PRIM3 Brief #21: The On-Chain Dollar Split in Two, and Founders Are Building for the Wrong Half
On September 5, 2026, a tokenized deposit network called Cari closed a $32.5 million round in which every single investor was a bank. First Horizon, Huntington, KeyBank, M&T Bank, Old National, SouthState, and Montana's Glacier Bank put the money in. About 30 banks have joined the network and roughly 40 more are in active talks, representing institutions with more than $10 trillion in combined assets, according to Ledger Insights.
That was the same week the total stablecoin float sat at $301.7 billion and had shrunk 1.5% over the prior 90 days, per Stablecoin Beat's tracker. Two numbers, one direction of travel. The dollar on-chain is no longer one market with different wrappers — it's two networks with different owners, different economics, and different rules about who is allowed to issue.
Most of the coverage filed this under "banks are finally adopting crypto." That read is comfortable and wrong. Banks aren't adopting the stablecoin stack. They're building the thing that keeps them from ever needing it.
What Cari Actually Is, and Why the Charter Matters
A Cari token records a US dollar deposit held at a participating chartered bank on a blockchain ledger. The deposit stays on the bank's balance sheet, stays FDIC-eligible, keeps paying interest, and never leaves the regulatory perimeter that bank already operates inside. That is the whole design brief, per Fireblocks, which built the custody layer.
Compare that to a stablecoin. USDT and USDC are bearer claims on an issuer, backed by a reserve pool of T-bills and cash held outside the banking system's credit-creation machinery. USDT holds $183.3 billion and USDC $73.6 billion of the $301.7 billion market as of September 3, 2026, roughly 60% and 24% respectively. We wrote about that concentration in Brief #10 and nothing about it has improved since. And under the GENIUS Act (Public Law 119-27), those issuers can't pay a yield to holders either, which removes the one lever an issuer would otherwise pull to defend share against a bank product that pays a policy-linked rate by construction.
So a US corporate treasurer choosing between them isn't choosing between two flavours of the same instrument. One pays interest, sits inside deposit insurance, and comes from the bank they already have a credit relationship with. The other doesn't, isn't, and doesn't.
Here's the detail that keeps getting missed: Cari runs on ZKsync, a public chain. So the perimeter isn't the chain at all. What draws the boundary is the liability, and a public ledger underneath a closed membership network changes nothing about who is allowed to mint on it. Founders who read "regional banks deployed on a public L2" as total-addressable-market expansion are reading the block explorer when they should be reading the balance sheet.
Why the Mainstream Read Is Incomplete
The consensus story of 2026 has been convergence — TradFi and crypto meeting in the middle, everyone eventually settling on shared rails. What's actually happening is a fork, and it's the second one this year.
In Brief #14 we argued that institutional tokenization concentrated behind Canton rather than the transparent public L1s that dominate the RWA charts. Assets split first. Now the money is splitting too, and JPMorgan just stitched both halves into one system by deploying JPM Coin onto Canton: deposit token, private chain, a closed loop running from issuance through settlement without ever touching a venue anyone reading this can access. Kinexys has processed more than $4 trillion cumulatively, with average daily volume above $7 billion as of June 2026, per JPMorgan.
Now compare the growth vectors. Stablecoin supply is flat-to-down over a quarter. Bank deposit-token volume is compounding off a base that already dwarfs most of DeFi. And it's compounding without any crypto-native distribution at all — no exchange listing, no liquidity mining, no KOL round, no community.
That last point is the one founders should sit with.
The convergence thesis quietly assumed banks would need the crypto ecosystem to get on-chain: they'd need our wallets, our custodians, our liquidity venues, our compliance tooling. Cari's answer is that a bank-governed membership network with Fireblocks underneath and the American Bankers Association as a channel partner doesn't need any of that. It needs other banks. Which it is currently signing at a rate of roughly one every few days.
What the Deal Tape Says
Look at where the money actually went in the first week of September 2026. Ten crypto and blockchain companies disclosed $292.35 million across the week of August 31 to September 5, and the theme was almost monotonous: plumbing between digital dollars and the banking system. Félix took $200 million for cross-border remittances, structured as $87 million of equity led by a16z plus a $113 million credit facility from General Catalyst, per crypto.news.
