Business

Know the Business: Nu Holdings Ltd.

Bottom line. Nu is a credit-led Brazilian retail bank wearing a fintech valuation. The engine is a fast-growing, mostly-unsecured consumer loan book funded by an unusually cheap deposit franchise and serviced by a digital cost base that incumbents cannot match. What the market most likely underestimates is the credit-cycle exposure of a 92%-unsecured book that has yet to be tested in a Brazilian recession at this scale; what it likely overestimates is how much of today's ~30% ROE is structural versus a 15% Selic borrowed from the Brazilian Central Bank.

1. How This Business Actually Works

Nu earns money the same way every retail bank does: it pays customers near-nothing for their deposits, lends those deposits to other customers at credit-card rates, and pockets the spread. The mobile app is the distribution channel; the credit book is the engine. Net interest income now represents 79% of revenue (FY2025), up from 43% at IPO — every period that share rises, this story looks less like fintech and more like a spread bank.

Four cost advantages compound, not one. Management calls these the "four low costs" — and unlike most management framings, this one is load-bearing.

The most important number in that table is the deposit cost at 88% of CDI, not the cost to serve. Brazil's interbank rate ended 2025 near 15%; a digital bank funding itself ~12 percentage points below where a wholesale issuer would have to pay, on US$42B, is generating roughly a billion dollars of structural funding advantage before anyone makes a loan decision. The credit-card book is built on top of that funding moat, not the other way around.

A simple analogy. Think of Nu as a credit-card issuer that, instead of borrowing from capital markets at SOFR+spread, has a free checking account franchise quietly bankrolling it at half the cost of every competitor. The mobile app feeds that funnel; the credit underwriting monetizes it.

What this is not. Not a payments company, not a software business. Every dollar of credit-card receivable consumes Tier-1 capital under Basel III; growth in the loan book has to be funded by deposits or equity, regardless of how cheap the tech stack is.

2. The Playing Field

Nu sits in the middle of LatAm financial peers by scale, but the comparison set is two very different groups: incumbent Brazilian banks (lower growth, lower multiples, leveraged balance sheets) and listed fintech challengers (smaller, more volatile, mostly payments-led). Nu is the only listed name that pairs incumbent-scale ROE with fintech-scale revenue growth, and that is what justifies the multiple gap to ITUB/BBD.

Where Nu actually sits on profitability vs. growth. Itaú prints higher absolute ROE on roughly 10× balance-sheet leverage, but at 2–3% revenue growth. Nu prints almost identical ROE at less than half the leverage and ~10× the growth rate. That is the asymmetry the multiple is paying for.

What the peer set reveals. Three things, in order:

1. The relevant comparable is not ITUB; it is "ITUB in 2010." Incumbent ROEs have been propped up by the same Selic that is helping Nu — but Itaú has stopped growing customers and its loan book is mature. Nu is buying that same ROE at a much earlier point on its growth curve.
2. The fintech-listed tier has decoupled. PAGS and STNE were valued like Nu in 2021 and now trade at 1.0× book. Their problem was leading with acquiring (a margin-compressing service business) rather than credit. The cautionary tale: payments-led fintechs collapsed; spread-led fintechs (Nu, Inter) compounded.
3. MELI is the only true peer on growth-with-profitability — but MELI is a marketplace with fintech attached, not a bank. Different capital intensity, different regulatory perimeter, different exit valuation.

3. Is This Business Cyclical?

Yes — and where the cycle hits is cost of risk on the unsecured book, not revenue. Nu's credit portfolio is 92% unsecured (67% credit card + 25% unsecured personal loans, only 8% secured payroll/FGTS). That is the highest-loss-severity slice of any retail bank's balance sheet, and the company has not yet operated through a full Brazilian unemployment shock at current scale.

How the cycle shows up in this specific business.

