Nu Holdings

IThe business

Nu Holdings owns Nubank, a bank with no branches. It started in São Paulo in 2013 with a purple credit card and no annual fee. At 30 June 2026 it had 138.9M customers: almost 118M in Brazil, 15.8M in Mexico and more than 5M in Colombia.⁵ About 83.5% of them used it in the month. It is a Cayman company listed in New York, reports in US dollars under IFRS, and files in English as a foreign issuer.²

It has not reported Q3 yet. Nasdaq lists 12 November, which the company has not confirmed in a filing.¹¹ So TTM on this page means the four quarters to 30 June 2026. In those four quarters IFRS revenue was $19.3B and net income $3.6B C (calculated).²³⁴ In Q2 alone net income passed $1B for the first time.⁵

The idea is simple. An incumbent Brazilian bank carries thousands of branches and tens of thousands of staff.² Nu has about 10,400 people and serves a customer for about $1 a month.⁶⁵ That gap let it charge no fees on the account, take customers the big banks did not want, learn from their payments, and lend to them. The lending is where most of the money is made.

How it makes money

Since Q4 2025 the company has shown a managerial income statement beside the IFRS one. It has limited assurance from KPMG but it is not IFRS.¹⁷ It is the clearest way to see the three engines. In Q2 managerial revenue was $5.88B.⁵

• Credit: $3.60B of income, 61% of revenue C (calculated). Credit cards are $26B of the book, unsecured loans $10.3B, secured loans $3.1B.⁵ Unsecured loans grew 45% at constant currency.⁶ After credit costs, credit gave 41% of gross profit.⁵
• Float: $1.45B, what it earns on deposits it has not lent. It holds about $42B of cash, government paper and central bank deposits C (calculated).³ Float gave 34% of gross profit, a share that exists because Brazilian rates are so high.⁵¹⁸
• Fees: $0.82B, mostly interchange on $43.4B of quarterly card spending.⁵ Fees gave 25% of gross profit.

Revenue per active customer was $17.1 a month in Q2, up 22% at constant currency. Cost to serve was $1.0.⁵ That spread is the whole model. Vélez said the oldest cohorts earn in the mid-$20s and incumbents $40 to $45, though he doesn’t expect Nu to reach that, because of fees it doesn’t charge and products it doesn’t sell.⁶ Those are company figures from a call, not filed numbers.

Where the growth comes from

There are three places. First, Brazil, where customer growth is near its end and revenue per customer is not. Nu says it is the primary account for over 62% of active customers.² In July it launched Croma for what it calls Super Core customers, a band between the mass market and the rich.⁵ It also serves 6.8M small businesses, by its count the most of any Brazilian institution, and about a third of the market.⁶

Second, Mexico. Nu reaches 16.5% of Mexican adults, the same share as Brazil in 2020, and earns $12.3 a month from each customer against $5.6 in Brazil at that stage.⁵ It broke even early in 2026, and became a full bank on 6 August.¹⁹³ A bank license brings payroll deposits and higher deposit insurance.⁶ Vélez’s base view is that Mexico could be 60% to 70% of Brazil.⁶

Third, the rest of the world, which is an option and not yet a business. Nu opened in the United States on 10 September through Lead Bank, with 3.50% on deposits and 1.5% cash back. It has preliminary conditional approval for a national bank charter. A new multi-currency account, Nu Global, pays yield in stablecoins.²⁰ Management caps the US cost at under 1 point of the efficiency ratio in each of 2026 and 2027.²¹ On 30 September it said it is not pursuing Monzo, after press reports.²²

What happened in October

Nothing happened at the company. Brazil voted. Lula won a fourth term on 25 October by about 51.8% to 48.2% on the count the press reported, over Flávio Bolsonaro.²³²⁴ The next day the Ibovespa hit its circuit breaker, the Treasury canceled two bond auctions, and the real fell hard.²⁴ The index lost about 18% in the week on my arithmetic from its closes, from 205,083 to 167,922; the press called it the worst week since March 2020.²⁵ Alejo Czerwonko at UBS put the worry in one line: “Investors want a clear and credible commitment to fiscal discipline.”²⁴

Nu fell 23.4% in a week with no news of its own. Two things sit behind that number. The Brazilian names on this table fell more or less together, so this was a country being repriced. And the stock had risen 13% on 5 October, the first session after the first round, so part of the fall gave back a bet on the election C (calculated).¹ The $11.61 close is a cent above the year’s low close, set in June.¹

IIThe moat

Most banks are protected by habit and regulation. Nu’s protection comes from being cheaper, and from what it learned by being cheaper. Each source below is real, but they aren’t equally strong, and the last one is out of the company’s hands.

