The Train and the Tracks

After spending the last century regulating the train. The risk is now moving to the tracks.

Jul 12, 2026

Traditional thinking treats money as an object and payments as the rail it travels on. A train and a set of tracks. You can, in principle, lift the train off one set of tracks and put it on another, and it remains the same train.

For digital money, I think this intuition is wrong, and dangerously so. Digital money does not exist outside the systems that operate it. The rail doesn’t just carry the money, it creates and constrains the monetary properties we care about. Par value, redeemability, finality, availability. These aren’t attributes of the token. They’re behaviors of the stack underneath it.

After 25 years inside various parts of banking infrastructure I can’t help think this is the framing the whole tokenization debate tends to miss. We passed the GENIUS Act in July 2025 and built an entire regulatory category around the quality of the train, 1:1 reserves, no rehypothecation, full disclosure, an enforceable right to redeem at par. All necessary. And all of it aimed at the balance sheet, while the risk has migrated to the tracks.

Par value is a property of the stack

I find it really helpful to think about the mechanism, because the mechanism is everything.

When you redeem a stablecoin, three separate systems have to work in sequence. First, a blockchain transaction moves your tokens back to the issuer. Second, the issuer sells Treasuries to raise cash, an off chain trade that settles T+1 and flows through broker dealers with finite balance sheets. Third, the banking system transfers deposits to your account. The token’s claim to be “worth a dollar” is really a claim that all three layers will function, at the same time, under stress.

The GENIUS Act addresses the first order question, whether the assets are there. So solvency, but solvency is necessary and not sufficient. The reserves also have to be convertible into bank deposits fast enough to meet redemptions, and that depends on the market layer working. In March 2020, foreign investors alone sold $417 billion of Treasuries in the dash for cash, and the dealers who intermediate that market were overwhelmed. Pricing dislocations in the deepest, safest market in the world. The stablecoin market today sits at roughly $300 billion, with forecasts of $2 to $4 trillion by 2030. At today’s size, a full scale redemption run would already be a meaningful Treasury selling event. At 2030 scale, it would be a market functioning event, hitting dealers who already operate near their supplementary leverage ratio floors.

And it runs both ways. Fear that the market layer might seize creates the incentive to redeem first, which stresses the market layer. Anyone who has watched a deposit run knows this loop, the expectation of illiquidity manufactures the illiquidity.

Then there’s the third layer, the one it feels like that bankers are least equipped to price, and that’s the technology. Stablecoins are ledger entries in smart contracts on blockchains the issuer does not control. The issuer controls the contract, the upgrade keys, the freeze function. It does not control whether the chain processes transactions at all. And liveness is an economic property. A chain halt suspends the redemption and arbitrage mechanism that holds the peg, the pause itself can create the discount, and the discount feeds the run.

If that sounds a bit theoretical, it isn’t. In October 2025, Paxos mistakenly minted $300 trillion of PYUSD, an internal transfer of $300 million with a fat finger on the zeroes, roughly two and a half times global GDP created by a single account with unlimited mint privileges. It was burned within about 20 minutes and nothing was exploited. But the lesson stands I think, the code path existed, and no reserve requirement in any statute would have stopped it. And in March 2023, when Silicon Valley Bank trapped a portion of Circle’s reserves, USDC traded down to 87 cents in the secondary market. The reserves were overwhelmingly intact, what failed was confidence in the conversion machinery over a weekend. The train was fine. The tracks were closed.

The oldest problem in American banking

If money whose value depends on redemption infrastructure sounds novel, it’s the oldest problem in American banking.

During the free banking era, before the National Banking Acts of the 1860s, the US ran on privately issued banknotes. A dollar note from a solid Boston bank and a dollar note from a distant country bank were both “dollars,” but they did not trade at the same price. Merchants consulted banknote reporters, printed tables of discounts, because a note’s value depended on the cost and credibility of getting it redeemed. The further you were from the issuing bank, the deeper the discount. Par was not a property of the paper. It was a property of the redemption network behind the paper.

New England solved this with infrastructure, not asset quality. The Suffolk Bank system created a clearing arrangement that made member banknotes redeemable at par across the region, and notes inside the network traded at face value while notes outside it did not. Same trains, with better tracks, and different money.

