Oct 13, 2026

Compliance is a line item now. American Express was fined $350 million over anti-money-laundering failings, and the interesting part of the announcement is that regulators applied deposit-era expectations to card portfolios. That means the cost of compliance scales with transaction volume, not with product type, and it lands directly on the P&L of every large issuer. Meanwhile synthetic identity fraud is surging and often goes unnoticed: criminals combine stolen data with AI to fabricate borrowers who look creditworthy, and banks book the losses as bad debt rather than fraud, so nobody can even measure the hole. The two stories reinforce each other. If detection fails, the fine is the fallback, and either way the margin shrinks. My read is that we are entering an era where the risk-adjusted profitability of payments depends less on interchange and more on how well your models distinguish real customers from synthetic ones. What would strengthen this: a second large card issuer hit with a card-portfolio AML action, or better, published loss-rate data that separately identifies synthetic fraud. What would weaken it: evidence that the Amex fine was a legacy-deposit matter wearing a card costume, which is possible.

The merchant payout is a fee-structure story in slow motion. Visa and Mastercard have now paid out more than $631 million under the $5.5 billion interchange settlement, even as Walmart and DoorDash keep fighting the injunctive relief piece. The cash is nice; the lasting pressure is the litigation machinery itself. Every renewal cycle now happens under the shadow of a merchant class that has learned it can sue over swipe fees. I have long held that announced aggregate capex materially exceeds what will be deployed, and that same skepticism applies here in reverse: the headline $5.5 billion overstates what merchants actually get, while the durable effect, sustained scrutiny of network pricing, is understated. Evidence that would strengthen the fee-pressure view: major merchants extracting routing or surcharge concessions in renewals. Evidence against: interchange rates quietly holding steady through 2027.

Compute is infrastructure, and infrastructure gets bombed. Ukraine's strike on three of Yandex's largest data centers took Russian banking and payment apps down with them. This is the physical-world rebuttal to the idea that concentrating compute in a few giant facilities is free efficiency. It also complicates the AI-financing debate. The IMF warns that AI borrowing could unsettle bond markets, and the standard rebuttal is that hyperscalers fund data centers from operating cash flow rather than debt, the position that hyperscaler AI capex is funded from operating cash flow, not debt. Both things can be true: the biggest players may be debt-light while the second tier borrows heavily, and collateral values could fall for reasons nobody controls, like a drone. What would strengthen the caution: rising credit spreads on data-center-backed paper. What would weaken it: capex plans holding while FCF stays positive next quarter.

One day China tokenizes its deposits and one day someone bombs the alternative; the future arrives unevenly.