Inter & Co welcomed 7 million new clients in 2025, its best annual performance, while Nu Holdings deployed its nuFormer AI credit model to production and launched more than 100 products and features. The customer acquisition surge marks a shift in emerging market lending, where digital banks using machine learning underwriting now outpace traditional competitors.
Nu's newer customer cohorts transact faster than older groups, a pattern Inter also reported. The acceleration reflects AI models identifying creditworthy borrowers traditional banks miss—or mispricing risk that surfaces later.
Brazil's November 2025 FGTS regulation changes triggered the first stress test. Nu's portfolio growth rate dropped from an expected 13-14% to roughly 10% as the AI model struggled to adapt to new lending constraints. The company typically sees 15-90 day NPLs tick up in Q1, indicating sensitivity to external shocks.
Inter expects its cost of risk in 2026 to reach 5.5-6%, with private payroll loan delinquency converging above 10%. The forecast suggests AI-driven growth creates concentrated exposure: models trained on stable periods may misread creditworthiness when macro conditions or regulations shift.
Traditional banks spread underwriting across human judgment and algorithmic inputs, diluting single-model failures. AI-native lenders face binary risk: their models either capture growth opportunities competitors miss or accumulate portfolios that deteriorate together under stress.
Nu's nuFormer deployment shows the upside—rapid feature rollout and customer scaling impossible with manual credit review. Inter's 7 million new clients demonstrate market demand for instant approvals. Both banks rely on algorithms to segment risk across millions of micro-decisions daily.
The delinquency projections reveal the downside. When external variables change—regulation, employment rates, government lending programs—AI models trained on historical patterns may continue approving profiles that no longer perform. Human underwriters adjust heuristics intuitively; algorithms require retraining cycles that lag market shifts.
Emerging market fintech growth now depends on model resilience, not just model accuracy. Banks that built portfolios in 2024-2025's favorable conditions will test whether their AI can navigate 2026's headwinds—or if speed-to-market came at the cost of durability.