Banks Are Not Disrupted

The moat holds, the IP inside it does not

The banking industry's classic moats, regulatory license, lending relationships, balance sheet scale, are real and durable enough that banks do not fit the standard disruption story, where a cheaper, worse product displaces an incumbent and improves over time. The sharper claim is that artificial intelligence is disruptive to something narrower: the intellectual property that sits inside the moat, meaning how a bank prices and manages risk, historically kept in spreadsheets and in the judgment of individual underwriters. As that pricing logic moves into models, the moat itself does not move, but who wins inside it can change quickly. The formulation is compact: the moat stays the same, the winners inside the moat change.

A competitor's concession, made against interest

Marc Rowan, whose firm Apollo competes directly with banks for lending business, supplies the strongest outside confirmation that the moat is intact. Asked in 2026 whether an AI-financing bill eventually lands on banks, he said the banking system is incredibly deleveraged and diversified, and in the best shape he had seen in forty-two years in the business, a concession volunteered during an active credit selloff rather than a calm moment.¹ His explanation for why is the more interesting part: banks are healthy specifically because the dangerous, long-duration lending left their balance sheets after the 2008 financial crisis, migrating through CLOs and business development companies to unlevered investors who can price it directly rather than warehousing it against insured deposits. The bank moat held, on his account, not because banks defended the territory but because they shed the part of it that was dangerous.

Where the client stays and the asset does not

Sixteen months earlier, in a separate interview, Rowan drew the boundary even more precisely, describing what his firm wants from a bank relationship and what it deliberately does not want. He does not want the bank's client, since Apollo cannot provide the surrounding bundle of advisory, foreign exchange, custody, and payments services a bank offers; he wants the asset, the loan itself.² If a bank loses a client to a competitor, it loses the entire relationship and every service inside it. If a bank loses an asset to a firm like Apollo, it loses only that asset, often because new capital rules made holding it unattractive in the first place. The migration is voluntary and regulatory in origin, not competitive displacement, which sharpens the moat-holds argument rather than contradicting it: the perimeter of what a bank chooses to hold has contracted by rule, while the client relationship, and the fee-bearing services layered on it, has stayed exactly where it was.

Malka's dating of the thesis

Micky Malka agrees with the moat-holds framing for a specific window and says that window closed. He describes incumbents genuinely catching up to neobanks on product between roughly 2021 and 2023, to the point that there was not much difference between an incumbent's financial product and a neobank's.³ Artificial intelligence then acted, in his words, as a nitro rocket for the technology companies that were still founder-led, separating them from the incumbents again. The names he gives as the winners of that re-separation, Stripe, Robinhood, Nubank, and Revolut, are notable for what they are not: neither scrappy startups nor the banks themselves, but founder-run technology companies that had already built the modern stack the AI wave rewards. Malka's account does not contradict the moat-holds thesis so much as time-stamp it. Absorption toward parity was real for a few years; it stopped being true once AI became the axis competition ran along.