AI agents could cost banks $500bn — by winning savers better rates

Lex argues that AI agents such as Meta’s Muse could erode one of banks’ easiest profit pools — idle money — by flagging better rates or automatically moving “lazy money” into money-market funds; banks earn about two-thirds of their income from the spread. JPMorgan, Bank of America and Wells Fargo hold $1.6tn of non-interest-bearing deposits (16% of liabilities); paying 3% on them would cost $47bn a year, almost half their combined earnings. Lex calls that scenario unlikely, saying talk of “agentic bank runs” ignores frictions, notably trust in tech companies and liability for misdirected funds. Its own estimate: the “core deposit intangible” recognised in US bank acquisitions is typically 2–3% of balances (Mercer Capital); applied to $20tn of US deposits, about 80% of them core, that implies roughly $500bn of equity value at risk — about one-tenth of the value of listed US banks (S&P Capital IQ). Offsets: AI also cuts costs, improves loan pricing and reduces errors (Lex cites Citigroup’s bungled $900mn wire transfer), and Brian Moynihan said BofA spent $400mn on AI and saw $800mn of benefits. Bigger worries for now: a Truist survey at end-September found more than half of bank investors see Fed rate hikes as the biggest valuation risk, with very few citing bots, and Jamie Dimon warns a credit downturn is long overdue. Caveat: the $500bn is Lex’s own back-of-envelope valuation estimate, not a forecast of losses.