The primary account is the one that matters

A customer whose salary lands at Nu pays Nu first. Nu’s data on that person is better than any bureau’s. Management says delinquency among customers with Nu as their primary bank is roughly half the portfolio average, which it calls being senior in the credit stack.⁵⁶ That is a strong claim, and a plausible one. It is also the company’s own cut of its own data, and it can’t be checked from outside. A better test is a full credit cycle, and the book hasn’t seen one at this size.

The data and the model

Nu built a foundation model for financial behavior, NuFormer, trained on more than a decade of transactions. It runs in cards in Brazil and Mexico and in Brazilian unsecured lending.⁵ Rob Livingston, the CFO, said on the call that these models beat logistic regression, which he called incontrovertible.⁶ I believe that. Whether it lasts is a different question. Every large Brazilian bank has more history, and Itaú launched its own AI product this year.⁷ A model is an edge only while its data is better than the next one’s.

What could wear it down

Pix is the example that cuts both ways. The central bank’s free instant payment system helped Nu take primary relationships, and 81% of Brazilians have used it.² It also means that no bank owns the rail, and that moving money away costs nothing. Open finance points the same way. The same state that opened the door for Nu can open it for the next entrant. MercadoLibre’s fintech arm has 88M monthly users, up 29%, and it is growing credit fast: its provision for doubtful accounts nearly doubled to $1.3B in Q2 C (calculated).³⁰

The rest is size. Nu has about 7% of the profit pool it targets, on Vélez’s own figure, and he puts Brazil’s gross profit pool at $100B.⁶ Small share is why it can pick its customers today. It is also why the protection is not yet tested against a real response from the incumbents.

IIIThe competitive advantage

Put Nu beside Brazil’s banks and two things stand out at once. Its costs are far lower. Its bad loans are far higher. Both are true because it lends small unsecured amounts to people the banks largely ignored.

Itaú’s book is about 7.5 times Nu’s, and its overdue ratio is 1.9% against 6.9% C (calculated). That doesn’t mean Itaú lends better. It lends secured money to richer people. What matters is whether the rate Nu charges covers the loss. In Q2 it did, by a wide margin. Credit cost $1.69B on a $39.4B book, about 17% annualized, and risk-adjusted net interest margin was a record 12.4% C (calculated).⁵ Write-offs in H1 were $2.33B against $1.51B a year earlier C (calculated).³

Q1 is the warning. Risk-adjusted margin was 9.5%, and 15-to-90-day delinquency rose 89 basis points in the quarter, mostly on seasonality by the company’s bridge.¹⁹²¹ Management told investors to expect a return to the level of H2 2025, which analysts on the Q2 call put at about 10.5% to 10.8%. Q2 came in well above that. Livingston said about a third of the beat was Desenrola, a government debt renegotiation program, and two-thirds credit and balances.⁶ He then said he expects to stay in that region, but corrected an analyst who called 12% a floor.⁶ On that same call Mario Pierry of Bank of America pressed on why 90+ delinquency has trended up for two years; Livingston put it down to mix.⁶ Intentional risk expansion added 24 basis points to early delinquency in Q2.⁵⁶

The banks are not having a better year. Bradesco said in October that the macro environment keeps pressure on delinquency, and it raised R$10B of equity.³¹ So the question is relative, and Nu’s relative position looks good so far. Its own figures show 90+ card delinquency steady in each income band since July 2025 while peer banks worsened.⁵

Funding is the quiet advantage

Nu pays 88% of the interbank rate on deposits. It also owes $15.5B to card networks, money that costs nothing because customers’ card purchases settle later.³⁵ The credit book is 87% of deposits. Net of those payables it is 53% C (calculated). In Mexico the ratio is 35%.⁵ As Vélez said, each dollar moved from government paper into a loan earns far more.⁶ The balance sheet is barely levered: assets are 6.2 times equity C (calculated).³ That’s the margin source that needs no new customers.

