regional banks, Huntington, First Horizon, M&T, KeyBank, Old National, and SouthState, building shared tokenized deposit infrastructure on ZKsync’s Prividium. Project Keystone is six banks on FIS’s Lyriq platform, with Citizens and Fifth Third alongside a familiar looking roster, and a plan to spin out into a bank owned entity its own people compare to SWIFT. Hazel Network is Vantage Bank and Custodia, building a community bank token that is an insured deposit inside the network and converts to a GENIUS compliant stablecoin when it leaves. Above all of them sit the rails the largest banks built first, JPMorgan’s Kinexys, which has moved more than $3 trillion in cumulative notional since 2020 and runs over $5 billion a day, with its JPMD deposit token live on Base and migrating to Canton, and Citi Token Services, live in dollars and euros across five money center jurisdictions.

It is tempting to read all of this as the banks finally answering the stablecoins, and there is real substance under it. The money center rails are in production and moving institutional volume, Kinexys has years of live settlement behind it, and the regulatory ground firmed up when the GENIUS Act became law in July 2025. But be precise about how early most of the rest is. Cari is pre-launch, its pilot only slated for later in 2026. Keystone went live this spring with a handful of banks doing simple transfers and net settlement, nothing programmable yet. The issuance side is real, but outside a few large banks it is closer to a starting line than a finished system.

All of them share the same blind spot. Each of these networks I’m sure settles beautifully inside its own walls. A Cari bank pays a Cari bank. A Keystone bank pays a Keystone bank. None of them answers the question that decides whether this becomes infrastructure or stays a row of expensive silos, what happens when a tokenized deposit at a Keystone bank has to settle against a tokenized deposit at a Cari bank, and the two sit on different ledgers under different rulebooks? Everyone is building a bridge. Nobody is building the bridge between the bridges. That gap really is the whole game, and the way it closes is probably not the way most people assume.

No one wins the war they are all fighting

The standard reassurance is that the fragmentation is temporary, that it tidies up once a winner emerges and everyone settles on it. I think that might have it backwards. Every one of these networks is fighting the same war, to become the single rail the others have to plug into, and that war has no winner. But no winner at the network layer is not the same as no winner anywhere. Someone is going to win this, and win it enormously. They are just going to win a different layer than the one everyone is fighting over, and you cannot see which layer until you see why the network war cannot be won.

Start with why every bank builds its own rail in the first place. A bank’s deposits are its cheapest funding and the base of its franchise, so the instinct when a new payment technology arrives is to build something that keeps those deposits moving inside the bank’s own perimeter. The more a bank spends on its own rail, the more it needs that rail used, and the less interested it is in anything that makes its deposits freely interchangeable with a competitor’s. That instinct does not produce one shared network. It produces JPMorgan’s rail, Citi’s rail, and a consortium for everyone too small to fund one alone. And most banks are too small. JPMorgan spends roughly $15 billion a year on technology, and of the nearly 4,000 banks in the US, the overwhelming majority have an entire expense base smaller than that single line item. The regionals did not pool into Cari and Keystone because they wanted to share. They pooled because they cannot out build the money center banks alone. That really is the barbell hardening, not a market converging.

A shared network only has value once enough banks are on it, which means every participant is waiting to see whether anyone else commits before committing themselves. Look at what the sophisticated players actually do. KeyBank, M&T, and Huntington are founding participants in Cari. KeyBank, M&T, and Huntington are also in Project Keystone. The same three regionals joined two competing networks, on two different technology stacks, at the same time. The easy reading is that they cannot make up their minds. The right reading is that they have made it up precisely. In a market where no network has tipped, you keep a foot in each camp so you are never stranded on the loser. Count the integrations, not the memberships. When the smartest participants refuse to be exclusive, they are not being indecisive, they are telling you the market will not tip to a single winner.

That is not a problem to be fixed. Consolidation onto one network was never the answer, and it was never even desirable, because a single bank owned network everyone had to settle through would be a tollbooth no one could route around. The interesting outcome was always going to be fragmentation at the network level, held together by something above it. Payments has been here before and built that something more than once, and not the way the famous example suggests.

Share Atomic Settlement

What payments history actually shows

Reach for a precedent and nearly everyone reaches for the same one, so let’s be exact about what happened with it. Bank of America launched BankAmericard in Fresno in 1958 as a single bank’s product. It licensed the program to other banks from 1966, and in 1970 handed it to National BankAmericard Inc., a not for profit association the member banks owned jointly, which renamed itself Visa in 1976. For its first three decades Visa was not a company in the sense people now use the word. It was a cooperative of competing banks, member owned, non stock, governed by the simple rule of cooperate on the rail, compete on the products. Visa was never a bank. It was the thing the banks built so they would not have to trust each other one by one.

