The Data Moat
Banks have been building their AI advantage for decades. Fintech lenders are just figuring that out.
TL;DR
Machine learning is rewriting credit underwriting, but the biggest winners aren’t the fintechs that pioneered it. Banks are quietly pulling ahead, powered by something no startup can replicate: decades of proprietary credit data. The smarter fintechs are adapting. The rest are exposed.
The Fintech Implication
There’s a version of the AI-in-lending story that flatters fintechs. They moved fast, ditched the FICO score, and built machine learning models while banks were still running loan applications through legacy systems and gut instinct. For a while, that was mostly true.
But the arms race has shifted. And the banks are pulling ahead.
The reason isn’t technology, it’s data. ML models are only as good as what you feed them, and on that front, no fintech lender born in the last decade can match a JPMorgan or a Wells Fargo. These institutions carry decades of proprietary credit performance data spanning multiple full economic cycles: the dot-com bust, 2008, the COVID shock, the rate spike of 2022–23. That’s not historical trivia, that’s the training set. And it compounds. The more loans you’ve made, the more performance data you accumulate, the sharper the model gets. JPMorgan is spending $18 billion annually on technology with a clear thesis: pairing AI to its proprietary data creates an advantage competitors simply can’t replicate.
Fintechs tried to close the gap with alternative data, rent payments, utility bills, real-time cash flows, behavioral signals. The pitch was compelling: find the creditworthy borrowers the traditional system was missing. Some of that is real. But alternative data has a ceiling, and it’s becoming visible. The core problem is cycle-testing. Most fintech underwriting models were built and optimized in a low-rate, low-default environment. They’ve never seen a real recession. Banks have, multiple times, and that stress-tested history lives in their models. A borrower whose cash flow looks fine today may look very different six months into a downturn. Banks know what that pattern looks like. For most fintechs, it’s an open question.
The smarter players have already pivoted. Rather than competing head-on, companies like Pagaya have repositioned as infrastructure, plugging their AI underwriting layer directly into bank workflows, approving borrowers that legacy models reject, and packaging the loans into ABS for institutional investors. It’s a capital-light model that sidesteps the data moat problem entirely by becoming part of the bank’s stack rather than fighting it. Banks originate the loans; Pagaya’s AI approves borrowers the banks would otherwise reject; the loan gets sold to ABS investors. The bank keeps the customer relationship. That’s a smarter model than pure disruption, and it’s where the fintech lending opportunity is quietly migrating.
The So What
Three things to watch:
JPMorgan’s AI build-out the bank has been explicit that its data advantage is the moat, not its models. Watch how its consumer lending margins evolve over the next two quarters as AI-driven underwriting scales. If loss rates diverge meaningfully from fintech peers during any macro softening, the data thesis gets confirmed in real time.
Pagaya’s network volume a live proxy for how much of the credit stack is migrating toward AI-as-infrastructure. Q1 2026 guidance came in at $2.5–2.7 billion. Sustained growth here signals the bank-fintech partnership model is the winning form factor, not direct competition.
The first real credit cycle test delinquency trends in consumer credit are the tell. If alternative-data models start showing unexpected stress relative to traditional bank portfolios, that’s the moment the data moat thesis stops being theoretical.
Next issue: CBDCs and stablecoins — two visions for the future of money are on a collision course. One is built by governments. One is built by markets.