NU › analyze
NU · Analyze
Full analysis · 2026-07-31 · Price $14.33 · Market cap ~$69.7B
Verdict: WATCH — excellent business, wrong price, and a mechanical downgrade catalyst 13 days away. Conviction 6.5. Fair value $9.50–12.50. Do not initiate before the Aug 13 print.
0. The one-paragraph version
Nu Holdings is a genuinely superior bank. It earns 2.5x the incumbents' return on assets with 44% less leverage, runs a 17.6% efficiency ratio against their ~40%, and took Brazilian card share at +36% against an industry +12.8% profitably, while its most dangerous rival bought share at a −6/−7% margin. None of that is in dispute and none of it is fragile. What is in dispute is the price. Three things the headline numbers hide: risk-adjusted net interest margin has been flat for two years while headline NIM hit a record — every basis point of spread expansion has been eaten by credit cost; Q1'26's earnings beat was a tax event, an 8.69% effective rate against 29.92% a year earlier, which sell-side consensus appears to have extrapolated; and ~41% of net income sits in the one product line that Brazil's payroll-lending boom is purpose-built to refinance away, where NU holds 0.4% origination share against Itaú's 18%. Corrected for currency and tax, fair value lands around $9.50–12.50. At $14.33 you are paying for 2028 with no margin of safety, into a quarter where even a clean print may not be rewarded.
1. Data integrity — read this before trusting any figure below
Six traps were found. Four would have changed a conclusion. This section exists because the agency has been burned by vendor feeds before ([[principle-primary-source-beats-vendor]]).
| # | Trap | Truth | Why it matters |
|---|---|---|---|
| 1 | Yahoo SharesOut = 3.81B |
Class A only. True total 4,861.7M (Class A 3,839.1M + Class B 1,022.6M). Diluted WAS 4,910.3M | 27% understatement. Every per-share figure breaks. Yahoo's own market cap silently uses the right number — only the raw field is wrong |
| 2 | Yahoo / roic.ai FY25 revenue $10.63B | IFRS Total revenue $15,774.7M; Managerial $16,319.6M | Vendor figure is a net construct understating revenue 32.6%. Any P/S or margin off it is wrong |
| 3 | Yahoo ECL allowance $4,165.7M | True $6,100M (Q1'26) | 32% understatement; coverage math fails without it |
| 4 | NU restated its entire history in Q4'25 ("Managerial P&L") | Q3'25 efficiency 27.7% → 20.3% (740bp). Q1'25 risk-adj NIM 8.2% → 9.3%. ARPAC Q4'24 $10.7 → $11.1 | The "every metric improving" narrative is materially cleaner post-restatement than it was in real time. Any series stitched from original releases has a step at Q4'25 |
| 5 | NPL basis switched | 2024–Q3'25 releases footnote NPLs "Brazil Only" (~10bp lower); Q4'25 onward consolidated | Genuine definitional break — do not mix |
| 6 | Q1'25 efficiency has two live values in NU's own documents | Press release prints 21.4%; NU's own spreadsheet computes 19.38% | The headline "−3.8pp YoY efficiency gain" is −1.7pp on NU's own mechanical basis |
One more, and it is the most consequential: Yahoo's quoted forward P/E of 12.39x runs off FY2027 consensus, not FY2026, and both years are built on a tax rate management has already guided away from. See §5.
Two of these generalize beyond NU and are being written to the knowledge base — dual-class foreign private issuers break Yahoo's share count, and a mid-series "managerial" restatement is a disclosure-quality event worth flagging wherever it appears.
2. Fundamentals
2.1 A note on method — FCF is meaningless here
NU is a bank. FY2025 OCF was $3.50B against capex of $341M, giving a headline "FCF" of $3.16B — a number that carries no information. That OCF embeds a −$12.04B change in receivables (loan growth) and +$4.13B in payables. For a bank, loan growth is capital deployment and deposit growth is funding; neither is discretionary cash. Capex is 98% capitalized software. EV/EBITDA, debt-to-assets and FCF CAGR are all excluded from this analysis as structurally distorted. The correct frame is retained earnings against regulatory capital growth.
2.2 Growth — strong and decelerating cleanly
| FY | Revenue (IFRS) | YoY | Net income | YoY | Diluted EPS | YoY |
|---|---|---|---|---|---|---|
| 2022 | $4,792.3M | — | −$364.6M | — | −$0.08 | — |
| 2023 | $8,029.0M | +67.5% | $1,030.5M | n/m | $0.2121 | n/m |
| 2024 | $11,517.1M | +43.4% | $1,972.1M | +91.4% | $0.4034 | +90.2% |
| 2025 | $15,774.7M | +37.0% | $2,871.7M | +45.6% | $0.5846 | +44.9% |
Revenue 3yr CAGR 48.8%. Net income 2yr CAGR 66.9%. Diluted EPS 2yr CAGR 66.0% — EPS tracking net income almost exactly, meaning dilution is not leaking growth.
