This is investment research, not personal financial advice.
Nu Holdings (NYSE:NU) rose 9.3% on Friday, from $13.93 to $15.23, after Nubank reported its first quarter with more than $1 billion of net income. The market added about $6.3 billion to the company's equity value, taking market capitalisation to $73,568 million. That was almost a one-for-one response to a result that beat the Visible Alpha profit estimate by 9.7%: $1.061 billion against $967.2 million (Reuters 2026; Yahoo Finance 2026).
The quarter was better than its headline. Risk-adjusted net interest margin, which deducts credit losses from lending spread, rose from 9.5% in Q1 to 12.4%. Early arrears improved. Deposit funding became cheaper relative to local rates. Yet the same report showed 90-plus-day non-performing loans rising to 6.9%, while common-equity tier-one capital fell to 11.9%. The share price now equals roughly 5.6 times June book value and 17.6 times annualised Q2 earnings.
Friday's reaction looks proportionate to the earnings surprise. The valuation left behind is harder to justify than the move itself. At $15.23, Nu needs something close to its present 33% return on equity to persist for years, even as risk expands and old loan cohorts pass into late delinquency. That is the tension: a bank with software-like operating economics still answers to bank arithmetic.
The tape paid almost exactly for the estimate beat
Nu's release separated three credit signals that the market had blurred after Q1. Its 15-to-90-day NPL ratio fell 16 basis points to 4.8%, mostly because the second quarter normally has fewer early delinquencies. The company deliberately moved into riskier, higher-return credit, partly offsetting that seasonal benefit. Meanwhile, the 90-plus-day ratio rose 35 basis points to 6.9% as Q1 arrears aged. Cost of credit declined 9% sequentially to $1.7 billion, and risk-adjusted NIM recovered by 290 basis points to 12.4% (Nu 2026a).
Those figures explain the relief. They do not prove that credit has become safer. Early arrears measure entry into trouble; late arrears measure accounts that have stayed there. A seasonal improvement in the first can coexist with deterioration in the second for another quarter. Management's claim is narrower: the portfolio produced enough additional spread to more than cover the expected loss, lifting credit's contribution to gross profit to 41%.
The income statement confirms substantial earnings power. Q2 interest and financial income was $4.761 billion and fee income was $753 million, giving IFRS total revenue of $5.513 billion. After $1.557 billion of funding and financial expenses, $1.482 billion of expected credit loss and $128 million of transaction costs, gross profit was $2.347 billion. Operating costs absorbed $1.106 billion. The remaining $1.061 billion was net income (Nu 2026b).
That $5.513 billion IFRS revenue figure is lower than the approximately $5.9 billion of "gross revenue" in management's release and the Reuters report. They are different measures, not a rounding error. The history table and valuation here use the face of the reviewed IFRS statements. Management's gross-revenue measure is useful for the operating narrative but should not be spliced into the filed series.
Revenue is no longer the scarce ingredient
Nubank makes money from revolving credit-card balances, personal loans, secured lending, card interchange, account and investment services, and the spread between deposits and earning assets. Its economic engine begins with a cheap digital account, then relies on activity and product adoption to turn a customer into a primary financial relationship. In Q2 it added about four million customers, reaching 139 million across Brazil, Mexico and Colombia. Monthly activity was 83.5%; Brazil passed 86%. Average monthly revenue per active customer, or ARPAC, reached roughly $17 (Nu 2026a).
Scale matters in two ways. First, the same technology and service operation can support more accounts without adding a matching branch estate. Second, transaction history improves product selection and underwriting. Nu says its credit models collect and learn from account-level behaviour, then price credit for net present value rather than the lowest possible NPL. That distinction matters. A lender can report higher arrears than a prime incumbent and still earn more after losses if pricing, funding and operating cost compensate.