Set that against the year. Crypto venture has done just over 300 deals in 2026 against about 840 in the same window of 2025, a 64% collapse in deal count, while dollars deployed held at $8.1 billion versus $9.4 billion, down only 13%. That's the barbell from Brief #12, sharper a year on. Fewer companies, bigger checks, and the checks are clustering on rails rather than applications.
But putting these two numbers side by side produces a second conclusion that almost nobody is drawing.
The obvious one: capital thinks payments infrastructure is where the next decade of usage sits. Fine, everyone says that.
The less obvious one: the largest checks are going to businesses that touch both dollar networks. Félix moves value between US bank accounts and Latin American payout rails. At Cari the crossing is bank-to-bank, which is a narrower boundary but a far denser one. Neither of these is a crypto-native application in the 2021 sense, and neither is exposed to whether stablecoin supply grows or shrinks next quarter. They earn on the crossing.
What This Means for Founders
If you're building a payments, treasury, or settlement product in Web3 right now, the practical question isn't which dollar wins. It's whether your product still works when the counterparty's dollar lives on the other side of the perimeter.
Three positions look defensible from where we sit, and one that a lot of decks currently occupy does not.
Defensible: the conversion edge. Somebody has to move value between bank-perimeter deposit tokens and public-chain stablecoins, with attestation, reconciliation, and compliance evidence that a bank auditor will accept without a six-month exception memo. Kima Network, an interoperability protocol in the PRIM3 portfolio that settles between chains and traditional bank rails without wrapping assets or running a bridge, was architected for exactly this shape of problem, and it reads better against 2026 conditions than it did on the day we backed it. Bridgeless settlement stops being a security preference and starts being a hard compliance requirement the moment one leg of the transaction is a regulated deposit, because a wrapped representation of an FDIC-eligible claim is not an FDIC-eligible claim and no bank risk committee will pretend otherwise. That single legal fact rules out most of the interop architectures the last cycle produced. It also explains why the teams closing bank pilots this year are the ones that never took custody of the asset in the first place, which was an unfashionable design choice in 2023 and is now the entire moat.
Defensible: the merchant-side and consumer-side edge. Oobit, also in the portfolio, lets a user spend crypto anywhere Visa is accepted, which is a conversion-edge business dressed as a consumer app. The economics of that edge get better, not worse, when the two networks diverge, because divergence widens spreads and multiplies the number of legs a transaction needs.
Defensible: institutional operations tooling. Crymbo sits on the treasury-operations side, and the operational surface banks are about to need (key management, policy engines, audit trails spanning two dollar systems) is being specified right now by compliance teams rather than product teams. So whoever is in the room while that spec gets written wins the contract.
Not defensible: "we'll be the stablecoin for X." Issuance is closing. GENIUS Act licensing in the US, MiCA enforcement in the EU since July 1, 2026, and now a bank-governed network that structurally admits chartered institutions and not startups. If your pitch is that you'll issue a dollar, you're entering a market where the incumbents own the charter, the interest, and the insurance, and you own none of the three.
There's a version of this advice that sounds like retreat. It isn't. Edges are where margin lives in every mature payment system — Visa doesn't issue currency either.
Where PRIM3 Is Placing Bets
When we underwrite payments and settlement infrastructure this cycle, three things carry the most weight.
First, does the business earn when volume crosses networks, or only when one network grows? Cross-network businesses are hedged against a question nobody can currently answer.
Second, can a chartered bank's compliance team actually deploy this? Not "is it compliant" in the abstract, but whether a named risk officer at a $40 billion regional bank will sign the memo. Most crypto infra fails this at the evidence layer, not the tech layer.
Third, is the team's distribution motion built for institutions or for Twitter? The teams closing bank contracts in 2026 look like enterprise sales organisations with a protocol attached. That's an uncomfortable identity for a lot of Web3 founders and it's the correct one for this sector.
We passed on two dollar-issuance deals in Q2 2026 for the reasons above. Both had good teams. Neither had an answer for what happens when the bank on the other side of the transaction has its own token, its own network, and no reason to route through a startup's.
The on-chain dollar is going to be plural for a long time. Build the exchange between the halves, not another half.
That's where PRIM3 is placing capital this quarter. More on our thesis at prim3.vc.