IVManagement

Who is in charge

David Vélez founded the company after running Sequoia’s Latin American investments. He is chairman and CEO, and he controls the votes. At March he and his wife held about 906M shares, 18.6% of the company by my count and 74.4% of the votes, through Class B stock with 20 votes each C (calculated).² At $11.61 that stake is about $10.5B. This is a lot of his own money, and he can’t be removed. He has also joined the OpenAI board: an analyst raised it on the Q2 call and he confirmed it.⁶

How they are paid, and what that pays them to do

Cayman law doesn’t require pay by person, and Nu doesn’t give it. Directors and key management together got $91.3M in 2025.² Pay is salary, profit sharing, and mostly restricted stock on time vesting, with “market conditions” awards for some.²

The most useful document is a Form 4 from this month. On 15 October Vélez received 13,577,733 restricted shares, about $200M at that day’s $14.73 close C (calculated). They vest over ten years to 1 January 2036 and replace his annual awards from 2027 to 2031. The form states only continued service as the condition.⁸¹ That is a long tie, which I like. It also pays him for staying, not for a return, a capital ratio or a credit outcome, and I found no such test in what I could read. A founder with this much stock has the right incentive anyway. The award is a small part of his wealth: about 1.5% of his holding.

People at the top do sell, in modest size. Cristina Junqueira, the co-founder, reported sales of 465,000 shares in 2026 at $14.71 to $15.80, a small part of her 131M. Henrique Fragelli, the Chief Risk Officer, sold 221,707 in August at $15.81.³²² I don’t read much into that. Directors got their stock grants in August, and none reported buying after the fall in the filings I saw through 27 October.³²

Turnover this year

Guilherme Lago left the CFO seat in July after five years, for Rob Livingston, who ran finance for Visa in North America.¹⁰⁶ Eric Young, the CTO, joined in August 2025 and leaves by 12 November. Vélez will run Global Platforms himself for now, in a year when the company calls AI its core priority.⁹²¹⁹ Roberto Campos Neto, who ran the central bank until 2024, joined in 2025 as Executive Vice-Chairman.² I can’t judge what his role is worth with Lula in office for four more years.

What they did with capital

The record over the one full cycle that exists is decent, and short. The company kept its IPO money through the 2022 credit scare, when the stock fell to $3.31, and that reserve let it grow while others pulled back.¹³³ It has paid no dividend. Its first buyback came in June 2026, at an average of $12.31.³³⁴ That’s good, and it’s a bit above today’s price. $499.6M remains under the program.³

Then there’s the Monzo episode. The company denied it. It added that its capital framework is unchanged and rests on returns against the cost of capital and value per share.²² Those are the right words. The United States is where they will be tested. Vélez called it extremely challenging himself, and said capital will go in stages.²¹

VFinancials

Nu went from a $365M loss in 2022 to $3.6B of net income in the TTM. Equity went from $4.9B to $13.3B with no new shares of size.¹⁴³ Return on equity was 32% in the TTM C (calculated). A few things need care before those numbers are taken at face value.

Dollars, constant currency, and two income statements

The company’s headline growth is FX-neutral. In dollars Q2 managerial revenue rose 56% against the 39% in the release, because the real was much stronger than a year before C (calculated).⁵²⁷ The real averaged 5.05 per dollar in Q2 and closed October at 5.42.²⁷ At 5.42, Q2’s Brazilian earnings would be about 7% lower in dollars C (calculated). Good years in dollars have partly been the real’s good years. It runs the other way too, and it did last week.

