By Brazil Stock Guide – Nubank (NYSE: NU) elevated its artificial intelligence strategy to the center of its long-term roadmap during its third-quarter earnings call, placing AI at the heart of product development, credit underwriting and customer engagement.
The call — where executives repeatedly emphasized the transformational role of AI — accompanied results that showed $783 million in net income and a 31% ROE, reinforcing the platform’s scale with more than 127 million customers. “Every single one of our metrics continues to grow,” CEO David Vélez told analysts, underscoring the foundation on which the AI strategy is being built.
Nubank’s AI engine is anchored in NewFormer, a proprietary system inspired by Large Language Models, or LLMs — the large-scale neural networks capable of interpreting patterns across vast datasets. The bank trained NewFormer on roughly 600 billion tokens and equipped it with 230 million parameters, an unprecedented scale for a Latin American financial institution. “We are building large, generalizable models based on the same principles that power world-class LLMs,” Vélez said on the call.
This leap is possible because Nubank was born cloud-native, free from the legacy systems that slow data ingestion and model deployment at incumbent banks. According to Vélez, this gives the fintech the ability “to train and deploy foundation models at scale,” a structural advantage as its regional footprint expands.
Early results are already visible. Nubank’s loan book grew 42% year over year to $30.4 billion, while its cost of risk fell 7%, a shift that CFO Guilherme Lago directly linked to improvements coming from predictive AI. “We have seen greater precision in some of the credit modeling techniques, including predictive AI,” he told analysts.
A key example discussed on the call was the multi-quarter rollout of higher card limits. AI-driven models allowed Nubank to expand limits primarily for lower-risk customers, boosting engagement without affecting delinquency. “We have introduced AI to sharpen how we increase credit limits for existing customers, and performance has been largely in line with expectations,” Lago said.
The strategy extends to monetization. Nubank surpassed $13 in ARPAC — Average Revenue per Active Customer — a critical measure of platform depth. In Mexico, ARPAC reached $12.5, nearly matching Brazil. Vélez called Mexico’s trajectory “as strong, if not stronger, than Brazil,” pointing to rapid credit adoption and an accelerating maturity curve.
Scale amplifies these gains. With over 60% of Brazil’s adult population and activity above 83%, Nubank operates one of the largest financial datasets in the hemisphere. Executives highlighted that NewFormer now learns from “trillions of tokens” representing behavior across Brazil, Mexico and Colombia.
Margins reflect both pressure and resilience. The traditional Net Interest Margin — or NIM — dipped as Nubank shifted toward safer credit and faced higher funding costs in Brazil. But the risk-adjusted NIM rose from 9.2% to 9.9%, evidence of stronger portfolio quality. “Even though asset yields came down, the much lower cost of risk led to an expansion of our risk-adjusted margins,” Lago explained during the call.
AI also strengthened recoveries. Lago noted that reactivating previously defaulted borrowers who later cured their debts improved collection levels beyond expectations. “Recovery levels were materially improved,” he said, attributing the gains partly to AI-enhanced strategies.
The call also highlighted Mexico’s rapid operational evolution. With 13 million customers, the cost-to-serve has fallen below $1, a milestone it took significantly longer to reach in Brazil. Lago stressed that Mexico has already surpassed Brazil’s early-stage metrics, driven by digital onboarding, credit expansion and AI-supported underwriting.
Beyond credit, Nubank described plans to transform its customer interface through AI. Vélez told analysts the company is building an “AI-native interface to banking,” one that anticipates needs, executes tasks automatically and uses conversational flows to simplify financial decisions.
Internally, AI accelerates engineering and data governance. Generative models already support development cycles, documentation and quality control, allowing the company to scale operations more efficiently across countries.
Fraud detection also benefits from higher granularity and earlier signal identification. Nubank pairs these advances with a governance structure to ensure model fairness, regulatory compliance and privacy.
Executives said AI will also support international expansion. The company’s application for a U.S. national bank charter reflects its belief that automation enables standardized, efficient operations across borders without the cost burden of physical infrastructure.
Throughout the earnings call, Nubank emphasized how AI strengthens its flywheel. Better underwriting improves credit quality, supporting higher risk-adjusted margins and capital generation. Personalization increases engagement, lifting ARPAC. Automation reduces operating costs and improves efficiency as the platform grows.
Despite the early gains, Vélez stressed that the strategy is still in its infancy. The NewFormer model has not yet been fully deployed in initial credit approvals, personal loans or across all geographies. “We are still just scratching the surface. It’s still day one for us,” he told analysts, positioning AI not as a feature but as the core of Nubank’s long-term competitive edge.
Nubank now presents itself as a global contender in AI-driven financial services, aiming to set a new standard for efficiency, monetization and risk intelligence across Latin America and beyond.
Leave a Reply