Nubank: a great bank, but the price already pays for a decade of 28% returns

Nu is the largest digital bank in Latin America, with 139 million customers, more than $1 billion of profit in the latest quarter and a 33% return on equity. None of that is in question. The question is what the shares already pay for it. At $13.29 the stock trades at 4.9 times book value, and the only way to justify that is high-20s returns on equity for the next ten years, with the Brazilian real, credit losses and new tax rules all staying out of the way. I do not find a margin of safety in that.

At $13.29 on October 1, 2026, the shares trade 23.6% above my triangulated Base value of $10.75, so the Base is 19.1% below the price. The Bear value of $5.19 is 61.0% below it and the Bull value of $14.68 only 10.5% above it. The required margin of safety of 50% puts the maximum buy price at $5.38. For existing holders this is a hold, but not a position to add to. For new money it is a watchlist name.

Route: SF-02 Banks, digital bank with card, loan and deposit franchises. Economic classification: Compounder (10/12), valued with a ten-year residual income model, the sector method for a bank earning well above its cost of equity. Cross-check: forward earnings multiple. Tangible book value is shown for reference. Values are present fair-value estimates, not 12-month price targets.

What Nu is

Nu runs a branchless digital bank across three countries: almost 118 million customers in Brazil, 15.8 million in Mexico (16 million by July) and more than 5 million in Colombia. In Mexico it received its bank authorization from the regulator on July 9 and launched the bank in August. In Brazil it is adding a full banking license, and in the United States it holds a conditional approval from the OCC for a national bank (January 29, 2026). The group earns money from three sources: interest on credit cards ($26 billion of balances), unsecured loans ($10.3 billion) and secured loans ($3.1 billion), a spread on $45.3 billion of customer deposits, and fees.

The unit economics are the core of the thesis. The average active customer brings in about $17.10 a month of revenue and costs about $1.00 a month to serve, and 83.5% of customers are active in a given month. Operating expenses are 19.5% of revenue. The shares are split into Class A (3.8 billion) and Class B (1.02 billion), a dual-class structure that leaves minority holders with limited influence.

The latest quarter

The growth rates quoted by the company are at constant exchange rates. In dollars they are higher, because the Brazilian real strengthened. The company says part of the improvement in early delinquency is seasonal, and that the 90+ day figure rose because first-quarter early delinquencies migrated into it. It also says it has intentionally expanded into higher-risk, higher-return segments, and that is visible in the accounts, as the next section shows.

What the credit book really costs

For a bank the hard metric is profit after the full cost of credit and funding, measured against the equity that supports it, not revenue or margin before losses. The IFRS accounts make the cost of Nu's growth visible.

Nu is paid for this risk: the net interest margin of 22.9% leaves a risk-adjusted margin of 12.4%. The company reports that its allowance covers 90+ day balances 2.4 times. But the book grew 37% in a year, a quarter of the card book is now in the highest risk band, and the mix is moving up that band. Write-offs lag lending: loans made in the last twelve months have not yet shown their final losses. That is the central uncertainty behind the 33% return, and it is why I do not extrapolate it.

Two other items matter for the quality of book value. Deferred tax assets were $3.65 billion at June 30, 27.5% of equity, up $1.14 billion in six months. They build because credit provisions become tax deductible only when loans are written off, so they are real but depend on the lending continuing. And Nu advanced $185.5 million to the Brazilian Credit Guarantee Fund (FGC) in the first half under an emergency recapitalization plan approved in February 2026, which requires member banks to make contributions over several years.

Capital, funding and shares

Equity attributable to shareholders was $13.25 billion at June 30, 2026. Goodwill of $409 million and intangible assets of $747 million (mostly software developed in-house) bring tangible equity to $12.09 billion. Total assets were $82.75 billion, so equity is 16.0% of assets, a leverage of 6.2 times. The release does not give regulatory capital ratios, so I could not check them against the Basel minimums, and that matters because the new Brazilian banking license may bring different capital rules.

Funding is strong but the cushion is shrinking. Deposits of $45.3 billion cost 88% of the interbank rate, 3 points less than a year ago. But deposits are 1.15 times the credit portfolio, down from 1.34 times a year ago, because credit grew 37% and deposits only 18%, and borrowings are $4.7 billion. In Mexico, deposits fell for a second quarter by design, and loans are only 35% of deposits there.