The tax rate is doing a lot of work

This is the number I would look at first. In H1 2026 pre-tax income rose 31%. Net income rose 62%. The difference is tax. The IFRS rate fell from 28.7% in H1 2025 to 11.8% in H1 2026 C (calculated).³⁴

Lago addressed this in May. He said the 8.7% in Q1 reflected structural changes to the global operating and corporate structure, and that it is recurring. He guided the IFRS rate to 15% to 20% for the rest of 2026, and the managerial rate to 30% to 35%.²¹ The note in the statements shows the pieces in H1: $281M from different rates in subsidiaries and the parent, $109M from interest on capital, and $254M of “other”, which covers foreign tax credits, tax-exempt sovereign bonds and incentives.³ The managerial statement grosses revenue up for all of this: $393.9M was moved from tax into other lines in Q2, so its 34.9% rate is about presentation and leaves net income unchanged C (calculated).¹⁷⁵

I don’t think anything is wrong here. Brazil’s rate on these entities went up, by law: CSLL on payment institutions rises to 15% and on finance companies to 20% from 2028.³ A structure that pays much less than that is a good thing to have, and a thing a government short of money might look at. At the 2025 rate of 25.8%, TTM net income would be $3.3B and return on equity about 29% C (calculated). The 33% headline is real, and flattered. Q2’s $175M tax line came with a $671M deferred benefit, while cash tax went up, so the current tax charge was $846M, more than double a year earlier.³

Cash flow doesn’t mean what it means elsewhere

Operating cash flow was −$1.2B in H1 2026 and +$3.6B a year earlier.³ That’s loan growth and deposit flows, and neither tells an owner much. For a bank, the honest test is how much of the profit has to stay inside to back the book.

Group equity is $13.25B. Regulatory capital across the three countries is about $6.2B, and the minimum about $4.2B C (calculated).³ In February Lago split December’s total capital of $8.9B into $3.6B required, $2.2B excess in operating entities and $3.0B at the holding company.³⁷ The rest of equity is mostly things regulators don’t count, and $3.65B of it is deferred tax assets, largely on credit provisions.³ Risk-weighted assets in Brazil rose from $20.1B to $35.7B in eighteen months, and the common equity ratio there went from 14.7% to 11.9%.³² Minimum capital is 10.5% of risk-weighted assets; total capital is 15.7%.³

So $3.6B of net income is about $3.2B an owner could take out if the book stood still in real terms, or about $2.8B at last year’s tax rate E (estimate). That’s 18 and 20 times earnings of that kind, against 15.8 times reported earnings per share of $0.73 C (calculated). And at this rate of growth, about $1.8B was truly free E (estimate). It’s a growth company wearing a bank’s balance sheet: the faster it grows, the less an owner receives today. That’s fine while each new dollar of capital earns a lot. On my numbers it does: net income is about 9% of risk-weighted assets, so about 60% of the 15% capital I’d keep against them C (calculated). That’s the number that makes the growth worth paying for, if it survives a credit cycle.

VIValuation

Start with how this goes wrong

Before any numbers I ask what would most likely destroy this business, or the investment, and work back from there. For a lender in Brazil the list is short.

Notice what is not on the list. Technology rivals are not. Slow customer growth is not. Those would hurt the value and leave the company standing. The three above are how an owner loses most of the money, and all three start outside the company. There’s a fourth way to lose: pay too much. That’s what the rest of this section is for.

What the price implies

At $11.61 the market value is $56.8B. That’s 15.8 times TTM earnings, 4.3 times book and 4.7 times book without goodwill and intangibles C (calculated).¹³ Consensus earnings are $0.86 for 2026 and $1.16 for 2027, though the feed may not reflect last week, which gives 13.5 and 10.0 times C (calculated).³⁹ Itaú is about 8.6 times its Q2 run rate at today’s real C (calculated).⁷⁴⁰²⁷ So the market pays Nu a premium over the best incumbent, and a much smaller one than a week ago, when the same arithmetic gave 20.6 times against about 11 C (calculated).

A bank’s price-to-book ratio can be read directly. It equals return on equity less growth, over the discount rate less growth. At 4.3 times book, a 12.5% rate and 8% growth for a long time, the price needs a return on equity of about 27% for good C (calculated). Nu’s is 32% now and 29% at the old tax rate. That’s a simple way to see it: the price needs the returns to stay about where they are, and no credit in between.

