Editorial
The court just told the Fed to show its math on debit fees
Every time you tap a debit card, a few cents move from the merchant to the bank that issued the card. For big banks, that amount has been capped since 2011 under the Federal Reserve's Regulation II at 21 cents plus 0.05% of the transaction, with an extra cent for fraud prevention. On October 26, the Eighth Circuit threw the 21-cent cap out.
The case is Corner Post v. Board of Governors, brought by a North Dakota truck stop. A unanimous panel of Judges Lavenski R. Smith, Duane Benton and David Stras affirmed a district court decision vacating the cap and sent the rule back to the Fed. Judge Stras framed the whole dispute in two sentences: banks charge fees every time a customer swipes a debit card, and "the question here is how high they can be."
The answer turns on two words in the Durbin Amendment: "incremental cost." Congress told the Fed to consider the incremental cost of authorizing, clearing and settling a particular transaction, and to leave out costs that are not specific to a particular transaction. The court read "incremental" much the way economists read "marginal": the cost of one more transaction. The Fed's 2011 rule never drew that line. With Chevron deference gone after Loper Bright, the panel did not defer to the Fed's reading. The Fed can use reasonable proxies, the court said, but at a minimum it must sort costs into those that are transaction-specific and those that are not, rather than "just pick a number and hope it is close enough for government work."
Here is what I think matters for the people who read this newsletter.
Nothing changes on Monday. The ruling takes effect only when the court issues its mandate, seven days after the deadline to seek rehearing, or after rehearing is denied. In cases involving a federal agency, parties get 45 days to ask. The current cap stays in place until then, and the Fed can still ask the full court or the Supreme Court to step in.
The pressure is downward. Merchants won this case on the argument that the rule let banks recover costs Congress excluded. Retail groups and Senator Durbin filed briefs on Corner Post's side; the big bank and credit union trade groups filed on the Fed's side. The Fed itself proposed lowering the base cap to 14.4 cents in 2023 and never finalized it. A rewrite built only on transaction-specific costs is very likely to land lower than 21 cents.
Issuers should plan for less debit revenue. For banks over $10 billion in assets, debit interchange is real fee income, and it helps pay for free checking and fraud tooling. If the per-swipe number falls, banks will look for that money elsewhere: account fees, a harder push toward credit cards, leaner debit perks.
Merchants should not book the savings yet. The Fed still has to write a new rule, take comments and defend it, and whatever it produces will be challenged too. That takes time.
The bigger lesson is about payments regulation after Loper Bright. For fifteen years the fight was over whether 21 cents was the right number. The court made it about the method. My bet is that every regulator setting prices on payment rails now has to show its work against the words Congress wrote, and that will reach well beyond debit cards.
Zen
Tangerines in Heaven
When Thich Nhat Hanh was a young novice, he asked his teacher whether there were tangerines in the Pure Land of the Buddha. Then he added a condition of his own: "If there are no tangerines I would not like to go there!" Decades later, in "Memories from the Root Temple: Tangerines in Heaven," he still wanted us to know he had really said it.
He pairs that memory with another, from a beach in California, where he led a group in eating a single tangerine over the course of an hour: the fragrance as the peel comes away, the joy of sharing it. One small fruit gave them plenty to be present for.
The boy had been promised a better place, and his first question was whether it held something he already loved. Most of us carry our own version of that promise, and waiting for it can become a habit, until even a good afternoon feels like time to get through on the way to somewhere else. Thich Nhat Hanh keeps pointing back to the life already within reach. The pleasure of a tangerine can sit beside grief or an unfinished worry without resolving either one first.
The same goes for the next promotion, the next funding round or the quarter when things finally calm down: the work in front of you this week is the only place any of it actually happens.
The Week's Best Articles
Payments
Southeast Asian central banks join Singapore's BLOOM digital-money effort
— Monetary Authority of Singapore via Future of Banking | 2026-10-23
The Monetary Authority of Singapore says Bank Negara Malaysia and the Bank of Thailand have joined its BLOOM initiative, working with industry partners on cross-border settlement in tokenized bank deposits and regulated stablecoins.
This is the unglamorous part of digital money that actually decides whether it gets used. A token that cannot move across a border is just a local payment app. Three central banks agreeing to work on the same plumbing is a real step toward regional rails, even if live volume is still to come.