There were 4.831 billion shares outstanding, after 40.66 million were bought back for $500.4 million in the half year (an average of $12.31, below today's price). Adding the shares that stock options and restricted stock units would create under the company's own treasury method gives 4.879 billion diluted shares, which I use for all per-share values. Another 67.8 million restricted stock units and 19.2 million options are outstanding, and share-based compensation cost $216.8 million in the half year, 11% of net income. Diluted book value is $2.72 per share and tangible book value $2.48.

Book value per share rose 17.9% in six months, but that includes a $371 million gain from currency translation as the real strengthened. Retained earnings, stock compensation and buybacks added 14.1%. Because Nu reports in dollars and earns in reais, pesos and Colombian pesos, a weaker real would reverse part of that gain, and I include a currency drag in the valuation.

Primary valuation: residual income

The sector-correct method for a bank that earns more than its cost of equity is residual income: value equals book value plus the present value of profits above the cost of equity. I model ten years explicitly and then a terminal period. The first draft instead said a fair price to book of "roughly 5.5 to 6.0x" and multiplied $2.74 by 5.8 to get $15.9, without a model. At a 27% return on equity and a 13% cost of equity, a 5.8x multiple means book value growing about 10% a year forever, in dollars. Growth of 6% to 8% a year gives only 3.0x to 3.8x.

Base case inputs, all estimates:

• Return on average equity of 27% for years 1 to 5 (the draft's normalized figure, below the 33% reported for Q2), fading to 20% by year 10, and 17% in the terminal period.
• Cost of equity of 13%, the middle of the draft's 12% to 14% range.
• 80% of earnings retained in years 1 to 5 and 70% in years 6 to 10; the rest returned through buybacks.
• A 3% annual currency translation drag on book value, for the expected depreciation of the real, peso and Colombian peso against the dollar.
• Terminal growth of 5%.

This gives $9.02 at June 30, 2026 and $9.30 on October 1, rolled forward at the cost of equity. That is 3.4 times book value. The terminal period is 22% of the value, so the answer does not hinge on the last years.

Secondary valuation: forward earnings

The cross-check uses next year's earnings and a market multiple. Consensus 2026 net income is about $4.17 billion (a third-party figure, estimated). The first half was $1.93 billion, so the second half must be $2.24 billion, 16% higher, which looks reachable given Q4 seasonality. I assume 2027 net income grows 20%, below the 27% consensus, to $5.0 billion, or $1.03 per diluted share. At 13x that is $13.33 at year-end 2026, or $12.93 today.

The 13x is an estimate. Nu's own forward multiple at year-end was 26x for 2023, 20x for 2024 and 21x for 2025 (data provider figures), and it is 14.9x on 2026 consensus now. I found no peer table on the same period and accounting basis, so the range of 11x to 15x is shown below. At the consensus 2027 figure of $1.13 per share, 13x gives $14.25, but consensus is not an input to my Base.

Historical cross-check and growth diagnostic

There is no long public history to anchor a valuation: Nu listed in December 2021, and its earnings grew from nothing, so its old multiples describe a different company. The shares are in the lower third of their 52-week range, which says the market has already marked down the growth premium, not that the stock is cheap.

The growth valuation cross-check (SM-19) is a diagnostic only. On consensus earnings the forward PEG is about 0.4, which looks very cheap. It is not meaningful as a valuation signal here. The growth it divides by comes from a credit book growing 37% into riskier segments, so part of the growth is borrowed from future credit losses, and the earnings per share consensus is a data provider's adjusted figure while the filings are IFRS. Growth durability: structural and normalizing. Cyclical distortion: medium, because Q2 is seasonally strong for credit and the real was strong. Flags: accounting basis mismatch (minor) and no synchronized peer table. SM-19 does not set fair value.

Valuation triangulation

The two methods diverge by 32.7%, above the framework's 30% threshold. The flag is raised, confidence goes down a grade and the margin of safety goes up. The gap has a clear cause. Residual income charges the full cost of equity and lets returns fade, while the earnings multiple uses next year's visible profits and the price the market is paying for them. The residual income method has the larger weight because it is the sector method and does not borrow the market's own multiple, but neither method is precise. The Bear and Bull values use both methods.

What is the market already pricing in?

I ran the residual income model backwards. To justify $13.29 at a 13% cost of equity, Nu must earn about 28.5% on average equity every year for ten years (26% at a 12% cost of equity, 30.5% at 14%). My Base is 27% for five years, then a fade to 20%. At a return on equity of 27% and a 13% cost of equity, today's 4.9x book value means book value growing 9.4% a year forever.

The price therefore needs Nu to hold its recent returns for a decade: 28%, 29% and 33% in successive quarters is a good start, but the credit book that produces them is only a few quarters old in its current risk mix. For upside, an investor must believe that returns stay near 30%, credit losses stay inside the pricing, Mexico and Colombia earn Brazilian returns, and nothing in the tax or capital rules changes.

