Nu Holdings' nuFormer AI model went live in 2025, enabling the digital bank to launch over 100 products and features in a single year. Inter & Co welcomed 7 million new clients in 2025, its strongest annual performance, with newer customer cohorts transacting faster than older ones onboarded through traditional methods.
The speed advantage disappeared when external conditions shifted. Brazil's November 2025 FGTS regulation changes caused Nu's portfolio growth to slow from an expected 13-14% to roughly 10% within weeks. Inter projects its cost of risk will hit 5.5-6% in 2026, with private payroll loan delinquency climbing above 10%.
AI-native underwriting creates concentrated model risk because millions of customers are approved using identical algorithms trained on similar data. Traditional banks spread risk across multiple human underwriters making independent decisions with different risk tolerances and local knowledge.
Nu typically sees seasonal increases in 15-90 day non-performing loans during Q1, indicating vulnerability to predictable external shocks. The FGTS regulatory change represented an unpredictable shock that AI models trained on historical data couldn't anticipate.
Digital banks achieve faster customer acquisition because AI models process applications in seconds rather than days, require minimal documentation, and operate 24/7 without human review. Inter's 7 million new clients in one year would require thousands of loan officers using traditional methods.
The tradeoff manifests during downturns. When economic conditions or regulations change, AI models continue approving customers using pre-crisis patterns until delinquency data accumulates and triggers model retraining. Traditional underwriters adjust risk appetites immediately based on news and local economic signals.
Emerging markets amplify this dynamic because regulatory frameworks evolve rapidly and economic volatility exceeds developed markets. Brazil's consumer credit regulations changed three times in 2025, each time creating lag periods where AI models operated on outdated assumptions.
The financial impact is measurable: Nu's 3-4 percentage point growth slowdown on a multi-billion dollar portfolio represents hundreds of millions in reduced revenue. Inter's projected 10%+ delinquency rate on payroll loans will require significant loan loss provisions.
Banks now face a fundamental choice: accept slower growth with diversified underwriting risk, or chase AI-driven acquisition knowing regulatory changes and economic shocks will cause sharp delinquency spikes.