2.3 Per-share discipline — the cleanest positive in the file
This is where NU separates itself from almost every fintech of its vintage, and it deserves explicit credit under the framework's per-share principle.
| FY2023 | FY2024 | FY2025 | 2yr CAGR | |
|---|---|---|---|---|
| Revenue / diluted share | $1.653 | $2.356 | $3.215 | +39.5% |
| Diluted EPS | $0.2121 | $0.4034 | $0.5846 | +66.0% |
| Diluted WAS (M) | 4,857.6 | 4,888.9 | 4,907.4 | +0.51%/yr |
| BVPS | — | $1.587 | $2.325 | +46.5% |
- Total shares +0.77% in FY2025, +0.12% in Q1'26. 28M of the 65M Class A increase was Class B conversion, not issuance.
- SBC fell 11.6% in absolute dollars in FY2025 while revenue grew 37% — from 3.24% of revenue to 2.09%, and 9.5% of net income in Q1'26. No options granted in 2023, 2024 or 2025. Remaining plan ceiling 4.69% of shares.
- $1.0B buyback authorized 2026-06-04 — NU's first ever, ~1.45% of cap. No dividend, none intended (20-F Item 8.A).
⚠️ Caveat: 34% of the Q1'26 equity build was FX translation (OCI), not earnings. That inflates the ROE denominator and explains most of the 33% → 29% ROE step-down.
2.4 Bank metrics — the core, and where the story turns
| Quarter | NIM % | Risk-adj NIM % | Credit yield % | Cost of risk % | Cost of deposits (% interbank) |
|---|---|---|---|---|---|
| Q1'24 | 20.01 | 9.63 | — | — | — |
| Q1'25 | 19.21 | 9.28 | 63.1 | 33.3 | 90 |
| Q3'25 | 18.58 | 10.81 | 59.6 | 24.8 | 89 |
| Q4'25 | 19.53 | 10.49 | 61.4 | 29.0 | 87 |
| Q1'26 | 21.15 | 9.48 | 62.7 | 35.4 | 88 |
Read the three comparable Q1s: headline NIM 20.0% → 19.2% → 21.1% (+110bp, a genuine record). Risk-adjusted NIM 9.63% → 9.28% → 9.48% — flat, −10bp over two years.
Two years of scale, AI underwriting and product-mix shift have produced essentially zero improvement in risk-adjusted margin. Every basis point of spread expansion has been consumed by credit cost. This is the single most important number in the analysis and it is not improving. Cost of risk at 35.4% is the highest in nine quarters. Credit yield is in structural decline — 71.0% → 62.7% over seven quarters — as mix shifts from revolving card to lower-yielding personal and secured loans.
Cost of deposits at 87–91% of interbank has been range-bound for two years. A genuine deposit franchise shows a falling cost of funds relative to benchmark. NU's has not moved. After the largest retail deposit-gathering exercise in Latin American history, that is an empirical result: NU has a distribution moat, not a funding moat. Brazilian incumbents fund partly through poupança (capped at 70% of Selic) and non-interest checking float — their blended deposit cost is structurally below NU's.
2.5 Credit quality — genuinely stable, and the strongest rebuttal to the bear case
| Quarter | NPL 15–90 % | NPL 90+ % | Allowance / portfolio % |
|---|---|---|---|
| Q3'24 | 4.5 | 7.0 (peak) | 15.7 |
| Q1'25 | 4.8 | 6.4 | 15.7 |
| Q4'25 | 4.1 | 6.6 | 15.2 |
| Q1'26 | 5.0 | 6.5 | 16.2 |
- 90+ NPL has held a 6.1–7.0% band for nine quarters and sits below the Q3'24 peak. On the back-book measure there is no deterioration.
- The 15–90 spike to 5.0% is seasonal and below Q1'24's 5.1%. NU's own bridge: +65bp seasonality, +17bp intentional risk expansion, +4bp mix, +2bp other. The 17bp of deliberate down-market expansion is the part to watch, not the seasonality.
- Coverage 249% of 90+; allowance at 16.2% of book is the highest reserve ratio in the series. Gross provisions ran 153.8% of new 90+ formation. NU is over-reserved, not under-reserved.
⚠️ But calibrate the peer claim honestly. NU's 90+ improved 10bp QoQ while StoneCo blew out 177bp and Inter +40bp — real outperformance. Yet 6.5% is the highest 90+ in the peer set (Itaú 1.9%, PagBank 3.05%, Santander 3.3%, Bradesco 4.1%, Inter 5.1%). Improving from the top of the range is a weaker claim than "best in class," and it reflects a deliberate risk-seeking posture — which argues for a higher cost of equity, not a lower one.
