Long-Term Thesis
Long-Term Thesis: Nu Holdings Ltd.
1. Long-Term Thesis in One Page
The long-term thesis is that Nu becomes the dominant mass-market consumer financial services platform across Latin America by 2030–2035, compounding tangible book value at roughly 18–22% annually on three pillars: a structural funding-cost edge (sticky retail deposits priced below the interbank rate), a structural cost-to-serve edge (cloud-native infrastructure at roughly $0.80 per active customer per month versus 40–50% efficiency ratios at incumbents), and the longest reinvestment runway in Latin American banking — roughly 7% of Brazil's $100B-plus profit pool and under 1% of Mexico's $43B pool today.
This is only a long-duration compounder if through-cycle ROE clears 20% on a credit book that has never been recession-tested at $32.7B and 92% unsecured, and if the deposit franchise holds below 95% of CDI as it scales toward $80B+. The 5-to-10-year case works only if at least one of Mexico, the US, or a deepening Brazilian wallet share matures into a second meaningful profit engine before Brazil saturates. The valuation already pays for the Brazil core (28.6× trailing earnings, 7.2× tangible book against 30% ROE); the multi-year return depends on what is added to that core, not on re-rating it. The single hardest fact to internalize is that today's 30% ROE is partly rented from a 15% Selic — the durable thesis is whether the operating machine still prints 18–22% ROE at Selic 9–11%.
| Thesis Strength | Durability | Reinvestment Runway | Evidence Confidence |
|---|---|---|---|
| High | Medium-High | High | Medium |
The 5-to-10-year underwriting question is not "is Nu a good bank" — it clearly is — but "does the operating machine compound when Selic normalizes, when Brazil saturates, and before international optionality earns its capital." The answer hinges on three things: deposit cost as a percentage of CDI as the base scales, through-cycle ROE on the unsecured book, and whether Mexico's ARPAC follows Brazil's curve in 2027–2029.
2. The 5-to-10-Year Underwriting Map
The drivers that must work, the evidence we have today, why each could endure for a decade, and what would break the case.
The driver that matters most. Driver 1 — deposit cost as a percentage of CDI as the base scales — is the load-bearing test for the entire 10-year case. The cost-to-serve edge survives essentially any scenario; ARPAC compounding follows almost mechanically from cross-sell if the customer relationship stays primary; international optionality is priced near zero. The funding moat is what turns 18–22% ROE into a defensible compounder versus a peak that mean-reverts. If deposit cost stays below 95% of CDI on an $80B base in 2030, the operating machine is intact. If it drifts above 95% sustained, the bull case requires a different argument.
3. Compounding Path
The long-term path is a book-value compounding story, not a revenue story. With ROE at 30% in 2025 and capital allocation 100% reinvested (no dividend, minimal buyback, $7M of FY25 capex), the returns calculus reduces to: at what through-cycle ROE does the book compound, and for how long can it compound before mean reversion?
Tangible book per share roughly doubled from FY2023 to FY2025 ($1.18 to $2.09) as the loss-making investment phase ended and retained earnings compounded into book. With ROE at 30.3% and dilution under 1.5% per year, the math is mechanical: book per share grows at ROE minus dilution minus the (currently zero) payout ratio, which is roughly 28–29% per year today.
The arithmetic. At 20% through-cycle ROE and no payout, tangible book per share roughly doubles every four years. From $2.09 today, base case lands at $5.20 by 2030 and $11.50 by 2035. At a P/TBV of 4× on the 2030 number, that implies roughly $20 per share — about 65% above the May 2026 price of $12.18. The bull case requires either a higher ROE (25%+) for longer, or a higher exit multiple sustained by international optionality earning its capital. The bear case is not that book stops growing but that the multiple compresses faster than book compounds — a 3× P/TBV on $3.60 of 2030 book is below today's price.
Three observations on the cash side. OCF/NI has held at 1.22× for three consecutive years — earnings convert to cash above 100%, which is structurally what a deposit-funded compounder should do. Capex is essentially zero ($7M FY25 in the cash flow statement; $341M in MD&A capex that includes capitalized software, still under 4% of revenue). The capital intensity of growth is not the OCF/NI line; it is the Tier-1 capital constraint of growing an unsecured loan book — and at 6.6× assets/equity versus 13–14× at incumbents, Nu has roughly 2× the balance-sheet headroom before leverage starts diluting ROE.
4. Durability and Moat Tests
Five tests — three competitive and two financial — that determine whether the moat survives the next decade.