My model does the same thing year by year. Run backwards, holding its margin, tax, capital and currency, the price needs revenue per customer to grow 5.7% a year at constant currency to 2035. Net income would then grow about 6% a year in dollars from 2026, against 9% in my base E (estimate).

In the last twelve months that figure grew 22%. Brazil’s inflation alone is 4.6%.⁵¹⁸ 5.7% is a low bar in reais. The currency is what eats it.

What an owner can take out

I value the cash left after the book has what it needs: net income, which is already after stock pay, less 18% of the rise in risk-weighted assets. The 18% is a 15% capital ratio, about where Brazil sits now, plus 3 points for deferred tax and software that regulators don’t count. On my path that retention takes 34% of net income in 2027 and 10% by 2035 E (estimate). I add $2.7B of capital above that target at face value. The holding company share of that is my estimate, from the $3.0B Lago gave for December less the June buyback.³⁷³

The gap between reported earnings and owner cash is the point. TTM net income is $3.6B. Owner cash at a standstill is about $3.2B, or 88% of it, and about 50% at this rate of growth E (estimate). I use the second kind of number in the model’s early years and the first by the end.

The discount rate, and what I can’t judge

I discount dollar cash flows at 12.5%. The capital asset pricing model gives 12.0%: the 4.94% ten-year Treasury yield, plus a 0.95 beta against the S&P 500 over 58 months, times 7.47%, which is Aswath Damodaran’s mature market premium of 4.23% plus 3.24% for Brazil C (calculated).⁴¹⁴³⁴² I add half a point because that country premium is from January, before the worst week in Brazilian stocks since 2020. That half point is a judgment. I can’t measure the right number today, so I show the range.

Three inputs here can’t be known, and I’d rather say that than hide it in a decimal.

• Brazil’s premium. It rests on who runs the finance ministry and what the budget does. As I write, nobody knows.²⁴²⁵
• The tax rate. The company says it is structural. I can read the reconciliation but not the structure.²¹³
• Normal credit losses. This book has grown about tenfold since 2020 and has never been through a deep Brazilian recession at this size. Q1 and Q2 of the same year gave risk-adjusted margins of 9.5% and 12.4%.⁵

The range runs from $9.74 to $19.01, and the price sits inside it. At 14.5% and 3% a year off the real, value equals the price C (calculated). Put the other way: the market asks for about 14.5% in dollars on my cash flows, which is 2 points over what I’d ask.

Bear, base and bull

My base has 190M customers in 2035 and $27.01 a month from each in dollars, which is about $35 at today’s currency. That’s above the mid-$20s of the oldest cohorts today and well below the $40 to $45 the company gives for incumbents.⁶ Net income reaches $9.4B, about what Itaú earns now in dollars C (calculated).⁷²⁷ Terminal value is 49% of the base, a low share for this site, because a stock at about 11 times my 2027 earnings doesn’t need much from the far future.

The bear isn’t a disaster. It has a credit year or two, a real that keeps losing 6% a year, and net income of $2.8B in 2027, about a third below 2026. The worse outcomes are in the table of three failures above and in the stress table below; they don’t belong in a case with a value on it, since they depend on a state decision I can’t size.

Set the three side by side and my weighted value, at 35%, 50% and 15%, is $13.37. I weight the bear heavily since the risk just rose. The weights are my judgment too.

The margin of safety

My margin-of-safety price is $9.87: the bear value of $7.40 divided by 0.75, the same rule as on every page here. At that price a holder would lose about a quarter if the bear happened, and would earn about 10.0% a year in dollars from there on my bear path. The price is 18% above it. The reason I set it from the bear is that the base is the case I can’t defend against a Brazilian crisis.

The price against the alternatives

A dollar here has to beat the next best use, and in Brazil’s case that has a strong rival: the overnight rate of 13.75%, in reais, with nearly no credit risk.¹⁸

So the choice isn’t Nu against Treasuries. It’s about 14.5% on my base and 8.9% on my bear, against 10% or so that needs no one to be right about credit. American Express is the best comparable on this site. It has the same kind of return on equity, 34.5%, a far safer country, and a price level with my value at a 9.5% rate.⁴⁴ Nu offers about 5 points more in the base case, for a risk of a very different kind. Itaú offers an earnings yield near 12%, slower growth and the same country.