2.6 Funding — the constraint with a clock on it
| Quarter | Deposits $B | Net credit portfolio $B | Loan/deposit % |
|---|---|---|---|
| Q1'25 | 31.56 | 14.00 | 48.5 |
| Q4'25 | 41.93 | 19.03 | 49.1 |
| Q1'26 | 42.45 | 22.80 | 58.3 |
LDR jumped +9.2pp in a single quarter — the largest move in the series — as deposits fell 4% FX-neutral QoQ while the portfolio grew 7% FXN. At 58.3% NU still has real headroom versus incumbents at 75–85%, and available funding is 1.71x the net credit portfolio, so this is not a liquidity risk today. But the direction is one-way and accelerating. At even half this pace LDR reaches ~75% within a year. This is the most likely delivery mechanism for the bear case, and it binds in roughly 2–3 years.
2.7 Book mix — the diversification that is not happening
Credit portfolio $37.2B: cards $24.3B, unsecured $9.9B, secured only $3.0B = 8.1%. Non-interest-earning transactor float fell from 50.4% of the book to 40.8%. Purchase volume grew +17% FXN YoY against a portfolio +40% FXN — the loan book is compounding 2.4x faster than customer spend.
NU is converting from a payments/float business into a lending business: trading capital-light, fee-based, regulation-resistant earnings for capital-hungry, cyclical, regulation-exposed earnings. That raises yield and risk simultaneously, and it is precisely why NIM rose while risk-adjusted NIM did not. Quality of earnings is deteriorating even where quantity is not.
2.8 Geography — and a real transparency gap
NU reports as a single operating segment. There is no country P&L, no country ROE, no country asset disclosure anywhere in the filings. That is a genuine gap on the most important question about the stock.
| FY2025 revenue* | $M | % |
|---|---|---|
| Brazil | 11,038.3 | 91.3% |
| Mexico | 808.1 | 6.7% |
| Other (incl. Colombia) | 237.3 | 2.0% |
* IFRS 8.33 entity-wide table, a narrower definition than the income statement — it totals $12.08B, not $15.77B. Computing mix against the headline gives ~70% Brazil and is wrong.
- Brazil — 115.5M customers, ~100M active, 86% activity (record). Largest private FI in Brazil. ~7% of a >$100B annual gross-profit pool.
- Mexico — 15.0M customers, 3rd-largest FI, $5.9B deposits. First profitable quarter in Q1'26, ahead of internal plan. Efficiency 120% → 42% over four years. <1% of a $43B pool. Full bank conversion Aug 6, 2026 — deposit insurance jumps ~16x, unlocking salary domicile.
- Colombia — 4.7M customers, ~$2.8B deposits, still loss-making.
- Brazil standalone ROE was disclosed once, ever: ">40%" in Q1'24. Nothing since. Do not use a Brazil ROE number without a fresh source.
2.9 Scorecard
| Metric | Value | Trend | Verdict |
|---|---|---|---|
| Revenue 3yr CAGR | 48.8% | ↓ 67→43→37% | Strong |
| Diluted EPS 2yr CAGR | 66.0% | ↓ | Strong |
| Dilution | +0.77% FY25 | → low & falling | Strong |
| SBC % revenue | 2.09%, −11.6% in $ | ↓ | Strong |
| ROE / ROA | 30.3% / 4.60% | ↓ from 33% | Strong |
| Efficiency ratio | 17.6% | ↓ (guide ~20%) | Strong |
| Revenue / employee | $2.02M (+48% 2yr) | ↑ | Strong |
| NPL 90+ / coverage | 6.5% / 249% | → stable, over-reserved | Adequate |
| Capital (Brazil CAR) | 16.6% vs 10.5% req. | → | Adequate |
| Cost of deposits | 88% of interbank | → flat 2yr | Adequate |
| Risk-adjusted NIM | 9.48% | → flat 2 years | 🚩 FLAG |
| Cost of risk | 35.4% | ↑ 9-qtr high | 🚩 FLAG |
| Gross margin | 35.3% | ↓ lowest in series | 🚩 FLAG |
| Gross profit QoQ | −4.3% | ↓ first decline in 9 qtrs | 🚩 FLAG |
| Effective tax rate | 8.69% | ↓ from 29.9%; guide 15–20% | 🚩 FLAG |
| Loan/deposit | 58.3% | ↑ +9.2pp in one quarter | 🚩 FLAG |
| Cost to serve | $1.00/mo | ↑ broke $1; ARPAC/CTS 18.8→15.9x | 🚩 FLAG |
| Secured mix | 8.1% | ↑ slowly | Weak vs strategy |
| Segment transparency | Single segment | — | Weak |
3. Moat
3.1 The verdict: NARROW, stable-to-eroding
| Source | Strength | Read |
|---|---|---|
| Cost advantage | ★★★★☆ | Real and durable — but only against incumbents |
| Brand / intangibles | ★★★☆☆ | Best-in-class; converts to CAC advantage, not pricing power |
| Data / underwriting | ★★★☆☆ | Real; the outcome labels are the asset, not the raw data |
| Scale | ★★★☆☆ | Buys funding cost and zero-CAC cross-sell, not exclusivity |
| Switching costs | ★☆☆☆☆ | Structurally near-zero — Brazil engineered them away |
| Network effects | ☆☆☆☆☆ | Essentially absent. Do not credit NU with these |
3.2 Reconciling the central contradiction
ROE is 3x incumbents (moat evidence) while risk-adjusted NIM has been flat for two years (moat not translating into improving unit economics). Both are true, and they answer different questions:
ROA is the truer signal about whether a moat EXISTS. Risk-adjusted NIM is the truer signal about where it is GOING.
