AI bank runs could shrink $5.8 trillion in deposit funding, Dallas Fed warns

I've spent years watching macro risk models quietly assume deposits are sticky. The Dallas Fed just poked a hole in that assumption. Their August 25 analysis doesn't mention Bitcoin once, but it's arguably the most important crypto-adjacent paper a central bank has published this year.

Deposits aren't what they used to be

Here's the core mechanic that keeps the banking system running: banks borrow short and lend long. Depositors can pull their money at any moment, but most never do. Balances sit for years. Rates on those balances barely move even when the Fed cuts or hikes. So banks treat deposits like they're long-duration funding, and they use that stability to hold mortgages, corporate loans, and other illiquid credit.

The Dallas Fed tries to put a number on this. Their model treats deposits as having an effective duration, roughly equal to the weighted average life multiplied by one minus the deposit beta. In plain English: deposits behave like long-term money mostly because people are lazy and rates are sticky.

Now add AI to the picture. Programmable rules and agentic systems could scan rates continuously, compare yields across banks, and move deposits instantly. The Clearing House announced a bank-led initiative back in June 2026 to build tokenized commercial-bank money with automated and agentic-commerce uses. That's the plumbing for this kind of behavior.

Think about what that does. A yield-sensitive deposit that used to sit dormant for four years can now flee within seconds when a competitor offers a better rate. The duration collapses. The stickiness disappears.

The Dallas Fed ran the numbers. On commercial-bank balance sheets as of July 15, they estimate roughly $7 trillion of asset-side interest-rate exposure in 10-year equivalents. About $5.84 trillion of that's supported by the deposit characteristics we're talking about. Those aren't small numbers. That's the funding base of America's credit system.

Their sensitivity cases are what matter. A 10% increase in deposit price sensitivity, assuming a four-year weighted average life, cuts aggregate duration-risk appetite by about $700 billion in 10-year equivalents. A 10% reduction in weighted average life chops modeled maturity-transformation capacity by about $580 billion.

Let me put that in terms anyone can understand. That's roughly seven hundred billion dollars of long-duration assets banks would no longer feel safe holding. Those assets are the loans that finance houses, businesses, and infrastructure.

Who pays when deposits get smart?

Banks have options if deposits become flighty, but none of them are free. They can issue more term debt, but wholesale funding costs more than sticky retail deposits. Those higher costs get passed to borrowers. The Dallas Fed says this directly.

Or banks can hold more reserves and Treasuries against faster, less predictable outflows. That also means less room for illiquid credit. Either way, the cost of credit goes up for consumers and businesses. There's no magic escape hatch.

We already have evidence this isn't hypothetical. The Central Bank of Brazil published a 2025 working paper on Pix, their instant-payment system. Heavier Pix usage increased banks' liquid-asset holdings and reduced liquidity transformation. Pix isn't a direct analog for US tokenized deposits, but it proves the mechanism works.

This is where the crypto connection gets interesting. Tokenized deposits aren't Bitcoin or stablecoins. They're bank-issued digital money on blockchains or similar infrastructure. The initiative The Clearing House announced includes major financial institutions. This is the banking system building the rails that make AI-directed deposits possible.

Here's the irony: the same technology crypto advocates have championed for years could undermine the fractional-reserve model that makes banks profitable. If money can move instantly, the entire concept of sticky deposits goes away.

What does this mean for crypto specifically? The obvious read is that tokenized deposits legitimize blockchain infrastructure. The deeper read is that they could reshape the competitive dynamics between banks and decentralized finance. DeFi protocols already offer instant yield on stablecoins. If traditional banks need to price deposits closer to market rates to retain them, the gap narrows. But the cost of credit rising is a tailwind for alternative lending platforms too.

From a risk perspective, this is a slow-motion structural shift. It won't happen overnight. The Dallas Fed authors note the magnitude is uncertain and tokenized deposits remain early. But the direction is clear: AI plus instant settlement erodes the funding advantage banks have relied on for decades.

What I think actually happens

The street is missing something here. Most commentary on AI in banking focuses on fraud detection or customer service. That's small potatoes. The real story is AI changing the liability side of the balance sheet, and with it, the entire maturity transformation engine.

My honest opinion: this is a feature, not a bug. A world where depositors get paid closer to market rates is a fairer world. The current system basically taxes lazy depositors to subsidize cheap borrowing for people who take out mortgages and business loans. That's a hidden transfer, and it's regressive. Wealthier households tend to hold more interest-bearing accounts or move money into money-market funds anyway.

But that doesn't mean the transition is painless. If banks lose $700 billion in duration-risk capacity, they either shrink their loan books or raise rates. And they'll almost certainly do both. The bet on home prices, on commercial real estate, on small business expansion, all of that gets more expensive.

Who wins? Whoever builds the best deposit product. The first banks to offer genuinely competitive rates with smooth UX will pull deposits from everyone else. That's a concentration risk. The largest banks have the balance sheets and tech budgets to compete. Regional banks are exposed.

Who loses? Borrowers. Particularly mortgage borrowers and small businesses. They're the ones who benefit most from the implicit subsidy that sticky deposits provide. When that subsidy vanishes, they pay for it.

So what should you do with this information? Watch the rate on your savings account. If it starts tracking the Fed funds rate more closely and more quickly, the deposit beta shift is happening. Also watch for tokenized deposit pilots from major banks. The announced initiative has real momentum behind it.

The numbers tell the story. $5.84 trillion in asset exposure supported by deposit stickiness. A $700 billion hit to duration-risk appetite from a modest 10% increase in sensitivity. Those aren't rounding errors. They're the shape of things to come.

The reality is we're moving toward a banking system where every deposit is a hot potato. That's more efficient. It's also more fragile. And it's going to cost borrowers more.

I'd keep an eye on the credit markets, not the crypto charts, to see when this starts to bite.