Deposits are the other half of the model. They reached $45.3 billion at June, compared with a $39.4 billion credit portfolio. The consolidated cost of those deposits was 88% of local interbank rates, three percentage points less than a year earlier. Mexico remained unusually liquid, with loans equal to only 35% of deposits, so Nu allowed some deposits to leave rather than pay for funding it could not yet deploy. Brazil provided the mature template; Mexico and Colombia are still paying for licences, product launches and customer acquisition.
This creates a useful split. Brazil must generate enough surplus to fund market expansion without weakening group capital. Mexico needs to turn deposits and 15.8 million customers into credit without importing a loss curve faster than it imports underwriting knowledge. Colombia, at more than five million customers, remains smaller. A future US operation would add another period of expense before revenue. The business can keep adding accounts; the scarce inputs are seasoned credit data and capital.
Thirty-three percent ROE comes from three spreads
Nu's reported 33% annualised Q2 ROE is not a single margin. It combines a funding spread, a credit spread after losses and an operating-cost advantage.
The funding spread improved because deposits grew and became cheaper relative to local interbank benchmarks. The credit spread widened because unsecured and other higher-yielding assets grew faster, lifting NIM to 22.9%. After credit cost, risk-adjusted NIM was 12.4%. Finally, a 19.5% efficiency ratio left far more gross profit for equity than a branch-heavy model would. That ratio worsened from 17.6% in Q1 because property, marketing and international spending shifted into Q2, but it remained low for a bank (Nu 2026a).
Brazil's 14.0% Selic target at the session date helps both sides of that equation. High rates can support asset yields, particularly when deposits cost less than the benchmark, but they also pressure borrowers and increase the return demanded by equity capital (BCB 2026). A high reported ROE in that setting is valuable only after separating nominal rate benefit from underwriting skill.
The five-year record shows the transformation. Revenue and profit are in US dollars as filed. Annual ROE below is author-computed as IFRS net income divided by average parent equity, rather than management's adjusted measure. The 2022 CET1 ratio is also author-computed from filed common equity tier-one capital and risk-weighted assets. Later CET1 and NPL figures are reported. Q2 2026 is a single quarter and is not comparable with the annual rows.
| Period | IFRS revenue ($m) | IFRS net income ($m) | ROE | CET1 | 90+ NPL |
|---|---|---|---|---|---|
| 2022 | 4,792.2 | (364.6) | -7.8% computed | 15.1% computed | 5.2% |
| 2023 | 8,029.0 | 1,030.5 | 18.2% computed | 11.4% | 6.1% |
| 2024 | 11,517.1 | 1,972.1 | 28.1% computed | 14.7% | 7.0% |
| 2025 | 15,774.7 | 2,868.9 | 30.3% computed | 13.0% | 6.6% |
| Q2 2026 | 5,513.2 | 1,061.1 | 33.0% reported, annualised | 11.9% | 6.9% |
Sources: filed annual and interim statements and company disclosures (Nu 2022; Nu 2023; Nu 2024; Nu 2025; Nu 2026a; Nu 2026b). Parent equity used in the computed annual ROE was $4.441 billion at December 2021, $4.891 billion in 2022, $6.406 billion in 2023, $7.646 billion in 2024 and $11.291 billion in 2025.
Revenue more than tripled from 2022 to 2025, while a $365 million loss became a $2.869 billion profit. Operating leverage is visible in the accounts: income grew much faster than the equity base. But capital ratios did not improve in a straight line. Nu moved from 15.1% CET1 in 2022 to 11.4% in 2023, rebuilt to 14.7% in 2024, then fell to 13.0% in 2025 and 11.9% by June 2026. Fast growth keeps asking for capital even when earnings are strong.