Leverage is a choice, not a moat, so ROE is contaminated. ROA is not — and on ROA the advantage is 2.5–3.1x and survives full tax and leverage normalization. That is a genuine, large, structural excess return.
But the first derivative is what determines terminal value, and the mechanism matters: NU's moat is being expressed as volume rather than price. It chose to grow the book +40% FXN at a constant 9.5% risk-adjusted margin rather than let the margin widen. That is defensible NPV logic while customers remain to acquire — but Brazil is ~68% penetrated (115.5M customers against ~170M adults) at a record 86% activity rate. Volume-expressed moats terminate; price-expressed moats compound. NU is nearing the end of the runway on which its chosen expression operates, and has not yet demonstrated it can convert the franchise into price.
3.3 Is record ROE a moat, or tax and leverage?
| Normalization | ROE |
|---|---|
| Reported (8.69% tax) | 29.0% |
| At 20% tax | 25.4% |
| At 25% tax | 23.8% |
| At prior-year 29.92% | 22.3% — indistinguishable from Itaú's ~21% |
Note the direction of travel the tax rate conceals: reported ROE fell from 33% to 29% while receiving a ~21pp tax tailwind. The pre-tax deterioration was larger than the headline shows.
But normalize ROA too and the answer flips back:
| NU (20% tax) | Itaú | Bradesco | |
|---|---|---|---|
| ROA | 4.03% | ~1.8% | ~1.5% |
| Equity multiplier | 6.59x | ~11.7x | ~9.7x |
| ROE | 25.4% | ~21% | ~14.6% |
Record ROE is not leverage — it is the opposite. NU earns 25.4% tax-normalized ROE on 6.6x leverage while Itaú earns 21% on 11.7x. Strip the tax and the ROE premium is only ~4pp, but NU earns it with 44% less balance-sheet risk. And this is not a benign-credit-cycle artifact: cost of risk is at a nine-quarter high and funding costs are +190bp. The cycle is already a headwind in these numbers, not a tailwind waiting to reverse.
Honest current-state ROE: 25.4%, not 29–30%.
3.4 Has the cost moat peaked?
Level: elite. Rate of change: peaked. Of the ~3.8pp efficiency improvement from Q1'25, roughly 1.3pp is durable and ~2.5pp is transient — management's own admission that only one-third is structural. Durable run-rate ~20.1%, essentially flat versus Q1'25.
Three arguments:
- Arithmetic exhaustion. Going 40% → 20% removes 20pp of revenue from the cost line. Going 20% → 15% removes 5pp. NU has banked ~80% of the ROE benefit the cost advantage can ever deliver. The remaining gap is a bigger source of future improvement for incumbents than for NU.
- The per-employee moat is the durable part — ~12,900 customers/employee vs roughly 700 for Itaú, an ~18x gap. The incumbent constraint was never technology; it is 90,000 unionized employees and a branch estate with political exit costs. Revenue/employee +48% in two years says this vector is alive.
- AI is a moat-NARROWING technology for NU, not a widening one. NU is already at the cost frontier — there is little fat left for AI to remove, which is why management can only claim one-third of a small gain. Incumbents sit at ~40% with enormous process fat and more capital to spend. Frontier models are purchasable; the binding constraint on incumbent cost reduction was organizational willingness, and 40%-efficiency banks under fintech attack are now willing. Any bull case treating nuFormer as a moat widener has the sign backwards.
Where the cost moat still wins decisively is defensively. At 17.6–20% efficiency NU profitably serves customers who are structurally unprofitable for a 40%-ratio incumbent. That is why Santander Brasil is exiting the demographic (mass-market clients 100M → 94M in a year) and why 339 sub-scale fintechs are out of BCB capital compliance as of this month, 679 by 2028, with Will Bank already liquidated. The cost moat is not producing rising margins — it is producing a shrinking competitive set, which is real and underrated.
3.5 Adversarial stress-test — quantified
Sensitivity constants: $125M net income = 1.0pp ROE · 100bp risk-adj NIM = 3.2pp ROE · $1B unsecured→secured = −1.06pp ROE.
The concentration that frames everything:
| Book | $B | % of book | Product ROA | % of NI |
|---|---|---|---|---|
| Brazil cards | 24.3 | 65.3% | 8.9% | ~57% |
| Brazil unsecured | 9.9 | 26.6% | 15.8% | ~41% |
| Secured | 3.0 | 8.1% | ~2–3% est. | ~2% |
~41% of net income sits on 26.6% of the credit book, in the single product line the consignado build-out is purpose-built to refinance away.
