Tokenized deposits could make bank funding less stable and ultimately raise credit costs for US households and businesses, according to a new analysis by two economists at the Federal Reserve Bank of Dallas — a rare institutional pushback against one of the blockchain industry’s most heavily promoted use cases.
Economists Rosie Levy and Srini Ramaswamy argued that instant settlement could allow depositors seeking higher yields to switch banks more quickly. Programmable deposit tokens and agentic artificial intelligence could automate those transfers, shortening the time deposits remain at individual banks and making them more sensitive to interest rates.
The numbers behind the warning
The analysis is stark. The economists estimated that if deposits became 10% more sensitive to interest rates, banks’ capacity to hold long-term loans and other assets could fall by about 700 billion USD. In a separate scenario, deposits remaining at banks for 10% less time could reduce that capacity by about 580 billion USD.
Both figures are expressed in 10-year equivalents and do not represent direct reductions in lending. The calculations are scenarios rather than forecasts, and the authors were careful to note they are not dollar-for-dollar estimates of lost credit. Still, the scale of the figures signals how seriously at least some Fed researchers take the potential side effects of putting bank deposits on programmable rails.
Why speed can be a liability
The core of the argument is a classic banking tension. Banks fund long-term loans — mortgages, business credit — with deposits that can be withdrawn at any moment. That maturity transformation works because, in practice, most deposits are sticky. Customers leave money sitting in checking accounts even when better yields exist elsewhere, largely because moving money is slow and inconvenient.
Tokenized deposits threaten that stickiness. When deposits become programmable tokens that can move between institutions around the clock, the friction that shields banks from sudden outflows erodes. An AI agent acting on a depositor’s behalf could continuously shuffle funds toward the highest yield, turning a stable deposit base into something closer to a real-time wholesale funding market.
Levy and Ramaswamy said banks could respond to more volatile deposits by holding larger portfolios of highly liquid assets, including reserves and US Treasurys. Banks could also lean more heavily on term debt to maintain their lending portfolios — though funding loans through wholesale debt would likely increase credit costs for consumers and businesses. The burden, in other words, would eventually land on borrowers.
The industry is building anyway
The warning comes as US banks race to build exactly the infrastructure in question. On Tuesday, 39 US state banking associations formed the BankChain Alliance to develop a nationwide network supporting tokenized deposits, stablecoins and automated settlement. The Clearing House is developing a separate network backed by JPMorgan Chase, Bank of America, Citi, BNY and Wells Fargo.
Banks have also begun connecting their tokenized-deposit systems across institutions. On Aug. 20, Standard Chartered and HSBC completed a live cross-border transaction through Swift’s blockchain ledger, which linked the banks’ separate systems and recorded their resulting obligations before settlement through existing payment infrastructure.
For the banks, the appeal is obvious: tokenized deposits promise programmability, instant settlement and 24/7 operation while keeping customer funds inside the regulated banking system — a deliberate contrast to public stablecoins. Regulators have taken notice of both sides of the debate, with the FDIC mulling guidance on tokenized deposit insurance and stablecoin applications.
A Brazilian precedent
The Dallas Fed authors cited Brazil’s Pix instant-payment system as a potential comparison, while noting it is not identical to tokenized deposits. A 2025 study by Brazil’s central bank found that heavier Pix use increased banks’ holdings of liquid assets and reduced credit intermediation — evidence that faster payment rails can, in practice, reshape how banks balance liquidity and lending.
If tokenized deposits amplify that dynamic with programmability and AI-driven automation, the credit-cost effects could be larger than anything Pix produced.
What it means for DeFi and banks
For the crypto industry, the report is a reminder that tokenization’s consequences cut in both directions. Faster, programmable money is celebrated as an efficiency gain, but the Dallas Fed analysis frames it as a potential tax on credit intermediation — one that could cost the economy hundreds of billions in lending capacity if deposits behave the way the scenarios suggest.
The debate is far from settled, and these are scenario estimates from a single regional Fed. But as BankChain, The Clearing House network and Swift-linked pilots move from demos to production, the questions Levy and Ramaswamy raised will move from academic journals to bank treasuries — and possibly to the interest rates ordinary borrowers pay.
700 billion in lost lending capacity from deposits getting just 10% more rate sensitive. dallas fed did not come to play
The 580 billion scenario feels more plausible to me. Deposits already move faster between banks than they did five years ago.
agreed on the 580b duration scenario. deposits already jump banks the moment rates move, instant settlement just removes the last friction. blockchains dont sleep and neither would the outflows
the fdic could tell the same story about money market sweeps and print the same number. instant settlement just deletes the teller line
money market sweeps still take a day and a human clicking the button. an agent rerunning the comparison every block changes the magnitude even if the mechanism is old
sweeps cap out at what a human can move before lunch. programmable outflows compound overnight, that difference is where the 700b lives
exactly, the dallas fed paper reads like the counterargument to three years of tokenization press releases. rate sensitivity cutting deposit franchise value is the whole point, not a footnote
Finally some sober pushback on tokenized deposits. Instant settlement cuts both ways and the industry only ever mentions the upside.
bank runs at programmable transfer speed, with agentic AI deciding when to move the money. what could possibly go wrong lol
the agentic AI part is what turns a slow run into a millisecond one. deposit contracts wont renegotiate at machine speed
agentic AI moving deposits at programmable speed is basically front running the fed funds rate lmao. the dallas fed wrote the horror version of the tokenization pitch deck
the agentic AI angle is the part nobody prices. deposits that rebalance themselves the second a bank tweets bad news, thats a bank run with no humans involved
700 billion in reduced loan capacity from a 10% jump in rate sensitivity is a wild number. People forget instant settlement cuts both ways, banks actually need sticky deposits to lend.
Exactly. The Dallas Fed scenario is about deposit duration shrinking, and agentic AI automating the switches makes it worse. Nobody in crypto wants to hear this part.
and that 700b assumes only a 10% sensitivity bump. AI rate shopping running continuously could blow way past that scenario
First serious institutional pushback on tokenized deposits I have read all year. Levy and Ramaswamy basically described a bank run that executes itself in code.
Levy and Ramaswamy described a bank run that executes itself in code, which is exactly right. humans hesitating at the branch used to be the brake
funny how the people who say banks are obsolete also expect credit to stay cheap while deposits get yield chased at machine speed. levy and ramaswamy did the math nobody wanted
10% more rate sensitivity wiping 700 billion of lending capacity is the single best number against the tokenized deposits pitch i have read all year
banks spent decades pricing deposit stickiness into loan books. the dallas fed just asked what happens when the stickiness becomes a config setting
a config setting is exactly right. every bank CIO who read that line just aged five years, and the 580b duration scenario is the polite version
That 700B figure is conservative. Real-world sensitivity could be 2-3x higher when AI agents start shopping rates across multiple platforms simultaneously
2-3x assumes agents get account level access everywhere. the realistic middle case is ugly enough, no need to inflate the number to make the point
Tokenized deposits always assume stable retail investor behavior. The AI angle makes this a completely different equation – automated withdrawals based on algorithms
Exactly – the study shows how tokenized deposits amplify existing banking fragility rather than solving it. The 700B number is actually conservative