The arrears number is both worse and better than it looks
Nu's 6.9% late-delinquency ratio looks poor beside an incumbent. Itaú Unibanco reported a 1.9% 90-day NPL ratio in Q2, alongside 24.3% recurring managerial ROE and 12.3% CET1 (Itau 2026). The comparison is directionally useful and mechanically imperfect. Itaú has a much broader corporate, mortgage, payroll and secured portfolio. Nu has a larger mix of credit cards and unsecured consumer loans, including customers that legacy banks underserve. Nu itself warns that its NPL ratio is not directly comparable with market figures because product and customer mix differ.
Risk-adjusted margin is therefore the cleaner scoreboard. If a lender charges enough spread to cover expected losses, funds itself cheaply and still retains customers, a higher NPL ratio can be economically rational. Q2 passed that test: risk-adjusted NIM expanded despite the 6.9% late ratio. The Reuters account also reported that JPMorgan analysts had expected about 11% risk-adjusted NIM even in a constructive reading; 12.4% cleared that level (Reuters 2026).
The counter-evidence sits in the cohort mechanics. Management said most of the early-delinquency improvement was seasonal and most of the late-delinquency increase came from Q1 accounts aging. It also said intentional expansion into higher-risk segments offset part of the early benefit. In other words, the next loss wave is not purely an accident of the calendar. Some of it was selected because the expected return looked attractive.
That makes 12.4% a starting margin, not a permanent cushion. Competition can narrow asset yields. Funding can become more expensive. Recoveries can disappoint. Brazil's macro rate can fall, and borrowers can weaken before pricing fully adjusts. Mexico has less seasoned customer data and a different recovery system. Nu's underwriting advantage is credible because the risk-adjusted result is visible; it is not yet proven through a full credit cycle at this scale.
Capital growth sets the owner-cash limit
Free cash flow is the wrong owner metric for a lender. Deposits and loans are operating raw material, not ordinary working capital, and regulatory capital limits how much accounting profit can leave the business. A more useful bridge starts with net income and subtracts the capital needed to support incremental risk-weighted assets, then treats share-based compensation as an economic owner cost.
Nu generated $1.933 billion of IFRS net income in H1 2026. Risk-weighted assets rose from $31.142 billion at December to $35.710 billion at June, an increase of $4.568 billion. Applying an illustrative 11.5% internal CET1 floor to that increase absorbs $525 million. Share-based compensation was $151 million. The bridge is author-computed from the reviewed statements; 11.5% is an analytical assumption, not management's policy (Nu 2026b).
| H1 2026 owner-earnings bridge | $m |
|---|---|
| IFRS net income | 1,932.5 |
| Less: capital for incremental RWA at 11.5% | (525.4) |
| Less: share-based compensation | (151.3) |
| Pre-distribution owner earnings, computed | 1,255.9 |
| Share repurchases | (500.4) |
| Residual retained after repurchases | 755.5 |
This is not a cash-flow statement and does not claim that $1.256 billion was legally distributable. It is a consistent way to show the tension between profit, growth and dilution. Nu used about 40% of the computed pre-distribution amount on repurchases. The remaining $756 million had to cover additional buffers, acquisitions, market entry and model error.
The balance sheet is liquid, with $13.6 billion of cash and $14.0 billion of securities at fair value through other comprehensive income. Deposits of $45.3 billion exceed the customer credit portfolio. Liquidity is not the immediate constraint. The narrower question is capital: CET1 at 11.9% sits 40 basis points above the monitoring floor used here, and management is opening more credit channels. A lower loss ratio creates capital internally; a credit surprise consumes it twice, through earnings and risk-weighted assets.
The operating moat is widening faster than the capital buffer
Three pieces of the moat have numerical support. Customer engagement rose to 83.5%; Brazil's rate exceeded 86%. ARPAC kept climbing to about $17. Deposit cost fell relative to interbank rates. Together these suggest more primary relationships rather than a dormant-account land grab. The economics also cross borders: Mexico had 15.8 million customers before the full-bank launch, and Colombia exceeded five million (Nu 2026a).