(a) Consignado crowd-out — the strongest structural bear point. CLT private payroll went 2% → ~6% of Brazilian non-mortgage consumer credit in a year (R$133B+, 10M workers). Origination share: Itaú 18%, BB 13%, Santander 10–11%, Nubank 0.4%.
| Migration | ΔROE | Resulting ROE |
|---|---|---|
| Mild (25% of unsecured) | −2.6pp | 27.7% |
| Base (50%) | −5.3pp | 25.0% |
| Severe (100%) | −10.5pp | 19.8% |
And NU has no good move. Management slowed private payroll citing 10–15% first-payment default and explicitly refused to price up. That is disciplined capital allocation and an admission: NU can cede the balances at −5pp of ROE, or write them at a default rate that destroys the 9.48% risk-adjusted margin anyway. There is no version where the unsecured profit pool survives at 15.8% ROA.
(b) Regulatory action on interchange or rates:
| Action | ΔROE |
|---|---|
| Credit interchange capped to 0.8% | −2.4pp |
| Credit interchange capped to 0.5% (debit parity) | −4.3pp |
| Revolving cap tightened, per 500bp of card yield | −3.2pp |
(c) Risk-adjusted NIM fails to recover:
| Outcome | ΔROE vs 10.5% guide |
|---|---|
| 9.5% (= current, two-year flat) | −3.2pp |
| 9.0% | −4.8pp |
| 8.5% | −6.4pp |
The base rate is unfavorable: risk-adj NIM has been flat at 9.28–9.63% for eight quarters. Guiding to 10.5% asks for a 100bp improvement that has not appeared once in two years. The market is capitalizing ~3.2pp of ROE that NU has never earned.
Combined bear (with overlap correction): ROE ≈ 17.6%.
3.6 The attacker NU cannot defend against is not a company
BACEN — Brazil's central bank — has a consistent, ideologically committed policy of competitively commoditizing every profitable layer of Brazilian banking.
| Action | Profit pool attacked |
|---|---|
| PIX (2020) | Payments, TED, boleto — annihilated |
| Open Finance (full since Apr 2024) | Data exclusivity, relationship lock-in |
| Law 14.690/2023 | Revolving card interest capped at 100%/yr |
| Prepaid interchange cap (in progress) | Fintech debit/prepaid interchange |
| PIX Parcelado — regulated credit, adaptation Mar 2026 | Directly attacks the $24.3B card book — brings ~60M cardless Brazilians into installment credit without a card |
| PIX Automático — mandatory Jan 2026 | Moves recurring billing off cards |
| Drex | Collateralized lending and capital-markets access |
NU's origin story is that it used BACEN's pro-competition agenda as a battering ram against complacent incumbents. That agenda has not stopped — it is now pointed at NU, the largest private FI in the country running a ~21% NIM on mass-market unsecured credit, in an election year. You cannot out-compete a central bank that builds free public infrastructure. NU can only adapt faster than incumbents, which it demonstrably does — but every adaptation cycle resets margins lower.
Attacks dismissed, with reasons: - Mercado Pago as existential — retracted. MELI deferred its Brazilian banking license ("Não é o momento agora", Feb 2026). More importantly, NU added card share at a 19% post-loss margin while Mercado Pago bought share at −6/−7%, growing its Brazil card book +36% against an industry +12.8%. Share gain at a 25pp margin differential is the cleanest evidence of competitive advantage in the entire dataset. - Big Tech — Meta discontinued WhatsApp business card payments in Brazil in January 2026, with ~99% smartphone penetration. PIX left no payments profit pool to capture. The irony: PIX destroyed NU's payment moat and Big Tech's entry vector. - Sub-scale fintechs — being regulated out of existence. - "It's a benign credit cycle" — no; cost of risk is at a nine-quarter high.
The attack with no rebuttal: funding. Cost of deposits flat at 87–91% of interbank for two years, with incumbents structurally cheaper. Combined with LDR at 58.3% and rising 9.2pp/quarter, NU has ~2–3 years before it must bid deposits up or go wholesale. Both compress the margin that is already flat.
3.7 Evergreen assessment
The company is evergreen; the current profit engine is not.
- Existence/position — very high confidence. NU will be one of the 3–4 largest Brazilian financial institutions in 2036. Scale banking franchises rarely die (Itaú 1945, Bradesco 1943, Banco do Brasil 1808 all still standing).
- Excess returns — much lower confidence. NU will not still earn a 21% NIM at a sub-20% efficiency ratio with no competitive or regulatory response.
NU has a cost moat in a deliberately commoditizing industry. A cost moat protects survival and decent returns; it does not protect extraordinary returns, because the regulator's explicit project is to compete those away. NU wins the war against incumbents — largely already decided — and then inherits an industry whose profit pool the central bank keeps redistributing to consumers.
Terminal ROE (10-year): 20.0% base, range 15–27%. Fade path: FY26 26–28% (tax-flattered) → FY27–28 24–26% → FY30 ~22% → FY33+ 20%. This is nominal BRL.