Bear, base and bull scenarios

Bear: credit losses rise as the new risk mix matures, returns fall to the low 20s and then to the mid teens, and the market pays a trough multiple. Base: returns stay at 27% for five years while credit is controlled, Mexico scales profitably and book value compounds strongly, then competition and size pull returns toward 20%. Bull: returns stay above 30% for five years, Mexico reproduces Brazil's economics, Colombia scales and underwriting stays superior. The cost of equity (13%), the currency drag and the retention rate are the same in all three cases, so the scenarios differ only in operating performance and the same risk is not counted twice. The probability-weighted value, at 25%, 50% and 25%, is $10.34.

Sensitivity: residual income value per share

Even the most favorable cell, a 30% return held for five years at a 12% cost of equity, gives $12.97 on this method, slightly below the price.

How the buy price is set

Confidence C starts at a 40% margin of safety. Five points are added because the two methods diverge by more than 30%, and five for the credit cycle and the concentration in unsecured consumer credit in Brazil. Nothing is subtracted: the balance sheet is strong but not predictable enough for a discount. That gives 50%. A 50% discount to the $10.75 Base gives a maximum buy price of $5.38. A deeper 60% discount gives $4.30, so the buy zone is $4.30 to $5.38. The current price is 2.5 times the top of that zone. A margin this large is a statement about uncertainty in the model, not a forecast that the stock will fall.

Dividend

Nu pays no dividend and is not an income thesis. Capital allocation is judged on book value per share growth and returns on the capital it keeps. In the first half it bought back $500.4 million of shares, about 1.6% of the market value of $64 billion on an annualized basis, and the share count fell 0.5% since December.

Reasons to own NU

• Exceptional returns. A 33% return on equity in Q2 and 38% on tangible equity, with the efficiency ratio at 19.5%, are far above what traditional banks earn.
• A scale franchise. 139 million customers, 83.5% of them active, $17.10 of monthly revenue against $1.00 of cost to serve, and the largest base of small business customers in Brazil (6.8 million).
• Evidence on credit. The company says 90+ day delinquency has improved in each income band since July 2025 while peer segments worsened, and that customers using Nu as their main bank have about half the delinquency of the average. Allowances cover 90+ day balances 2.4 times.
• Cheap, growing funding. $45.3 billion of deposits at 88% of the interbank rate, exceeding the credit book.
• A long runway. A Mexican bank with 16.5% of adults as customers, Colombia, a Brazilian banking license and a conditional US charter, and AI underwriting in production on three portfolios.

What could go wrong?

• Credit. The portfolio grew 37% while 90+ day delinquency reached 6.9% and the higher-risk share of the book rose. A downturn in Brazilian household credit would hit earnings and book value together. This is the main thesis breaker.
• Currency. Book value and earnings are translated from reais, pesos and Colombian pesos. A $371 million translation gain flattered book value in the half year, and a weaker real would reverse it.
• Tax and regulation. From January 1, 2027, Brazilian financial services move to the new IBS and CBS consumption taxes at an initial 10.85% rate, rising to 12.5% by 2033. The company is still assessing the impact. The Brazilian banking license, the FGC contributions and the Basel capital rules could also change capital needs.
• Capital allocation. Mexico, Colombia and the US absorb capital before they earn Brazilian returns. On September 26 and 28, reports said Nu was in talks to buy the UK bank Monzo for GBP 8 billion to GBP 10 billion; the shares fell 10.0% on September 28. On September 30 Nu said it is not pursuing a transaction, and the shares recovered, but the episode shows how far the company is prepared to look.
• Valuation and governance. At 4.9x book value, small changes in assumed returns move the value a lot, and the dual-class structure limits minority influence.

Management execution

Management has delivered a rising return on equity (28%, 29%, 33%), the first $1 billion quarter, a Mexican bank launch on schedule and $500 million of buybacks at an average price below today's. It also discloses its credit risk in detail and says openly that it is expanding into higher-risk segments on purpose. Against that, the efficiency ratio rose from 17.6% to 19.5% in a quarter, the company does not give earnings guidance that could be tested, and its Monzo statement came only after the shares had fallen. Management credibility scores 16/20, good.

Stock Analyza scorecard

The valuation score is 26/100: discount to Base 8/40, protection against the Bear 3/20, agreement between methods 5/15, historical valuation 5/10, market implied expectations 3/10, data and model penalty 2/5. The business scores high, the price does not.

