By Brazil Stock Guide – Nubank wants to convince the market that artificial intelligence will become the next major growth layer for the bank. But for the first time, that thesis is being tested in the one place where no technology narrative can survive without numbers: the cost of credit.
In a videocast released Wednesday, Eric Young, Nubank’s Chief Technology Officer, and Rohan Ramanath, who leads the company’s Core AI team, presented AI as a strategic priority for the institution. The ambition goes beyond using models to automate existing processes. The idea is to redesign products, decisions and workflows around an AI-first logic.
Ramanath framed the distinction simply. “Adoption is like doing something faster,” he said. Transformation is something else: “if AI can do most of this work, then what is the work?”
That distinction is central to understanding Nubank’s thesis. The bank is not merely saying that it can improve customer service, write code faster or reduce costs with AI. It is saying that proprietary models, data at scale and low operational friction can enable a new way to make credit decisions, price risk, fight fraud, renegotiate debt and personalize the financial experience inside the app.
At the center of that thesis is nuFormer, Nubank’s proprietary foundational model trained on the company’s own data. According to Ramanath, the model allows for more granular decisions than previous approaches and has already been integrated into the system responsible for credit-card limit increases in Brazil’s mass-market segment — the bank’s largest individual segment in its main market.
“With precision come inclusion and profitability at the same time,” Ramanath said. The phrase captures the promise: better models would not only reduce losses, but also allow Nubank to lend more to customers who might previously have fallen outside approval curves — and to do so profitably.
It is a powerful thesis. It is also one that now has to prove it works at scale.
In the first quarter of 2026, Nubank’s provisions for credit losses reached US$1.79 billion, up 33% from the previous quarter. Its 15-to-90-day NPL ratio rose to 5.0%, while risk-adjusted NIM fell to 9.5%, from 10.5% in the fourth quarter of 2025. The company attributed the move to three factors: seasonality, portfolio growth and product mix.
Nubank’s defense is that there has been no structural deterioration. The 90-plus-day NPL ratio declined to 6.5%, below the 7.0% peak recorded in the third quarter of 2024. The credit portfolio, meanwhile, continued to expand, reaching US$37.2 billion, up 40% year over year. The bank also reported revenue above US$5 billion, net income of US$871 million and an annualized ROE of 29%.
In other words, Nubank remains highly profitable, is still growing quickly and continues to operate with an efficiency level that traditional banks would struggle to replicate. The issue is different. The market has started to ask whether the acceleration in credit is coming with a higher cost of risk than expected.
Nubank says nuFormer provides better signals. But Ramanath was careful to separate signal from decision. “nuFormer gives us a better signal; it does not tell us what to do,” he said. The final decision, according to him, remains anchored in risk-adjusted economics, credit policy, governance, fairness testing and independent review by risk teams.
That distinction matters because it anticipates the most obvious criticism: that AI could simply be giving a more sophisticated language to a more aggressive credit expansion. Nubank rejects that interpretation. For the fintech, better models allow risk to be calibrated more precisely — not ignored.
Still, the nature of the market’s scrutiny has changed. When Nubank talks about AI, investors are likely to focus less on the elegance of the model and more on three balance-sheet lines: provisions, early-stage delinquencies and risk-adjusted NIM.
Beyond credit, Nubank says it is using AI in fraud prevention, collections, deposit-pricing, software engineering, product development and internal productivity. In fraud, the promise is to increase security without creating excessive friction for the customer. In collections, the company sees room for AI-assisted conversations that help customers renegotiate debt in a more personalized and less intimidating way.
The most ambitious front is the so-called AI Private Banker. Ramanath described it as a set of capabilities inside the app designed to help customers organize their finances, choose the right credit product and better manage their debt. The provocation is clear: private banking has historically been a luxury service; AI could make personalized financial advice accessible to low- and middle-income customers.
If it works, the impact could be meaningful. A smarter app could increase engagement, improve cross-selling, lift revenue per customer and reduce the cost to serve. That is the bridge between the technology story and the equity thesis.
But that bridge only holds if AI improves the quality of decisions. And for a bank, the most important decision is still credit.
Nubank built one of the most impressive stories in global finance by simplifying the banking experience. It is now trying to prove something harder: that artificial intelligence can personalize that experience, increase monetization and expand credit without weakening risk discipline.
The promise is that AI can better separate good risk from bad risk. The first quarter of 2026 showed that this promise will be tested in real time.
It is too early to say the strategy is going wrong. Nubank’s profit, growth and returns remain too strong for that conclusion. But it is fair to say the bar has moved higher.
From now on, the market will not judge Nubank’s AI by what it promises to do inside the app. It will judge it by what it delivers in the balance sheet.
Leave a Reply