What history says. Brazil's 2022–2023 retail-credit cycle pushed sector credit-card NPLs above 14% (post-Americanas, post-rate-hiking shock). Nu's reported NPL 90+ touched 7.0% in Q3'24 before recovering to 6.5% in Q1'26 — meaningfully better than the sector. The disclosed coverage ratios (NPL 90+ coverage 249%, Stage-3 coverage 81%) show management provisioning aggressively rather than relying on the cycle. That is the right answer — and it is also why a 30%+ ROE in the up-cycle is fragile. A 10% normalization in up-cycle ROE is not bearish; it is base case once Selic normalizes from 15%.

4. The Metrics That Actually Matter

For a credit-led digital bank, the headline numbers that move the stock (revenue growth, net income) are downstream. The five metrics below are upstream — they tell you whether the franchise is compounding or showing strain. Watch these every quarter; they will move 1–2 quarters before the P&L.

Why not just watch ROE? ROE in Brazil is partly an artifact of Selic. At Selic 15%, even a mediocre lender can print 18–20% ROE; at Selic 8%, a great lender prints 15%. ROE direction matters less than whether ARPAC is rising while cost of risk is stable — that combination tells you the unit economics are compounding independent of macro.

ARPAC dipped when Nu went mass-market (prepaid cards, NuAccount-only customers brought the average down) and has roughly tripled since 2021 as those same customers were cross-sold credit cards, personal loans, and investments. The Q1'26 print of US$16 extends the line. If this curve flattens for two consecutive quarters at constant FX, the bull case is in trouble.

5. What Is This Business Worth?

The right lens is price-to-tangible-book vs. through-cycle ROE, with growth optionality on the side — not SOTP, not EV/EBITDA, not P/FCF in isolation. A spread bank's value is the equity it carries times the return that equity earns net of the cost of equity, discounted for the risk that the spread is partly cyclical. SOTP is genuinely the wrong lens here: Nu reports as a single operating segment, has no listed subsidiaries, no holding-company discount, and ~99% of revenue runs through the same balance sheet engine.

A simple framework. At ~7.2× tangible book and ~30% ROE, Nu prices to roughly 24× through-cycle earnings if ROE normalizes to 18–20%, or roughly 15× if ROE holds at 25%+. The P/E of 28.6× on reported FY25 is misleading because reported ROE is at peak Selic. The honest underwriting question is not "what is the multiple," it is "what ROE will this business earn through a full cycle, and how much is the loan book worth growing at 20%+ for another five years."

6. What I'd Tell a Young Analyst

Five things you cannot get from a screen. In order of importance:

1. The credit cycle has not happened yet at this scale. Nu reached 100M+ customers in 2024. Its loan book grew through a benign Brazilian credit environment with falling sector NPLs. The 6.6% NPL 90+ Nu is reporting today is good — but it is also the best-case version of Nu's credit performance. The right mental discount on the ROE is "what does this look like with NPL 90+ at 8.5% and cost of risk at 50% of pre-provision revenue."

2. Read the deposit-cost line, not the customer-count line. Customer growth is a vanity metric at this scale — Brazil is saturating at 62% of adult population. Whether deposits stay at 88–91% of CDI as the base grows toward US$60–80B is the actual moat test. Incumbents will fight harder for deposits as Open Finance enables portability. If deposit cost drifts toward 95% of CDI, the funding edge halves.

3. The Mexico option is real but small until it isn't. Mexico hit break-even in Q1'26 at <1% gross-profit market share of a US$43B pool growing 15% CAGR. If Mexico ARPAC follows Brazil's curve (took 4–5 years from break-even to ARPAC > US$10), this is a 2028–2029 story, not a 2026–2027 story. Do not underwrite Mexico as if it were Brazil two years ago.

4. PIX is a feature, AI is mostly marketing. Nu's commentary about "rebuilding banking around AI" is real for collections and underwriting cost reduction, but unit economics in this business are driven by the credit decision and the funding mix, not the model. Treat AI capex as the new SaaS R&D — necessary, not differentiating.

5. What would change my mind. Three things, in order: (a) a single quarter where ARPAC rises but NPL 15–90 also rises by >50bp — revenue being bought with relaxed underwriting; the next two quarters' provisions will hurt; (b) deposit cost moves above 95% of CDI for two consecutive quarters — funding moat closing; (c) Mexico stops compounding ARPAC while still losing money — the international expansion math no longer works.