3.8 Moat-defense scorecard
| Defense | Grade | Read |
|---|---|---|
| Geographic (Mexico/Colombia) | A− | The only accretive defense. Mexico card ROA 14.1% > Brazil card 8.9%; <1% of a $43B pool; profitable Q1'26; full bank Aug 6. Worth ~+2.5pp terminal ROE |
| Ecosystem / SME | C+ | 5M SME customers at ~zero CAC is genuinely underrated, but no disclosed unit economics. Contributes 0 to base case. Brazilian SME is where the sector is currently blowing out |
| AI / nuFormer | B reliability, D magnitude | ~1.1pp of ROE, verifiable. But direction of trade is wrong — see §3.4 |
| NuPay / Amazon | B− strategic, C economic | Correct response to the MELI flywheel gap, but a rented surface, not an owned asset. Defends interchange by converting it into credit risk |
| Secured lending | D as moat, B as necessity | Mechanically ROE-dilutive: −1.06pp per $1B migrated. NU's principal answer to the consignado attack lowers ROE by construction |
4. Sentiment & intelligence
4.1 The framing correction that matters
NU is −24.5% off its high but +27.9% off its 52-week low of $11.20, set 2026-06-02/03. The 50-DMA is $13.14. This is not a stock in decline — it is a stock eight weeks into a recovery from a capitulation low, on 2.5–3x normal volume. Any analysis treating "−25% off the high" as an ongoing derating is reading the wrong tape. Equally, per [[principle-down-a-lot-is-not-cheap]], the bounce describes positioning clearing, not value.
4.2 The derating was four discrete events — and the big one was governance, not credit
| Date | Event | Move |
|---|---|---|
| Late Jan | Peak $18.98 | — |
| Feb 26 | Q4'25 print — record ROE 33%, but 2026 framed as an "investment year", efficiency guided up | −14.5% |
| May 14 | Q1'26 print — EPS $0.18 vs $0.20 cons., provisions +75% YoY, risk-adj NIM −100bp | −12% |
| Jun 1–3 | ★ CFO departure + downgrade cascade | 52-wk low $11.20 |
| Jun 4 | $1.0B buyback announced | the bottom |
| Jun–Jul | Mexico license (Jul 10), Brazil license deal (Jul 20), JPM PT→$20 | +27.9% |
On 2026-06-01 NU announced Rob Livingston (ex-Visa) as CFO, succeeding Guilherme Lago after 5 years. BofA cut to Underperform, PT $10, the next day. Lago gave detailed forward guidance on May 14 and announced his exit 18 days later, with credit metrics deteriorating — the worst possible sequencing. The market repriced a governance event as a credit event. Livingston started July 13; Lago stays through Aug 31.
4.3 Management tone — a structural shift, and it is the real signal
Comparing the Q4'25 and Q1'26 transcripts in full:
- Q4'25: asset quality got one paragraph — "we see no signs of deteriorations and remain comfortable." Credit was an afterthought inside a victory lap.
- Q1'26: credit became the architecture of the entire call. Vélez pre-empted it in opening remarks. Lago built a quantified attribution bridge (growth $423M / seasonality $267M / intentional risk expansion $69M / mix $16M — "growth and seasonality account for 86%"), with dedicated slides on cohort economics and coverage.
Three defensive tells, all new: 1. Lago volunteered the bear case by name — "I want to address directly a concern... Brazil's household debt service ratio." His rebuttal ("limited predictive power") is an argument you only make when the number is against you. 2. He pre-empted the tax question — "I want to be direct about our effective tax rate... because we know it may be a focus." 3. Vélez quantified a downside cap on the US — "even in a scenario where we do not find product market fit, the cost is less than 100 basis points on our efficiency ratio."
Credit where due: Lago's disclosure on private payroll was genuine analysis, not spin — "we're seeing 10% to 15% for payment default. That's a very high FPD for supposedly a secure product." Three of five analyst questions were on credit. BTG's Rosman raised the local-vs-foreign investor divergence explicitly; Lago named the three foreign concerns himself: asset quality, internationalization, AI.
4.4 Macro and regulatory
Brazilian consumer credit is deteriorating — in FLOW metrics, not STOCK metrics, and the deterioration is rank-ordered almost perfectly by credit growth rate and unsecured mix.
- System delinquency 4.7% in June 2026 — highest since the series began in 2011; individuals in free credit 7.6%; household debt/income 49.9% (all-time high); debt service ex-housing 27.1% (record since 2005); Serasa negativados 83.3M (record, 16 straight monthly increases).
- Selic 14.25%, and the market repriced hawkishly all year: Focus YE26 went 12.5% (Aug'25) → 14.00% (Jul 24). BCB raised its own odds of breaching the 4.5% inflation ceiling from 30% → 79%.
- Fast growers (MELI +87%, NU +40%, Inter +33%) show sharp provision flow moves; slow growers (Itaú +6.8%, Bradesco +8%) show mild ones. This is a seasoning problem in unseasoned 2025-vintage cohorts, not a household default wave — StoneCo said it outright.
