Nubank’s AI Credit Model Drives 23% Jump in Net Interest Income

Nubank's AI Credit Model Drives 23% Jump in Net Interest Income

Brazilian digital bank Nubank reported that its proprietary artificial intelligence credit-scoring model contributed to a 23% year-over-year increase in net interest income, reaching $2.8 billion in the fourth quarter, according to results released Tuesday evening.

The São Paulo-based fintech, which trades on the New York Stock Exchange under the ticker NU, has been deploying machine learning algorithms to refine its underwriting decisions across its 110 million customer base in Latin America. The AI system analyzes thousands of data points beyond traditional credit bureau scores to assess borrower risk, allowing the bank to expand lending while maintaining credit quality.

For CFOs watching the intersection of AI deployment and actual financial results, Nubank's numbers offer a concrete case study. The company's net interest margin—the spread between what it earns on loans and pays on deposits—expanded even as it grew its loan portfolio by 34% to $22.1 billion. That's the kind of outcome finance leaders want to see: AI tools that demonstrably improve unit economics, not just operational efficiency.

The AI model's impact shows up most clearly in Nubank's credit card business, which represents roughly 60% of its loan book. By more accurately identifying creditworthy customers who traditional banks might reject, the system has allowed Nubank to approve more applications while keeping its non-performing loan ratio at 4.8%, below the Brazilian banking sector average of 5.2%.

Here's the thing everyone's missing: this isn't about automating existing processes faster. Nubank's AI is making fundamentally different credit decisions than a human underwriter would make—and those decisions are generating measurably better returns. The company's return on equity hit 28% in the quarter, up from 24% a year earlier, suggesting the model isn't just approving more loans but approving better loans.

(I should note: Nubank hasn't disclosed the specific architecture of its AI model or exactly which features it weighs most heavily. That's proprietary sauce. But the financial results suggest it's working as advertised, which is more than you can say for most "AI-powered" fintech claims.)

The broader implication for finance leaders: AI's impact on lending businesses may be arriving faster than on other financial functions. Credit decisioning is a perfect use case—massive datasets, clear success metrics, and immediate feedback loops. If you're a CFO at a bank or lender, your competitors are probably already testing similar models. If you're not, you're likely falling behind on both approval rates and credit quality simultaneously.

The question now is whether Nubank can maintain these economics as it scales. The company added 5.2 million customers in the quarter, and its AI model will need to perform equally well on this newer, less-seasoned cohort. The real test of any credit model comes 12-18 months after origination, when you see how those loans actually perform through a full economic cycle.

For now, though, Nubank has delivered what finance leaders always want to see: a specific AI application with specific financial results. Not a pilot program, not a proof of concept—actual revenue growth tied directly to the technology.