Quality and price are two questions

On quality: this is a very good business. Low cost is the most durable edge in banking, and it’s earning about 60 cents a year on each dollar of required capital. On price: 19% below my base, in a week when the whole country was marked down. I don’t think the price is unreasonable, and the market may well be right. Its case is plain. The fiscal path may be bad. The tax benefit may not last. The credit book is young, and it just expanded risk on purpose at the top of a cycle in rates. A good business in a country that can change its rules deserves a higher rate than my 12.5%, and at 14.5% there’s nothing left. This looks like a very good business at a price that pays me for Brazil, if Brazil gets no worse.

A simple argument beats a long one, so I multiplied mine out. My bull needs five things at once: Mexico near Brazil’s size, say 50%; the upper-income push working, 60%; margin at 33%, 50%; tax at 33%, 65%; and the real losing 1.5% a year, 40%. I would not lean on those odds, and they aren’t independent. Multiplied, they give about 4%. The base needs less, and each miss takes off 4% to 23%, as the stress table shows. The price needs less again, but it does need the first failure not to happen.

Where this could be in 5 and 10 years

Each value is my model’s value at that date: the present value, in that date’s dollars, of the owner cash that comes after it, at 12.5%. It uses today’s 4,892M shares and assumes no buybacks. Owner cash is summed, not discounted, and includes $0.56 a share of spare capital that the model counts at face value, which I treat as paid out in 2026. Dates count from the end of October 2026.C (calculated)

Dollars per share, rounded to 10 cents. The return is the October 2036 value plus that cash, against $11.61, over ten years. It is not the “Return a year” row in the three-cases table, which solves for the rate. Here the 2036 value is set at 12.5%.C (calculated) Values are my estimates.

For the base to hold at five years, customers reach 182M in 2031, revenue $43.8B, net income $8.0B and owner cash $6.6B. At ten years, in 2035, it is 190M, $51.6B, $9.4B and $8.5B. That takes a pre-tax margin near 28%, a tax rate of 35% and 18% of the rise in risk-weighted assets held back as capital. Value climbs because the cash gets closer, and the base has $12.40 a share of owner cash to give back on the way.E (estimate)

The cases differ most on three numbers: the real, falling 6%, 3% or 1.5% a year; the 2035 pre-tax margin, 23%, 28% or 33%; and growth in revenue per customer, 5.9%, 8.9% or 10.7%. The 2036 value is all cash after 2035, so the rate matters too: at 14% the base is $19.30, not $22.70.E (estimate)

The most likely way the base is wrong is the real. If it falls 6% a year, not 3%, the five-year value is $14.20 and the ten-year value $17.70.C (calculated)

These are estimates of value, not forecasts of the share price. The market price can sit above or below value for years.

VIIThe case for the price

What most people miss

First, capital. The 33% return on equity is earned on a pile of equity far larger than the book needs. At June net income was about 9% of risk-weighted assets, and at a 15% capital ratio that’s about 60% on the capital the business needs C (calculated).³ Second, the balance sheet is under-lent. Net of the free $15.5B owed to card networks, the credit book is 53% of deposits, and Mexico is at 35%.³⁵ So margin can rise without one new customer. Third, the buyback has started, and $499.6M is left to spend, at prices below the $12.31 paid so far.³

The bull case

Management makes the strongest version of it. On the Q2 call Vélez said Nu has about 7% of a profit pool he puts at $100B gross in Brazil, that older cohorts earn in the mid-$20s against $17 today, and that Mexico could be 60% to 70% of Brazil, or as large if real-time payments take hold.⁶ Livingston said a risk-adjusted margin near 12% looks sustainable, though not a floor.⁶ In May Lago called the lower tax rate a recurring feature of the structure.²¹

Large investors have put money behind it. Baillie Gifford held 229.9M shares, 5.99% of Class A, at 30 June, and BlackRock 8.4% at 31 March.⁴⁵⁴⁶ Coatue held 29.3M and Tiger Global 11.0M at 30 June, both unchanged on the quarter.⁴⁷⁴⁸ ARK held 1.1M.⁴⁹ Those are positions, not arguments, and they predate the vote. Berkshire Hathaway held 40.2M at the end of 2024 and none at March 2025.⁵⁰ I found no published view on this company from the other investors the page usually checks, and my search was cut short.