Then the part I think everyone forgets too easily. In 2008 Visa went public, two years after Mastercard did, in what was then the largest IPO in US history. The cooperative the banks had built became an independent, for profit company that no longer answered to them. The banks kept issuing the cards and earning the interchange, but the network in the middle, the mark that makes any bank’s card clear at any merchant, was now its own master, worth more than almost any single bank it serves. That is what “Visa won” actually means. The banks won the issuing business, which fragmented across thousands of them. The shared acceptance layer they jointly created won everything else, and then floated away from them. This is worth holding onto, because the banks are building shared acceptance layers again, and Keystone is already planning to spin its network out into a separate entity its own people compare to SWIFT.

The bit that should make anyone cautious about the Visa analogy, though, is in US payments, Visa is the exception, not the rule. Most of the time fragmentation does not resolve into a winner at all. The ACH network moves tens of trillions of dollars a year and has no Visa. NACHA writes the rulebook and operates nothing. The switching runs through two competing operators, the Federal Reserve’s FedACH and The Clearing House’s EPN, and when a payment crosses between them one operator simply hands it to the other. Final settlement happens in central bank money at the Fed. There is no network mark, no interchange, no profit stream, and no winner, because ACH was built to cut cost rather than enable profit, and a system built to cut cost stays a thin utility. Glenbrook, whose book is the standard text on this, calls that the difference between thick networks and thin ones, and the key point is that both kinds were bank owned. Cards went thick because interchange handed them a tax to fund a brand. The ACH stayed thin because it had no such tax. The economics decided which became a Visa and which became plumbing, not the technology.

The same split runs through the rest of the system. Large value payments run on two rails that coexist rather than consolidate, the Fed’s Fedwire and the bank owned CHIPS, and both settle finally in reserves at the Fed. Foreign exchange settlement, the closest structural cousin to the problem in front of us, runs through CLS, a neutral utility the banks own jointly that settles both legs of a trade at the same instant so neither side is left holding a claim that fails. It exists because in 1974 a German bank named Herstatt was closed mid trade and took a chain of counterparties down with it, and the industry decided the cure was a shared settlement utility nobody competed with. Two token networks settling against each other is the Herstatt problem in brand new clothes, and CLS is what the answer has looked like for fifty years, ie not a winner, a utility, settling in central bank money.

And when none of that gets built, fragmentation simply persists. The US has two instant payment rails, The Clearing House’s RTP, launched in 2017, and the Fed’s FedNow, launched in 2023, and they do not interoperate at all. A bank connects to one, the other, both, or neither. The community banks largely would not join RTP, because the largest banks own it, and waited for the Fed to build a neutral one. Nine years on there is still no bridge between the two, because nothing forces one. That is not a transitional state, it is the equilibrium when no neutral layer is built and no rule compels interoperability, and it is exactly where tokenized deposits sit today. You cannot send a JPMorgan dollar into a Citibank account. The rails exist. The bridge does not.

The prize is a layer, and it can be built two ways

So there is a valuable layer here, the acceptance layer that makes a dollar a dollar regardless of which network issued it, and I think history says it gets built in one of two shapes. Thick, like Visa, a for profit network funded by a fee on every transaction, which over time becomes the most valuable seat in the system and stops answering to the banks that built it. Or thin, like the ACH and CLS, a neutral utility funded to cut cost, settling in central bank money, that nobody wins because there is nothing to win except lower unit costs. Which shape it takes turns on the same thing it turned on for cards, whether there is a profit stream the layer can tax.

A proprietary deposit token is a real asset for the bank that issues it. JPMD lets JPMorgan’s clients do things they could not do before, and because it is a deposit and not a stablecoin it can legally pay interest, which the biggest banks will use to defend their corporate balances. But a single bank’s token is not durable at the level of the system. The moment a neutral acceptance layer exists, the token can be routed around, the way an issuing bank’s brand disappeared behind the Visa mark. And it cannot deliver singleness of money on its own, because singleness, the plain fact that a dollar is a dollar regardless of which bank stands behind it, is a property of the layer above the issuers, not of any one issuer.

