Nu Holdings Crosses $1 Billion Quarterly Profit Threshold as AI Platform Drives Record Risk-Adjusted Margins
Q2 2026 earnings call, August 13, 2026
Nu Holdings delivered its first-ever $1 billion net income quarter, a milestone founder and CEO David Velez framed as validation of a 13-year thesis that a branchless, technology-first bank could serve hundreds of millions of customers more cheaply than legacy incumbents. Net income hit $1.1 billion, up 17% quarter-over-quarter and 49% year-over-year, with return on equity at a record 33%. The company's customer base reached 139 million across Brazil, Mexico and Colombia, with gross revenue of $5.9 billion and an efficiency ratio of 19.5%.
Risk-Adjusted Margin Surprise Draws Analyst Scrutiny
The standout metric was risk-adjusted net interest margin, which expanded to a record 12.4% from 9.5% in the prior quarter, well above the 10.5% to 10.8% range management had guided to on the Q1 call. Bank of America's Mario Pierry pressed CFO Rob Livingston directly on the gap, asking whether better-than-expected credit models or faster growth explained the beat. Livingston attributed roughly one-third of the overperformance to Desenrola, a government debt renegotiation program not fully visible at the time of Q1 guidance, with the remaining two-thirds split between stronger-than-expected credit performance and higher-than-planned interest-earning balances from loan growth. Credit income contributed 178 basis points to the sequential improvement, versus 152 basis points in Q1, while lower cost of credit added another 115 basis points. Livingston was careful to temper expectations going forward, telling Morgan Stanley's Jorge Kuri that 12% should not be read as a floor: "I didn't say that it was a floor. I said we'd be in that ballpark. I don't want to overcommit there."
Pierry also flagged an inconsistency between two slides in the deck — one showing a steady or improving credit card delinquency trend by income bracket, the other showing overall 90-plus delinquencies rising over two years. Livingston clarified the divergence is a mix-shift issue rather than deterioration within any single product: the credit-card-only slide is smoothed and disaggregated by income segment, while the portfolio-wide seasonality slide reflects a shift toward asset classes and cohorts intentionally taken on for higher risk-adjusted returns, not credit quality erosion in unsecured lending specifically.
nuFormer: The AI Platform Now Touching Nearly Every Decision
Velez used a significant portion of his prepared remarks to detail nuFormer, a foundation model for financial behavior the company has built over the past year, describing it as central to Nu's competitive positioning. The architecture recently moved to hybrid linear attention — the same approach underpinning frontier models like Kimi K2 and Qwen 3.5 — trained with the Muon optimizer used in today's most efficient large language models. Velez said the new generation quadrupled context length, training speed and inference speed while cutting production costs, and that the model's efficiency gains are compounding: "Today, we can achieve the same predictive performance with 20 million fine-tuning data rows that previously required over 400 million, cutting development cycles from weeks to days."
The platform was first deployed in Nu's flagship Brazilian credit portfolio, replicated in Mexico through 2025, and extended this year into unsecured lending, next-generation core credit models, SME credit cards and Colombian underwriting. Beyond credit, AI agents now handle more than 60% of customer support conversations in Brazil at or above human satisfaction parity, and more than 100 marketing campaigns have been run using nuFormer-driven targeting. Livingston, responding to a question on whether AI-assisted underwriting is riskier in unfamiliar markets, disclosed that Nu is capping U.S. market entry costs at no more than 100 basis points of efficiency ratio impact, because rebuilding credit models with sufficient data depth for a new geography will take between 12 and 30 months even with the platform architecture transferring quickly.
Mexico Banking License Clears Path to Full Digital Bank Status
Mexican regulators approved Nu's banking license this month, a milestone Velez called the completion of the company's "transformation from a credit-first fintech into a full-scale digital bank." The license unlocks payroll direct deposit, higher deposit insurance, and a broader credit product suite. Nu now serves over 16 million customers in Mexico, having reached 16.5% of the country's adult population — roughly the penetration Nu had in Brazil in 2020, but monetizing faster: ARPAC in Mexico stands at $12.30 versus $5.60 in Brazil at the equivalent stage, driven by higher income per capita and better unit economics. Velez noted Nu reached breakeven in Mexico in six years versus eight in Brazil, and reiterated a base case that Mexico could ultimately reach 60% to 70% of Brazil's business size despite a smaller population, aided by a new central bank mandate requiring standardized payment interfaces across the financial system by year-end — a regulatory push Velez compared directly to the network effects Pix created in Brazil.
Chroma Targets the "Supercore" Gap Between Mass Market and Affluent
Nu launched Chroma in July, a subscription tier aimed at a segment it calls Supercore — customers who sit between the mass market and the roughly 1 million Ultravioleta high-income customers (whose purchase volumes and assets under custody grew 41% and 37% year-over-year, respectively). Velez said Chroma addresses a segment with "an even larger profit pool than high income" where Nu already has deep penetration but low share of wallet, citing free ChatGPT subscriptions, accelerated savings products and lifestyle benefits as retention tools. Asked about market share ambitions, Velez said the opportunity is less about acquisition than monetization: "we already have 3 out of 5 Brazilians in this bracket as customers of Nu... we just haven't given them the product set and the bundle that they should need."
Credit Quality: Seasonal, Not Structural, Despite Macro Caution
Early delinquencies (15-90 days) improved 16 basis points to 4.8%, while 90-plus delinquencies rose 35 basis points to 6.9%, which management characterized as normal seasonal migration rather than deterioration. Coverage over 90-plus NPLs stood at 244%. Velez, addressing concerns about 2027 growth into a cautious macro backdrop, reiterated that Nu's underwriting assumes conditions will be worse than history by default and that the company holds "a pretty significant cushion" against that assumption. He also pointed to Nu's status as primary bank for over 60% of its mass-market Brazilian customers as a structural credit advantage, noting delinquency among primary-relationship customers runs roughly half the portfolio average — effectively making Nu senior in those customers' credit stack.
Private Payroll Lending: Deliberately Slow-Walked
Velez addressed the private payroll loan product directly, explaining Nu has intentionally throttled its rollout because early adopters skewed toward high-risk borrowers rather than the low-risk refinancers the product was designed for: "counter to most people's intuitions, the first people that started taking these loans were very high risk... it just didn't make a lot of sense for us to be opening the door for that." He said behavior is shifting as better-quality customers begin seeing refinancing value, and that Nu is "getting close to a system that makes a lot of sense" and expects to be among the market leaders within 18 to 24 months, framing the product as a source of portfolio diversification and reduced cyclicality longer term.
CFO Transition and OpenAI Board Seat
The call marked Rob Livingston's first as CFO, succeeding Guilherme Lago, who spent five years in the role and will remain as special adviser. Separately, Velez confirmed his recent appointment to OpenAI's board, framing it as an opportunity to gain insight into AI development while contributing to what he called ensuring "great AI companies build something great for humanity." He declined to elaborate further, noting it was not OpenAI's earnings call.