Against reality

Today revenue per customer is growing 22% at constant currency, the price needs 5.7% and my base 8.9%.⁵ So on that line reality is well ahead. Risk-adjusted margin averaged about 11% in H1, about what my base needs C (calculated). Tax is 14% against a guide of 15% to 20%. The weak line is the one the company doesn’t control: the real lost 7% in a week, and my base has 3% a year from here.²⁷

If they keep margin near 11%, keep tax under my 35% managerial rate, and add 50M customers by 2035, mostly in Mexico, then a price of $14.30 is possible at 12.5%. The bull’s $24.20 needs Mexico close to Brazil and a stable real. On H1’s numbers they are on the base path, in the only months of data I have since the risk expansions. The test is a full year of credit, and it hasn’t happened.

What a 10x would take. Ten times $11.61 is $116.10, about $568.0B of market value. By 2036 that’s 26% a year C (calculated). At 15 times earnings it needs about $37.9B of net income, about four times Itaú’s run rate today and twice my bull’s 2035 figure C (calculated).⁷²⁷ At a 20% net margin that’s about $189.3B of revenue. On 200M active customers it’s about $79 a month each, nearly twice the $40 to $45 the company gives for incumbents.⁶ It would also be about two-fifths of the $100B gross profit pool Vélez cites, as net profit C (calculated). Tenfold runs from about $57B have happened in technology and chips. Micron has done more than that since April 2025, from about $70B.⁵¹ I found no bank that has, and banks can only grow with their country’s credit. It’s close to impossible.

Where I come down

The return on required capital moved me: it’s the reason a bank can deserve this multiple. The election didn’t move my view of the business, and it moved my rate by half a point. What would raise my values is a credible fiscal plan and a full credit year with margin near 11%, which would let me drop that half point and lean toward the bull.

VIIIRisks, and what would change my view

Credit comes first. Nu lends unsecured to lower-income people at high rates in a country with a 13.75% policy rate. Its allowance of $6.6B is 16.9% of the book, and the full weight of its downside scenario would add only about $0.5B C (calculated).³ That scenario is the model’s own; a real recession may be worse. Management says it assumes the future is worse than the past in every loan, and that short loans let it react fast.⁶ Both are true of good lenders and of lenders who then get hit.

The currency is second. A US-listed share in a dollar-reporting company holds a claim on reais. From 2014 to 2025 the real lost 6.4% a year.²⁷ The company began a net investment hedge on part of its Brazilian holdings in 2026 and shows a $120.6M loss on it in equity, but that covers a portion, and for a while only.³

The state is third. Brazil has already raised tax on fintechs by law and capped prepaid and debit interchange. A cap on revolving interest was discussed in 2023.³² Public debt and the budget now sit at the center of the market’s worry, and a government under strain has a large, popular target with high rates. Nu’s filing lists ad hoc caps as a risk.² I can’t put odds on this. It’s in the stress table at 3 and 6 points of margin, worth −11% and −23%.

Rates are smaller than people think. Liquid assets of about $42B earn close to the overnight rate, and deposits cost 88% of it, so a move in Selic roughly nets out on float E (estimate).³⁵ Bonds are a different matter: 1 basis point on the Brazilian curve moves fair value by $0.7M, so the rise in yields last week will show up.³

Governance is the last. One man controls the votes, chairs the board, runs the company and now runs its technology for a while, and sits on the board of an AI lab. The company is in the Cayman Islands and gives no individual pay. I think he’s been a good steward. That’s a judgment about a person, and it can’t be tested until something goes wrong.

The next dates are simple. Copom meets on 4 November. Q3 results are expected on 12 November. The first Investor Day is on 8 December in New York, which is the natural place to give a capital return policy and medium-term targets, neither of which the company has today in any filing I read.⁵²