Reading the moat trajectory. The cost-to-serve advantage strengthens in absolute terms as scale grows but narrows versus the digital-peer cluster (Inter, C6) that runs on similar rails — net, the lever drifts from a 5 to a 4. The deposit-funding edge is the most vulnerable lever over a decade because a Selic normalization mechanically halves the absolute funding advantage even if percentage holds, and Open Finance enables structural deposit portability — most likely lever to step from 5 to 3. The credit-data, engagement, and scale-economy levers strengthen as cohorts mature and AI capex compounds. The contractual switching-cost lever is regulatorily suppressed and will not improve. Net: the moat broadens in some directions and narrows in others; the durability question is whether the structural cost edge and behavioral engagement are enough to substitute for a compressing funding gap.
5. Management and Capital Allocation Over a Cycle
Founder David Vélez has been CEO for 13 years through the IPO, two LatAm launches, the 2022–23 credit cycle, the profitability inflection, and now the AI/US pivot. His personal $11B stake (74.4% of votes on 18.6% of economics) is the largest absolute founder skin-in-the-game of any newly-listed bank globally, and his voluntary 2022 termination of a $423M Contingent Share Award — to save shareholders up to 2% of dilution — remains the single most concrete piece of alignment evidence in the file. CFO Guilherme do Lago has held the seat through the entire profitable phase; CEO Brazil Livia Chanes (the 80% of revenue) has operated through the 2024 NPL peak and recovery. Roberto Campos Neto's mid-2025 appointment as Executive Vice-Chairman has strong operating logic (regulatory dialogue, US bank build-out) but represents a textbook revolving-door optic for a bank whose payment-license framework he supervised seven months earlier.
The capital allocation track record is consistent with the long-term thesis: net income redeployed 100% into the loan book, Mexico/Colombia expansion, and now the US bank, with essentially no dividend, no material buyback ($6.2M FY25), and no M&A bloat (Hyperplane $1.5M FY25 and Tyme $250M minority stake are the largest disclosed transactions in three years). At 30% reported ROE and 6.6× leverage versus 13–14× at incumbents, the math says retaining all earnings is the right answer until either the international franchise is meaningfully profitable or Brazilian saturation closes the reinvestment runway — neither has happened yet.
The 10-year capital-allocation question is when management initiates a payout. At current pace, Brazil ARPAC compounding plus Mexico ramp should consume reinvestment capacity through 2028–2029; after that, a dividend or buyback program becomes the natural next step. Whether management initiates capital-return discipline before book per share compounds past a level where reinvesting at 20%+ ROE is no longer plausible is the test. Two precedents to watch: ITUB and BBD both initiated heavy dividend programs in their early-2000s maturation; Nu has not yet shown what its capital-return cadence looks like, and the disclosure machinery (founder-controlled FPI, no individual NEO comp, no Say-on-Pay) is weaker than it would be at a US domestic.
The bear-side governance critique is real but not load-bearing for the 10-year case: insider activity since IPO has been sell-only (Vélez August 2025 $462M block, Junqueira March 2026 $4M open-market sale, Sequoia 13G/A cut 57% in May 2025); the introduction of the Managerial P&L framework in Q4 2025 with ISAE 3000 limited assurance only reduces backward comparability; ~$243M of FY25 development spend was capitalized rather than expensed. None of this is fraud — Nu sits in the forensic "Watch" band, not the elevated band — but the machinery for catching aggressive judgment calls runs through fewer hands than at a US domestic, which argues for a margin-of-safety adjustment on the multiple rather than a thesis breaker.
6. Failure Modes
Six specific ways the long-term thesis breaks, each with observable early warnings rather than generic "execution risk."
The single thesis breaker. A Brazilian unsecured-credit recession layered onto a Selic normalization cycle in 2026–2028 — not either one alone, but both together. Each in isolation produces a one-year ROE compression that the operating machine absorbs; together they compress ROE toward 14–15% for two years, force coverage rebuilds that consume retained earnings, and break the "primary-deposit stickiness substitutes for switching costs" pillar simultaneously. The 10-year compounder thesis can survive either; it struggles with both at once.
7. What To Watch Over Years, Not Just Quarters
Five observable multi-year milestones, each with its time horizon and the validating/refuting reading.
The long-term thesis changes most if deposit cost stays below 92% of CDI on an $80B+ base in 2030 and Mexico ARPAC clears $15 with positive segment net income before 2029 — that combination converts the "Brazil credit-led compounder" thesis into the "LatAm consumer financial services platform" thesis the multiple is being asked to underwrite. The signal that closes the case fastest is the deposit-cost line through a Selic cut cycle; the signal that opens the optionality is the Mexico segment break-even durability.