Nu Holdings Ltd.
NU · NYSE Arca · Brazil
Accepts deposits in Brazilian reais, Mexican pesos, and Colombian pesos and lends that money out using machine learning models trained on 127 million customers.
NuBank accepts deposits in Brazilian reais, Mexican pesos, and Colombian pesos under three separate central bank licences, and converts those deposits into credit cards and personal loans priced by machine learning models trained on transaction data from 127 million customers. Every new customer who transacts adds data that sharpens the models' default predictions, which lets NuBank offer thinner credit spreads and pull in still more customers — but the speed of that loop is capped by how fast NuBank can accumulate retained earnings to satisfy Brazil's Banco Central capital adequacy rules, because every new real lent must be matched by capital on hand. Competitors cannot shortcut their way in by simply funding a licence application, because the models depend on years of transaction history from customers already living and spending in those specific markets, and that history cannot be bought or compressed. The whole structure leans heavily on Brazil — it provides the majority of the customer data and sits under the dominant regulator — so a Brazilian recession, a sharp fall in the real, or a restrictive move by the Banco Central would hit the credit book and degrade the models at exactly the same moment.
How does this company make money?
NuBank earns money in three main ways. First, it collects net interest income — the difference between the low rate it pays depositors and the higher rate it charges borrowers on credit cards and personal loans. Second, every time a customer uses a NuBank credit card, Mastercard processes the transaction and NuBank receives an interchange fee from the merchant. Third, customers who upgrade to premium account features or buy insurance products through the mobile app pay subscription or product fees.
What makes this company hard to replace?
Many Brazilian customers have their payroll deposited directly into NuBank accounts, and switching that arrangement with an employer typically takes 30 to 60 days. NuConta also acts as the hub for automatic bill payments and standing transfers, so moving banks means manually reconnecting every recurring payment. Customers who have used NuBank's credit card for years have also built up rewards points that would be abandoned if they left.
What limits this company?
Brazil's Banco Central requires NuBank to hold a minimum amount of its own capital for every real it lends out. Because Brazil is by far the largest market and the source of most of the 127 million customers, this Brazilian capital requirement effectively caps how fast the company can grow everywhere. Expanding into Mexico or Colombia does not help, because each country adds its own separate capital requirement on top of the Brazilian one rather than sharing the same buffer.
What does this company depend on?
NuBank cannot operate without five things: the Brazilian Central Bank banking licence and Mexico's CNBV authorisation, which are the legal foundations for accepting deposits and making loans; the iOS and Android app stores, which are the main channels through which new customers sign up; the Mastercard payment network, which processes every credit card transaction and generates interchange revenue; AWS cloud infrastructure, which handles transaction processing and runs the data analytics; and customer deposits held in Brazilian reais, Mexican pesos, and Colombian pesos, which fund the loans.
Who depends on this company?
Brazilian merchants rely on the interchange fees generated every time a NuBank credit card is swiped — if those transactions stopped, that revenue would disappear. Residential mortgage borrowers in Brazil who use NuBank personal loans to cover down payments would face reduced access to credit. Small businesses in São Paulo and Mexico City that use NuBank business accounts for daily payment processing would have their transactions disrupted.
How does this company scale?
The machine learning models get cheaper and more accurate as more customers transact — each new data point tightens default predictions and reduces losses, so the cost of serving an additional customer falls over time. What does not get cheaper is geographic growth: entering each new Latin American country requires applying for a separate banking licence, hiring local regulatory compliance teams, and building local deposit relationships from scratch. That part cannot be centralised or automated.
What external forces can significantly affect this company?
Brazil's Banco Central is rolling out a digital currency called DREX, which could reduce demand for private digital banking services like NuConta. When the U.S. Federal Reserve raises interest rates, money tends to flow out of emerging markets like Brazil and Mexico, making it harder and more expensive for NuBank to raise capital. Swings in the value of the Brazilian real and Mexican peso affect how much local deposits are worth in funding terms and squeeze the purchasing power of the customers who are supposed to be repaying loans.
Where is this company structurally vulnerable?
If Brazil's Banco Central revoked NuBank's Brazilian banking licence or imposed hard limits on how much the company could lend, the Brazilian transaction stream — which provides the bulk of training data for the credit models — would be cut off. The models would become less accurate at exactly the moment the loan book is most at risk. The Mexican and Colombian models would also weaken, because they are calibrated against a data mix that would no longer exist.