Nubank plans to bring its nuFormer AI underwriting model to the US

For the chief credit officer at a regional bank

Nubank plans to bring its proprietary AI underwriting model to its U.S. operations, founder and CEO David Velez said on the company's third-quarter earnings call on Nov. 12.1

Nubank expanded into the United States in September, according to finainews.com; the source does not say where in the U.S. it operates or how many customers it has there.1 Velez said the model is a well-tuned system that can iterate fast, which lets Nubank develop new models and test and learn faster.1

According to finainews.com, Nubank began developing nuFormer, an AI foundation model that analyzes financial transaction data to support credit underwriting, fraud detection and prevention, and behavior modeling; the article says the model launched in May, and Velez said it is on its third version.1

In Brazil, Nubank has more than a decade of customer data to train on, and about 30% of Brazil's population used it as their primary institution by the end of 2025, according to NPS Prism by Bain & Co. data published in July.1

Velez said that in the U.S. a positive credit bureau and a FICO score are readily available for anyone new to the market, and that Brazil lacked this.1

Velez said bringing the model to the U.S. will take time and money: "You need losses so that your model learns from the losses." He said he thinks of these losses as capital expenditure or R&D investment.1

Why it matters

Nubank says it will fund model training in the U.S. with credit losses, starting from bureau data that every U.S. lender can use. Competitors should watch whether the model learns enough from its own lending to build an underwriting edge, since Nubank has the earnings to absorb the losses: it reported record quarterly net income of $1.1 billion.

From the brief

Our take. Nubank earned $1.1 billion last quarter and says it will treat US credit losses as R&D. Your bank books the same losses as losses. Price your credit book knowing a rival can afford to pay for its lessons.

Action. Model risk: Test any transaction-data credit model for fair-lending disparate impact and explainable adverse-action reasons before using it in underwriting.

From our records

Sources

  1. ReportNubank to expand AI underwriting into US market Finainews