- BRL appreciated ~9.5% over 12 months — a tailwind to USD-reported results, not a headwind.
★ The Brazilian banking license is the defining 2026 corporate action and carries a hard date. Resolução Conjunta 17/2025 bars BCB-authorized institutions without a commercial/multiple-bank license from using "Banco"/"Bank" anywhere customer-facing, with full compliance by ~2026-11-28. NU's stack (payment institution + SCFI + brokerage) has no bank license. Response: it agreed on 2026-07-20 to acquire Banco Porto Real de Investimentos — value undisclosed, subject to BCB approval, not yet closed. The brand is Nubank. This is a binary, brand-existential, hard-dated risk that appears unpriced.
★ Mexico confirmed: CNBV granted final operating authorization 2026-07-10; bank operations begin 2026-08-06. Obtained in a visibly tightening window — CAME was revoked, Finsus denied, Mercado Pago still pending.
Tax: LC 224/2025 raised CSLL on Nu Financeira to 17.5% effective 2026-04-01 (20% in 2028). Q2'26 is the first full quarter under the new rate. Offsetting: Lei 15.270 exempts income to R$5,000/month (~16M beneficiaries), landing on NU's core cohort — Lago flagged it as a tailwind not yet in provisions.
Election: Oct 4 / Oct 25. Lula 40–45% vs Flávio Bolsonaro 31–36% in R1. Largely orthogonal to NU's drivers — the consensus election trade is state-linked names (BB, XP), not NU.
4.5 Positioning
- Analysts: consensus still 1.64 "Buy" across 22, mean PT $17.98, range $10–$22 (2.2x spread). Cuts: BofA→Underperform $10 (Jun 2), Susquehanna→Neutral $13 (Jun 3), Citi→Neutral $13 (Jun 15). Then JPM raised to $20 (Jul 7) and Needham initiated Buy (Jun 26). Tell: when Citi cut on June 15, the stock did not make a new low.
- Institutions: 7 of top 10 added in Q1 — BlackRock +12.7%, Morgan Stanley +16.7%, JPMorgan +13.2%. Notable exit: Capital Research −21.7%. ⚠️ All 13F data is as-of 2026-03-31 and predates the collapse it is being used to interpret. Q2 13Fs land ~Aug 14, one day after earnings.
- ✅ Berkshire Hathaway does NOT hold NU — verified absent from the Q1 2026 13F (29 positions, $263.1B). No Berkshire overhang remains. Exit date unverified.
- Insiders — the cleanest contrarian negative. The stock fell from $19 to $11.20 and not one insider bought. Cristina Junqueira (co-founder) sold 300,000 shares at ~$14.81 on 2026-03-23; Anita Sands (Director) sold 21,000 at $12.24 on 2026-05-15 — the day after the Q1 print. Livingston's 2.19M-share award at $0.00 is a new-hire inducement grant, not a purchase ([[pitfall-yahoo-insider-purchases-counts-rsu-grants]]). The company bought; the people did not.
- Short interest: 3.2% of shares out, 2.06 days to cover, easing. Not a crowded short.
- Retail bullish, foreign institutions skeptical, Brazilian locals constructive — a divergence Rosman named from inside the call, being resolved by heavy volume (762M and 675M shares in consecutive July weeks vs ~200–250M typical).
| Dimension | Rating |
|---|---|
| Corporate | Bullish — operating record intact; the Q1 miss was provision-driven, not demand-driven |
| Sector | Bearish — every flow metric deteriorating; system delinquency at a 2011-series high |
| Macro | Bearish → stabilising — Selic 14.25% repricing hawkish, but IPCA cooling, unemployment at a series low, BRL the world's best performer |
| Retail | Bullish |
| Institutional | Neutral, turning positive — but Q2 13Fs unseen |
| Insider | Bearish — two sales, zero purchases across a 41% drawdown |
5. Valuation
5.1 The debate round changed this section materially
The Moat Analyst challenged the Valuation Analyst's currency specification. The challenge was correct, the Valuation Analyst conceded it, self-reported two further errors, and the corrected model moved fair value from ~$11.75 to ~$8.90 — about −25%. This is recorded because the process is the evidence.
The error: NU's reported ROE is BRL-nominal (Brazilian ops have a BRL functional currency; the FX largely cancels in the ratio). The Valuation Analyst's 14% cost of equity was built in USD. Discounting a BRL-nominal return at a USD discount rate credits the full nominal BRL return in the numerator while charging only the USD cost of capital — you get the currency depreciation for free.
Correct treatment: value the BRL stream at COE_BRL = (1+δ)(1+COE_USD)−1, translate at spot. Note (COE − g) is FX-invariant, so the terminal denominator does not change — but the residual-income spread (ROE_BRL − COE_BRL) does. Closed-form check: a bank at 20% BRL ROE, δ=3%, COE_USD 12% is worth 1.30x book; the mixed specification gives 1.67x — 28% too high.