Confidence score

What I would watch from here

Two of these readings are yellow today, which is another reason the verdict is not a buy. The next review points are Q3 2026 results, expected around November 12, and the company's first Investor Day, reported for December 8, 2026.

The three most important thesis breakers

1. 90+ day delinquency above 7.5%, or cost of credit rising faster than the net interest margin.
2. Return on equity below about 22% for more than two quarters.
3. International expansion or an acquisition that consumes capital without returns above the cost of equity.

Adversarial review

I rebuilt every figure from the second quarter 6-K and the interim financial statements. The first draft's company data were right: customers, revenue, net income, return on equity, delinquency, credit portfolio, deposits and equity of $13.25 billion all match the filings, as does its arithmetic (price to book, 2.74 times 5.8, the 30% margin of safety). The valuation did not hold up.

What changed and why. First, the primary value was a multiple chosen by judgment, and it was the whole valuation: 5.8x book implies about 10% book value growth in dollars forever, and the Bull value of $22 implies 8x book and more than 10% forever. Second, the secondary value was a range with no earnings path or multiple, and its 3.1% agreement with the primary was therefore not independent evidence. Third, the sensitivity grid multiplied "forward" book value by a multiple without discounting; its lowest book value of $2.75 is today's $2.74, so the grid mixed today's and future values. Fourth, the draft counted basic shares (and called them issued shares) and used no tangible book value, although the bank framework works on tangible book. Fifth, the business quality components added to 88 but the headline said 85 with no component changed, and the confidence of 20/25 and 25% base margin of safety ignored that the return on equity, the cost of equity and the multiple are all estimates; its extra 5% for emerging-market complexity is not a framework modifier. Sixth, the draft left out the currency effect on dollar book value, the deposit cushion falling from 1.34 to 1.15 times, the rise in higher-risk lending, the FGC advance, Brazil's 2027 tax reform and the Monzo-related 10% drop. It also ended by ranking NU against other stocks, which does not belong in an analysis. The verdict is unchanged: hold, but "do not add" and a far lower buy price.

The strongest counter-argument is that Nu's returns are structural: a 19.5% efficiency ratio, $1 cost to serve, cheap deposits and AI-driven underwriting could keep returns above 30% for years, and the market has a long record of pricing quality banks above their book. That is plausible. It is the Bull case, and my model reaches the draft's $16 only when three favorable inputs hold together: a 30% return for five years, a 12% cost of equity and a 2% currency drag. The Base does not assume them before the credit book has been through a cycle.

The most generous combination is a 30% return fading to 22%, a 12% cost of equity, a 2% currency drag, a 19% terminal return and 15x on consensus 2027 earnings. The most cautious is a 24% return for three years fading to 18%, a 14% cost of equity, a 4% drag, a 15% terminal return and 9x on earnings of $0.78. Single changes keep the Base between $9.52 and $11.76, so the maximum buy price stays between $4.76 and $5.88, far below $13.29 every time. Only the stacked generous case, at $16.15, is above the price, and even there the maximum buy price is $8.07. All corrections lowered the value, so I tested the other direction: the consensus-based test and the generous case both raise the Base, but not enough to change the verdict. The publication gate status is PASS_WITH_WARNING.

Final verdict: hold at $13.29, do not add

Nu has the economics of an exceptional bank: scale, low cost to serve, cheap deposits and returns above 30% at the moment. But the price of 4.9x book value already assumes that those returns last a decade while a fast-growing, increasingly risky credit book goes through its first full cycle.

Verdict: HOLD at $13.29, do not add. Triangulated Base value $10.75, downside to Base 19.1%, Confidence C (14/25). Required margin of safety 50%, maximum buy price $5.38.

For an existing holder, the franchise and the returns support holding, and a very large position is a reason to trim, not to add. New money waits for $5.38 or below with the thesis intact, or for evidence that returns and credit quality hold through a cycle.

The main thesis breaker is a deterioration in credit quality that pushes returns on equity toward the low 20s.

Sources

Primary sources include Nu's second quarter 2026 earnings release (Form 6-K), the interim financial statements for the period to June 30, 2026 and the Q2 2026 earnings call. The market price, price history and consensus earnings come from StockAnalysis, which uses S&P Global data. The company's statement on the Monzo reports is here. Diluted share count, tangible book value, the residual income model inputs, the earnings multiples and the scenario figures are my own calculations or estimates, labeled as such.

The reference market price is $13.29, the October 1, 2026 regular session close. Balance sheet figures are as of June 30, 2026. Growth rates are at constant exchange rates unless stated.