The underwriting moat is less settled. Nu can observe payment flows, account balances and product behaviour at enormous scale. Its 2025 annual report describes a self-learning credit engine designed around long-run net present value, not minimum arrears. The 12.4% risk-adjusted NIM supports that claim. Yet a model is only as durable as its training environment. Rapid movement upmarket, deeper unsecured lending, Mexico and a prospective US operation all change the data distribution. Historical success in Brazilian mass-market cards does not automatically transfer.
Management's capital choices sharpen this test. The company repurchased $500 million of shares in H1 while CET1 fell. Repurchases reduced outstanding equity at a time when risk-weighted assets increased by 14.7%. They also offset a large part of share-based compensation. The choice can be coherent if management believes book value will compound at very high returns and the capital buffer is sufficient. It leaves less margin for a credit model that proves too optimistic.
The strongest anti-thesis to a cautious valuation is simple: traditional bank mean reversion may be the wrong frame. A branchless institution with 139 million customers, cheap deposits and primary-account data can sustain a lower cost base and make better product decisions than incumbents. Itaú's 24.3% peer ROE shows that high Brazilian bank returns are not exclusive to Nu; Nu's 33% adds a meaningful operating premium. If that premium persists while Mexico moves from surplus deposits into profitable loans, book value could compound fast enough to make today's multiple look less extreme.
The strongest disconfirming fact is equally plain. The post-move valuation needs that premium to persist. There is little room for Nu to become merely a good bank.
$15.23 is a demanding residual-income equation
A bank is best valued from book equity and the spread between ROE and cost of equity. The model here begins with June parent equity of $13.250 billion and 4.905 billion diluted shares, or $2.70 per diluted share. It projects five years of residual income, fades ROE toward a terminal level, retains most earnings to fund growth and discounts the return earned above an assumed US-dollar cost of equity. Scenario values are in the NYSE quote currency.
| Case | Five-year and terminal frame | Value per share |
|---|---|---|
| Severe downside | ROE fades to 13-16%; cost of equity 17-18%; terminal growth 2-3% | $2.0-$3.1 |
| Bear | ROE fades to 18-21%; cost of equity 15-16%; terminal growth 4-5% | $4.1-$6.4 |
| Base | ROE fades to 23-26%; cost of equity 13.5-14.5%; terminal growth 5-6% | $8.0-$12.9 |
| Bull | ROE remains 28-31%; cost of equity 12-13%; terminal growth 6-7% | $16.0-$28.3 |
These ranges are outputs, not brackets drawn around the market price. The base range sits below $15.23. The post-result price begins just under the bull range, where terminal ROE stays near 30%, the cost of equity falls toward 12-13% and most earnings remain available to compound book value.
The model is unusually sensitive because Nu starts with a large return spread over book. Holding retention at 91%, terminal growth at 5.5% and initial ROE at 33% produces the following values:
| Terminal ROE | 12.0% cost of equity | 13.5% | 15.0% | 16.5% |
|---|---|---|---|---|
| 18% | $8.21 | $6.31 | $5.03 | $4.11 |
| 22% | $11.76 | $9.01 | $7.17 | $5.85 |
| 26% | $15.88 | $12.15 | $9.65 | $7.87 |
| 30% | $20.61 | $15.76 | $12.50 | $10.18 |
Reverse-solving the same model for $15.23 shows what the market is asking. At a 13% cost of equity and 5.5% terminal growth, it requires terminal ROE of about 28%. At 14.5% cost and 5% growth, the required terminal ROE rises to roughly 33%. A 16% cost and 4% growth requires nearly 38%. The debate is therefore not whether Q2 was strong. It was. The debate is how much of a 33% quarterly ROE survives competition, maturity and credit seasoning.
Two results will separate seasonality from selection
Q3 2026 is the first checkpoint. Early arrears improved in Q2 because of seasonality, while late arrears rose as Q1 accounts migrated. If the 90-plus-day ratio peaks near 6.9% and risk-adjusted NIM remains above 11%, the company will have converted higher-risk credit into owner economics rather than just nominal spread. If late NPLs pass 7.2% for two quarters without a matching margin response, the credit engine is consuming more of the operating advantage.