It feels like stablecoins are replaying this dynamic with the labels changed. USDC at 87 cents was a banknote discount, quoted on Coinbase instead of in a Chicago broadsheet. The secondary market is today’s banknote reporter, continuously pricing not the reserves but the redemption stack. I think the free banking era also tells us where this goes. The discounts ended when the redemption infrastructure was mutualized, standardized, and eventually absorbed into a central bank. Which points at the question nobody in the stablecoin debate wants to own, whether issuers eventually need access to the Fed’s balance sheet. The status quo keeps them outside the discount window. Every option that makes them safer pulls them further inside the perimeter.

The neutrality trade

The issuers have understood all of this faster than their regulators, and you can read their conclusions in their infrastructure decisions.

Stripe and Paradigm are building Tempo, a payments first chain targeting 100,000 transactions per second with no native token, gas payable in stablecoins. Circle is building Arc, where USDC itself is the gas token. Tether is behind Plasma and Stable. These purpose built chains exist for one reason, the issuers concluded they cannot accept liveness risk, validator politics, and fee volatility on neutral public chains they don’t control. If the rail defines the money, own the rail.

But look at what they’re giving up. The pitch for public blockchains was always credible neutrality, no single party can censor you, halt you, or change the rules. The purpose built chains trade that neutrality away for control, every one of them run by the issuer or a consortium of the issuer’s friends. Will the market value that trade? I think the honest answer is that nobody knows yet, and the answer probably differs for a corporate treasurer and a crypto native fund.

Meanwhile the incumbents are making the identical trade from the opposite starting point. Citi moves nearly $6 trillion in payments daily, and Citi Token Services now runs tokenized dollar and euro transfers integrated with 24/7 USD clearing, fully embedded inside the bank’s existing stack. Permissioned, controlled, and pointedly not marketed as blockchain at all. The clients using it are moving money on weekends for M&A closings and e-commerce settlement, and most of them never hear the word blockchain. The issuers are building banks’ control models onto crypto rails, the banks are building crypto’s settlement properties into bank rails, and both have concluded the same thing. Control of the track is the product.

The last deterministic act

The other story running through finance right now is AI agents, and on the surface it has nothing to do with stablecoin plumbing, but I’m increasingly convinced it’s the same story.

Every conversation about agentic AI in financial institutions circles the same discomfort. These systems are probabilistic. Ask the same question twice, get two different answers. Ask a room of practitioners who uses AI daily and every hand goes up; ask who would trade a $50 million corporate loan on its output and the hands come down. FINRA is already watching agents hallucinate decade old settlement rules, T+7 in a T+1 world, and is fielding questions about clients bringing their own agents to the firm’s front door. And there’s a systemic version, if institutions run similar models at similar speeds, correlation and speed become new risk variables, and you cannot fight a machine speed failure in human time.

The industry’s emerging answer is to make the guardrails deterministic even when the intelligence is not. Compile regulation into machine checkable rules. Wrap agent actions in verifiable receipts and zero knowledge proofs. Constrain the probabilistic layer inside a deterministic envelope.

Now connect the two halves. As the decision layer of finance becomes probabilistic, the settlement layer becomes the only place where certainty still lives. An agent negotiating and executing payments at machine speed cannot wait for T+1 batch settlement, cannot tolerate a chain halt, and cannot resolve disputes over the phone. It needs settlement that is instant, final, and cryptographically verifiable, because the deterministic settlement record is the ground truth that makes the probabilistic activity above it auditable at all. I’d posit that deterministic settlement is about to become the scarce asset in a probabilistic financial system. That, more than weekend M&A, is what the new rails are for.

Pricing the tracks

Pull the threads together and I think the pattern is hard to miss. The value propositions everyone celebrates, 24/7 availability, instant finality, programmability, live in the settlement layer. The risks everyone worries about, liveness, treasury market bottlenecks, mint bugs, correlated AI models, also live in the settlement layer. The value and the risk have both left the token and moved into the rail, and our regulatory architecture, our diligence frameworks, and most bank strategy decks are still inspecting the train.

For bankers, “should we issue a stablecoin” is probably the wrong question. I think the better question is which settlement stacks your clients’ money will depend on in five years, and what your position is on those tracks. For investors, the companies worth backing are the ones becoming the rail, the settlement, verification, and control infrastructure that everything above it will have to trust, rather than the ones putting another token on somebody else’s.

The free banking era ended with the banks that owned the clearing system setting the terms for everyone else. That’s the epitaph worth writing down now! Money is not an object that rides the rails, the rails are what make it money.

Nobody ever paid much attention to the tracks. That was the point of them. It’s also why they’re about to become one of the most valuable things in finance.

References

Stablecoin plumbing and risk

Purpose-built settlement chains

Incumbent rails

Atomic Settlement

Discussion about this post

Ready for more?