Business

Know the Business

Nu is a spread lender wearing a tech wrapper: 85% of revenue is interest income on credit-card and personal-loan balances, funded by sticky retail deposits at 81% of the Brazilian interbank rate, distributed through one app at a monthly cost-to-serve of about $0.80 per active customer. The moat is unit cost, not pricing — and the moat compounds because every additional customer makes the underwriting models smarter without adding meaningful overhead. The market is correctly pricing a high-quality compounder (P/E ~20, P/B ~5, FY2025 ROE 30%); what it may be underestimating is the gap between blended monthly ARPAC of $15 and mature-cohort ARPAC of $27, and what it may be overestimating is the durability of a 26.7%-of-revenue cost-of-risk line in a Brazilian unsecured-credit cycle the company has not yet been tested through at this scale.

How This Business Actually Works

One sentence: Nu rents money from depositors at sub-policy-rate pricing, lends it to consumers at credit-card and personal-loan rates, eats a 27%-of-revenue credit-loss line, and keeps the difference — because operating costs are 19.9% of revenue, less than half of what an incumbent's branch network costs.

The chart on the right is the most important picture in this report: funding cost plus credit losses consume 56 cents of every revenue dollar before a single employee is paid. Marketing at 1.9% of revenue and customer support at 4.1% are rounding errors next to a 1 pp move in cost of funding or NPL formation. This is why "tech multiple" framing is misleading — earnings variance is driven by two bank lines, not software cost curves.

Bargaining power is asymmetric and sits in three places. Customers have very little: switching is technically easier than ever with Open Finance, but Nu's NPS above 85 and primary-banking-relationship penetration over 60% mean the friction is psychological, not structural. Depositors have meaningful power in a high-Selic regime — Nu pays 81% of CDI today, and that ratio is the single best operational signal for whether the franchise is intensifying or weakening. Regulators hold the most leverage: a revolving-credit interest cap (already in force since 2023), an RWA methodology revision, or an interchange reset can compress a layer of the margin stack overnight.

The Playing Field

Three things this peer set reveals. First, Nu is alone in its quadrant: only MELI matches the ROE-and-growth profile, but MELI is an e-commerce platform earning bank-like returns from a non-bank base, priced at 11.7x book versus Nu's 5.2x. Among pure financial peers, the incumbents (ITUB, BBD) trade at 2x book or less because their ROE is mid-to-high teens with no growth, and digital challengers (INTR, PAGS) trade below 1.5x book because they have not yet shown they can earn an incumbent-class ROE. Nu has done both — 30% ROE with 27% revenue growth — and is the only LatAm bank for which a 5x book multiple is rationally defensible. Second, what "good" looks like in this industry is converging Itaú-level ROE (around 20%) with digital-bank cost structure; Nu has already overshot the ROE benchmark by ~10 points and has further unit-cost room to run. Third, the bear case sits in the peer table too: Inter and PagBank are also digital, also Brazilian, also Open Finance-enabled — and their ROEs are 14% and 14.5%. The premium ROE Nu earns is not a structural property of being digital; it is the property of being the digital franchise that achieved scale first.

Is This Business Cyclical?

Moderately to highly cyclical, with the cycle hitting through cost of risk and cost of funding before it touches revenue. Nu has never reported through a deep Brazilian unsecured-credit downturn at current scale — it became profitable only in FY2023, and the 2023 NPL spike (Brazil's last real consumer credit stress) was managed without distress, but the loan book has grown roughly 4x since.

The provision-growth chart tells the most underappreciated story in the Nu file: the 2022 232% jump in provisions was the company's stress test, and it survived without distress — but the trajectory of 33% provision growth in FY2025 against 27% revenue growth means cost of risk is still rising slightly faster than the topline. That is normal in a growing unsecured book, but it is the line to monitor: a flip where provisions outpace revenue by 10+ pp in a single year would be the signal that the cycle has turned. Nu's reported leading indicator (15-90 NPL at 3.9% in December 2025, down from 4.1% at Q3) is the cleanest real-time gauge.