That layer is the thing worth owning, and the asset that controls it is not a technology, it is the rulebook. It feels like interoperability will be decided less by a technical protocol than by which legal rulebook collects enough signatures to make bank issued tokens fungible at scale. That is the cornered resource, the asset everyone needs and only one body holds, and it is what separates a strategy here from a wish. I I’m just think “picking the winning token” or “joining the right consortium” is a bit of a wish. Owning a piece of the layer that clears between tokens, the registry of who may issue, the rail that turns any inbound bank token into par-value cash, the settlement asset the cross network leg lands in, is a strategy. Tony McLaughlin, who built Citi’s institutional token business, left to build exactly the clearing piece at Ubyx, and Barclays put money into it in January 2026. His bet is that clearing can carry a fee, a few basis points against the card networks’ two to three percent, which I think is a bet this layer can be built thick. The Fed and utility path is the bet that it stays thin. Both are live, and which one wins is really the game in town.

Who gets to build it

If the thin, neutral version is the one that wins the whole market, then ownership decides everything, and the instant payments story already told us how this goes. RTP is good infrastructure, and the community banks would not join it, because the largest banks own it. They waited for a Fed operated rail and went there instead. A network owned by the money center banks does not win the long tail, no matter how good it is, because ownership is a signal everyone who is not an owner reads correctly. Now look again at the big bank tokenized deposit network forming inside The Clearing House, same operator, same ownership, probably the same signal. It may well become the settlement layer for the banks that own it, which is a large slice of the system. It may be challenging to become the acceptance layer for all 4,000 for the reason RTP did not.

I do wonder where the Federal Reserve is on all this. Governor Waller has been explicit that the Fed has dropped the language of a wholesale CBDC, because bank reserves are already a digital wholesale settlement asset and the work is to tokenize the instrument that exists rather than mint a new one. That is the missing floor. The cleanest way for a Cari token and a Keystone token to settle against each other with finality is for both legs to land in a common tokenized reserve, which is the Fedwire role and the CLS principle rebuilt for tokens. Project Agorá, the BIS project that includes the New York Fed and more than forty private banks, demonstrated precisely that and moved to real value testing in late May 2026, settling tokenized central bank reserves against tokenized commercial bank deposits across seven jurisdictions, atomically.

Stand back and the endgame stops looking mysterious and starts to look like every other payment system that grew up. Carlota Perez calls this moment the turning point, when a technology stops throwing off incompatible experiments and starts getting wired into the dull connective tissue that lets it run at scale. The six networks and the multi homing are the experiments. The turning point is the unglamorous work that ends them, a tokenized reserve as the neutral settlement asset, a thin rail that clears between issuers, and a rulebook with enough signatures to bind. Project Agorá moved to real value testing in late May 2026, settling tokenized central bank reserves against tokenized commercial bank deposits across seven jurisdictions, atomically. That is the floor being poured, the way it was poured for checks, for the ACH, for wires, and for foreign exchange. Every time but cards.

The proprietary rails survive that world, the way thousands of banks kept issuing cards long after the association they built floated away from them. Kinexys keeps running its $5 billion a day. JPMD keeps paying the interest a stablecoin legally cannot. What changes is that these tokens interconnect through a neutral layer instead of straining to be it. Issuance fragments, acceptance consolidates, and the only live question is the one that decided cards fifty years ago, whether the layer doing the consolidating is a thin utility the banks share or a thick toll they build and then hand to someone else. The thick version can carry a fee, a few basis points against the card networks' two to three percent, which is reason enough for someone to try to build it. The thin version, funded to cut cost and settling in central bank money, is the one nobody wins because there is nothing to win but lower unit costs.

Because the banks figuring out which network to join are fighting over the issuing side of a question that gets settled on the acceptance side. They are one layer too low. The layer above them, the one that turns any bank's inbound token into par value cash, is the one with the moat, and nobody owns it yet. That is the bridge between the bridges. Right now it is sitting there unbuilt.

References

The US tokenized-deposit networks

GSIB rails

Payments system precedents

Benson and Loftesness, Payments Systems in the U.S., Glenbrook Press, 2nd edition (2014). The open loop / closed loop and thick vs thin framework.
NACHA: the ACH network, rules, and the FedACH / EPN dual operator model
Federal Reserve: Fedwire Funds Service and the National Settlement Service
The Clearing House: CHIPS and RTP
CLS Group: payment-versus-payment settlement for FX

The settlement fabric and the neutral layer

Regulation

Atomic Settlement

Discussion about this post

Ready for more?