Nu Holdings Ltd.

The trailing multiple sits below where the shares finished each of the last three fiscal years, and the forward multiple sits lower still, while reported earnings growth has accelerated. That gap between a compressing multiple and an expanding earnings base is what produces the sub-one PEG figure in the panel, and it is the central fact of the current setup.

Three engines run in parallel, and they do not depend on each other in the way a conventional bank's segments do. Credit cards supply the entry product and the behavioural data. Unsecured and secured lending supply yield. Fees and float supply earnings that carry no credit cost at all. What makes the structure work is the order in which a customer passes through them. Nu acquires almost entirely through referral and brand at a cost per customer that traditional banks cannot approach, monetises first through the card, then through deposits it holds cheaply, then through lending priced off data the first two products generated. Each stage funds the next without external capital.

The fastest-growing line last year was float income on assets held at amortised cost, which is the deposit franchise doing its work. Deposits are gathered at a fraction of the interbank rate, and the gap between that funding cost and the yield on the credit book is where the economics live. Unsecured lending grew fastest among the credit lines, which is also where the risk sits, and the second-quarter margin expansion shows management pushing that mix on purpose.

Operating leverage originates in the cost line, not the revenue line. Monthly cost to serve an active customer sits below a dollar, and it has stayed there while revenue per active customer has roughly doubled over three years. That single ratio explains the efficiency ratio and the return on equity together. The risk is that the leverage is now partly spent: management has committed the incremental margin to AI infrastructure, Mexican banking operations and a United States build, so the ratio improves more slowly from here even as the underlying cost advantage holds.

Rate the mechanism, not the brand. The cost advantage earns the top mark because it is arithmetic rather than reputation: no branch network, no legacy core system, and a support function where AI agents now handle the majority of Brazilian conversations at or above human quality. An incumbent cannot match that without dismantling assets it already owns, and a new entrant cannot match the scale over which the cost is spread. That is the pillar most likely to survive a downturn intact.

The data pillar is stronger than it looks from outside and weaker than the company's framing implies. A decade of transaction history across more than a hundred million customers is a genuine input advantage, and NuFormer is now in production across three portfolios with two more in testing. The evidence management offers is that delinquency among customers who bank primarily with Nu runs at roughly half the portfolio average. That is a real finding, though it also describes selection: primary-relationship customers would be better credits at any lender. The underwriting edge is demonstrated in Brazil across income bands; it is asserted, not yet demonstrated, in Mexico and Colombia, and management told analysts it expects to need between one and three years of local data before it trusts United States models.

The deposit franchise is the pillar under most pressure, because it is the one competitors attack directly. Consolidated cost of deposits has improved against interbank rates, which shows the franchise holding. Brand and referral acquisition carry the lowest rating among the strong pillars for a simple reason: the mechanism has never been tested outside Latin America, and the United States launch prices against incumbents with distribution Nu does not have.

Sequencing matters more than the list. Brazilian monetisation has to work first, because it funds everything else from internal cash, with no call on capital markets. Nu already serves most of the Brazilian mass market and holds a small single-digit share of a domestic profit pool worth more than a hundred billion dollars a year, so the growth is no longer about adding customers. It is about revenue per customer, which is why Croma and Ultravioleta matter: they move the company into the segments where incumbent banks still earn most of their profit.

Mexico is second in sequence and first in evidence. The licence converts a credit-first fintech into a full institution able to gather payroll deposits, and the cohort data management presented shows Mexican customers monetising earlier than Brazilian customers did at comparable penetration. If Mexico compounds the way Brazil did, it removes the single-country concentration that the current multiple partly reflects.

The United States sits last and should be read that way. Management said profitability there will take a number of years, against eight in Brazil, and capped the investment inside the guided efficiency envelope. The launch runs through a partner bank while the charter clears the Federal Reserve and the FDIC, which means shared compliance and shared revenue in the interim. Nothing in the next two years depends on it working. Everything in the next decade might.