Inter & Co recorded 65.6% CDI (Certificado de Depósito Interbancário) funding costs in Q4 2024, the lowest rate among Brazilian financial institutions serving over 20 million clients. The metric reflects borrowing costs as a percentage of Brazil's benchmark interbank rate.
Nu Holdings kept risk-adjusted net interest margins flat quarter-over-quarter after removing the impact of FGTS (Fundo de Garantia do Tempo de Serviço) program changes, according to CFO Guilherme Lago. The government program adjustment reduced margins industry-wide, but Nu's AI systems maintained underlying credit performance.
Both banks deployed production AI credit models during 2025. Nu's nuFormer system processes loan applications using machine learning trained on over 100 million customer data points. Inter's newer client cohorts transact faster and more frequently than earlier users, indicating improved targeting from algorithmic underwriting.
Traditional Brazilian banks operate with funding costs 15-20 percentage points higher than Inter's rate. The gap stems from digital banks' data-driven customer selection and automated risk pricing, which reduce default rates and allow lower interest offers to depositors.
Inter became Brazil's fastest-growing financial institution among those exceeding 20 million clients. Growth correlates with the bank's ability to price loans more precisely than competitors lacking real-time risk models.
The funding cost advantage creates a compounding effect: lower deposit rates improve margins, which funds customer acquisition, which generates more training data for AI models. Digital banks with this flywheel can underprice traditional lenders while maintaining profitability.
Brazil's Central Bank requires all banks to maintain minimum capital ratios regardless of loan performance technology. AI models help digital banks optimize within these constraints by identifying lower-risk borrowers who would receive identical treatment under traditional underwriting.
The 2026-2027 period will test whether AI credit advantages persist through economic cycles. Brazilian interest rates remain volatile, and machine learning models trained during growth periods may underperform in downturns. Non-performing loan rates across both digital and traditional banks will determine if current efficiency gaps reflect permanent structural advantages or temporary market conditions.