Q4 and the 2026 annual report add the capital evidence. CET1 needs to remain near 12% while risk-weighted assets and Mexico's loans grow. A ratio below 11.5% alongside RWA growth faster than retained earnings would narrow the room for expansion and distributions. The annual filing will also show whether H1's $500 million repurchase was a one-off use of surplus or part of a recurring capital plan.
Customer economics resolve more slowly. Activity below 82% or ARPAC below $16 for two quarters would suggest that additions are no longer deepening the relationship. Stable or rising figures would support the cheaper-deposit and richer-data loops that distinguish Nu from a conventional consumer lender. Mexico's low 35% loan-to-deposit ratio provides room, but the return arrives only when credit seasons.
Friday's 9.3% rally was a reasonable response to a roughly 10% profit beat and a larger surprise in risk-adjusted margin. The close at $15.23 now prices much more than one clean quarter. It prices a credit model that keeps high-20s or better ROE after late arrears catch up, enough capital to fund three markets, and a cost base that remains structurally below incumbent banks. Q3 will show whether seasonality bought time. The 2026 annual report will show whether Nu used it to widen the return spread without thinning capital again.
Source notes, confidence and missing information
The Q2 accounts were reviewed by KPMG under IAS 34, not audited as a full-year report (Nu 2026b). The annual history comes from fetched SEC filings; computed ROE and the 2022 CET1 ratio are labelled above. Yahoo Finance supplied the exact close, prior close, move, market capitalisation and share count; its 14 August snapshot reconciles to $15.23 divided by $13.93, less one, or 9.33% (Yahoo Finance 2026).
The Finance API sidecar resolved Nu Holdings Ltd. and found historical filings, facts and events, but its daily-price series stopped on 8 August and its filing list did not yet contain the 13-14 August Q2 packet. Those stale sidecar fields were not used for the market snapshot or trigger. SEC EDGAR, the filed 6-K documents, Yahoo's dated quote packet and Reuters' independent account supplied the point-in-time evidence instead. Better Fetch MCP tools were not available in this runtime, so the market scan used Yahoo's full-session gainers and losers feeds, StockAnalysis mover tables, primary filings and fetched news pages.
The residual-income values are author calculations, not company forecasts. Their cost of equity, retention, fade and terminal-growth assumptions are disclosed because small changes move the result sharply. There is no clean public vintage table by origination month, product and geography in the filing packet, so the precise loss curve behind the 6.9% NPL ratio remains the largest missing input.
References
- SEC 2026a: SEC EDGAR company record for Nu Holdings Ltd. (NU), used for legal identity and filing history.
- Yahoo Finance 2026: NYSE session snapshot for NU on 14 August 2026, used for close, prior close, move, volume, share count and market value.
- Nu 2026a: Nu Holdings Q2 2026 results release, dated 13 August 2026.
- Nu 2026b: Nu Holdings reviewed interim financial statements for Q2 and H1 2026, filed 14 August 2026.
- Nu 2026c: Nu Holdings Q2 2026 earnings presentation, filed 14 August 2026.
- Nu 2025: Nu Holdings 2025 Form 20-F annual report.
- Nu 2024: Nu Holdings 2024 Form 20-F annual report.
- Nu 2023: Nu Holdings 2023 Form 20-F annual report.
- Nu 2022: Nu Holdings 2022 Form 20-F annual report.
- BCB 2026: Banco Central do Brasil Selic target-rate series through 14 August 2026.
- Itau 2026: Itau Unibanco Q2 2026 results release, used for peer ROE, CET1 and NPL context.
- Reuters 2026: Reuters report on Nu's Q2 result and estimate comparison, syndicated by The Star on 14 August 2026.