The Metrics That Actually Matter

Forget headline EPS. Five numbers will tell you whether Nu is creating or destroying value, in roughly this order of importance:

The two metrics worth more attention than they get are ARPAC and cost-to-serve, together, because their spread is the unit economic. ARPAC has tripled from $4.50 to $15 since 2019; cost-to-serve is essentially flat at $0.80; mature cohorts are already at $27. If a mid-cohort customer worth ~$15 in revenue today and ~$2 in net contribution can plausibly become a $27-revenue/$2-cost customer over a decade, the company's intrinsic value compounds well above what consensus is modeling — because the marginal customer needs no capex or branch.

What Is This Business Worth?

The right lens is P/B against sustainable ROE, with explicit modeling of the ARPAC compounding curve and cost-of-risk normalization through a cycle. This is one economic engine, not a sum-of-parts: Brazil, Mexico and Colombia are the same playbook at different stages of customer-base maturation, and the consolidated numbers do not hide materially different businesses. SOTP would force-fit segment math the company does not disclose at the necessary granularity.

The peer P/B-vs-ROE chart frames the valuation question precisely. If Nu's sustainable through-cycle ROE is 30%, the 5.2x book multiple is rational versus the bank-peer line that prices Itaú at 2.2x for 21% ROE. If sustainable ROE is closer to 20% — i.e., cost of risk normalizes higher in a cycle and ARPAC compounding slows — the multiple should compress toward 3x book, implying meaningful downside even before any growth disappointment. The asymmetry: ARPAC has a clear path higher (mature cohorts at $27 vs blended $15) and customer growth has multi-year runway; both push the numerator up. The risk: cost of risk is more sensitive to Brazilian unemployment than to anything in management's control.

What would change the thesis materially: (1) cost of funding rising above 90% of CDI for two consecutive quarters; (2) 15-90 NPL rising above 5% with no seasonality explanation; (3) Mexico/US capital absorption pushing equity growth below earnings retention; (4) revolving-rate or RWA regulatory move that compresses NIM by 100+ bps. Absent these, the franchise is doing what a high-quality compounder does and the multiple is defensible.

What I'd Tell a Young Analyst

Stop thinking of this as a tech stock and stop thinking of it as a bank. It is a digital balance sheet — a bank by economics, a tech company by cost structure. The dollar of revenue you should model is not "ARPU times users"; it is (spread on credit balances + fees per active customer) minus (cost of funding + cost of risk + $0.80 cost-to-serve). Everything else is decoration.

Three watch items, in priority order:

1. The ARPAC-versus-cost-to-serve gap. If ARPAC compounds at 15%+ per year and cost-to-serve stays under $1, every other concern is secondary. If cost-to-serve creeps up — because Mexico support takes more headcount, or US compliance adds layers — flag it immediately; the unit economic is the franchise.

2. 15-90 NPL, monthly. This is the leading indicator that prints before earnings. Brazil's BCB publishes consumer-credit NPLs monthly; Nu's own disclosure is quarterly. Watch the BCB series in the months between Nu prints — a 30 bps move in 15-90 NPL on the unsecured book is the kind of early signal that moves the stock when the next quarter lands.

3. Cost of funding as % of CDI. Today 81%, the company's chosen ceiling. If a quarterly print comes through at 85%+ without a corresponding 100 bps spike in Selic, that is the deposit franchise weakening — possibly because Inter, Mercado Pago or BTG Digital is bidding deposits up. This is the metric where the moat shows up first, and where the first crack would appear.

What the market may be missing: mature customer cohorts already generate $27/month versus a $15 blended average — that is not aspirational, it is a fact in 2025. If the rest of the base ages into that curve, FY2030 ARPAC could be 60%+ above today on essentially the same cost-to-serve. What the market may be overestimating: that Mexico replicates Brazil at Brazil's profitability. Mexico has Mercado Pago co-leading the market, a different policy-rate dynamic, and Sofipo-licensed competitors with deposit rates that already eat into Nu's cost-of-funding edge. The Brazil playbook will not run with Brazil's economics.

What would actually change the thesis is not a quarterly miss on ARPAC or a single bad NPL print — it is structural: the OCC charter consuming more capital than the LatAm franchise generates, a Brazilian recession driving 90+ NPL past 9%, or a revolving-rate cap that compresses card NIM by 200+ bps. Those are the events to underwrite against.