Fintech
Big US banks push back on outside AI agents
— Gizmodo via Fintech Wrap Up | 2026-10-16
Major US banks are blocking third-party AI agents from logging in with customer credentials and moving money, steering them toward controlled, read-only access instead. The stated worry is simple: an agent using your login looks exactly like you, even when it is making decisions you never approved.
There is a commercial motive here too. An agent that manages money across banks makes deposits easier to move and weakens the bank's hold on the relationship. Both motives are real. The open question for the industry is who eats the loss when an agent gets it wrong, and until that is answered, expect banks to keep the door narrow.
Ripple wins its national trust bank charter, and XRP shrugs
— CoinDesk Daybook | 2026-10-27
The OCC gave final approval to Ripple National Trust Bank, which will manage the reserves behind Ripple's RLUSD stablecoin and offer institutional custody. It is a trust charter, so no deposits or lending. XRP barely moved, and CoinDesk noted Bitcoin slipping below $80,000 the same day.
The market got this one right. The charter strengthens Ripple's stablecoin and custody business, not demand for the token. It is another sign that crypto's most valuable assets are increasingly the regulated businesses, not the coins.
Whop's CEO on a marketplace where AI agents outnumber humans
— The Fintech Blueprint | 2026-10-26
Whop CEO Steven Schwartz tells Lex Sokolin that almost 100,000 people earn money on the platform each month selling software, courses, communities and physical goods through a single API-first stack. Whop settles nearly $500 million a month in income to users, and Tether invested $200 million at a $1.6 billion valuation.
The interesting part is the payouts thesis. Schwartz argues stablecoins are the only rail that pays a seller in Nigeria or Vietnam the same way as one in Brooklyn. He also says Whop now has more agent users than human ones. If that holds, the next merchant platform is being built for software customers first.
US Economy
Debt worries rise even among consumers who pay on time
— PYMNTS | 2026-10-23
A PYMNTS Consumer Expectations Index survey of 2,087 US consumers finds debt confidence falling even among people who are keeping up with their bills.
That gap matters for lenders and merchants. An on-time payment tells you about today, not how much room a household has left. Worry is not delinquency, but it is often where pullbacks in spending start.
AI News
Anthropic files its S-1 publicly
— Understanding AI | 2026-10-29
Anthropic made its S-1 registration statement public on Monday, following a confidential filing in June. Understanding AI walks through the financials, which run through June 30.
For anyone who has wondered what the economics behind frontier AI actually look like, this is the first audited look. The filing does not set a date or price, but it gives investors something better than leaks to argue about.
AI-assisted cyberattacks put pressure on defenders
— Nate's Substack, citing Microsoft | 2026-10-24
Nate argues that agents are becoming part of both attack and defense because they can study unfamiliar software, test a theory and keep going after a failure. Microsoft's 2026 Digital Defense Report describes attackers using AI to compress the attack lifecycle and automate phishing and reconnaissance.
For payments and banking teams, the takeaway is persistence. Attacks that used to stall when a human got bored now keep running. Defenses need to assume the attacker never sleeps.
AI labs are buying data from dead startups
— MIT Technology Review | 2026-10-13, recirculated 2026-10-28
MIT Technology Review reports that wind-down services are brokering code, Slack logs, emails and workspace data from shuttered startups to AI companies building training environments.
If you have ever signed a SaaS contract, read this one. The permission a customer gave a startup was not written with model training in mind, and a failed company's data room may hold your information. It is worth checking what your vendor agreements say happens to data in a shutdown.
Anthropic seeks input from religious and cross-cultural groups
— The Associated Press via PBS News and The Neuron | 2026-10-26
The AP reports that tech companies are increasingly consulting faith leaders on AI. Anthropic says it has met with scholars, clergy, philosophers and ethicists from more than 15 religious and cross-cultural groups to inform Claude's values and constitution.
Wider input can surface assumptions a small team would miss. The harder part, as experts in the AP piece note, is that traditions often disagree, and consultation alone does not tell us how those disagreements get resolved.
Cheaper AI coding does not make owning software free
— The VC Corner | 2026-10-27
The VC Corner asks whether AI coding tools make it cheaper for companies to build what they now buy, using an illustrative comparison of about $7,700 in engineering time against a $30,000 annual software bill.
The real distinction is between writing version one and owning the tool for years. Maintenance, security and the engineers it ties up all belong in the math. Build-versus-buy just got more interesting, not settled.
See you next week,
-dg