Two further errors the Valuation Analyst found in self-audit: the residual-income engine computed net income on beginning book while using average-book ROE inputs (understating NI ~13%/yr, which raised value); and the 27.5% starting ROE was tax-flattered, leaving ~2pp of non-recurring tax benefit in the highest-PV years. The currency error and the average-book error were partially cancelling — the original $10.48 was accidentally close to right for the wrong reasons.
5.2 Models
| Model | Output | Weight | Why |
|---|---|---|---|
| Residual income (BRL, multi-stage fade) | $7.66 (own COE) / $9.31 (market COE) | 40% | Only model that fades ROE explicitly and anchors on hard book value |
| Normalized P/E | $10.00–12.90 on FY26E $0.77 | 25% | Corrects the tax distortion consensus is running on |
| Warranted P/B vs ROE | reverse-engineers the market's assumption | 15% | Correct framework, but single-stage Gordon is hypersensitive at 5.5x |
| Reverse DCF | implied return 7.3%/yr USD at spot | 10% | The decision gate, not a fair value |
| Sum-of-the-parts | $20–34B (28–49% of cap) is MX/CO/US | 10% | Kills the "free optionality" framing |
| Peer relative (P/B per ROE pt) | NU 0.184 vs ITUB 0.105, KSPI 0.082 | sanity gate | Cleanest evidence on the COE question |
| Bogle expected return | +8.8 to +10.8%/yr | cross-check | Independent confirmation |
| Graham IV | $6.15 | 0% | Structurally invalid — the 22.5 cap is a prohibition on high-ROE, low-book financials, not a judgment |
| DDM / DYT | N/A | — | No dividend, none intended. The buyback is a one-time 1.45%, not a yield |
5.3 🚩 The tax finding — highest-confidence, and it has a date
This survived every challenge and both analysts converged on it independently.
Q1'26 pretax was $954.3M; net income $871.4M at an 8.69% IFRS effective tax rate vs 29.92% a year earlier. Management guides 15–20% IFRS for the rest of 2026 (30–35% managerial). Q1'26 EPS of $0.178 is not a run-rate.
Consensus FY2026E of $0.886 implies $4.28B net income. Q1 delivered $871M, so Q2–Q4 must average $1.14B/quarter — a +30% sequential step-up held for three quarters — while management has guided the efficiency ratio worse (17.6% → ~20%) and said two-thirds of the Q1 opex beat was timing that reverses.
| Consensus | Tax-normalized | |
|---|---|---|
| FY2026E | $0.886 → 16.2x | $0.72–0.77 → ~18.6–19.9x |
| FY2027E | $1.156 → 12.4x | ~$0.95 → ~15.1x |
The quoted 12.39x forward P/E is FY2027 consensus built on a tax rate management has already disowned. The honest current multiple is ~18.6x — a ~50% understatement. PEG 0.85 is contaminated by the same error.
This forces a mechanical downgrade cycle that has nothing to do with credit, and Q2'26 is the first full quarter under both the guided 15–20% rate and the higher CSLL rate.
5.4 The robust conclusion — what $14.33 requires
Terminal ROE (BRL-nominal) needed to justify the current price:
| COE_BRL | Required terminal ROE | vs Moat's 20% base (range 15–27%) |
|---|---|---|
| 14.5% (market-calibrated from Itaú) | 26.2% | inside range — at the very top |
| 15.5% | 27.6% | outside |
| 16.0% (build-up) | 28.3% | outside |
| 17.5% | 30.4% | outside |
Even granting the market's own generous Brazilian cost of equity, $14.33 requires a terminal ROE at the extreme upper bound of the moat analysis — and above it under every other calibration. The COE debate moves fair value by ~$1.65; it does not move the verdict, because COE affects NU and its peers identically.
5.5 Manager's adjudication on the COE — I do not take the lowest number
The Valuation Analyst disclosed, unprompted, that its own build-up COE of 16.0% BRL implies Itaú is 30% overvalued, Bradesco 62%, Inter 32%. That is the tell. A model asserting that an entire liquid, heavily-covered sector is 30–60% mispriced more likely has a wrong discount rate than a genuine sector-wide arbitrage. I therefore weight the market-calibrated figure (COE_BRL ~14.5%, reverse-solved from Itaú), not the build-up.
I also decline the Sentiment Analyst's counter-anchor. Their $13.50–14.00 relies on a terminal growth rate of 11–12%, which is not a terminal rate — it is a decade-one growth rate. At ~4–5% Brazilian nominal USD GDP growth, an 11–12% perpetual rate means NU eventually becomes the Brazilian economy. Rejected. Their methodological point stands, though, and is worth carrying: a 130bp change in terminal g moves fair value 32%, so no single-point residual-income output deserves to carry a hard price line by itself.
Sentiment also could not reproduce the Valuation Analyst's "market implies 10.3% COE" claim, getting 12.8–13.4% instead. That specific figure is withdrawn from this report as unverified. The COE argument — that NU should carry a risk premium at or above Itaú's, not below — survives on independent evidence: an 8.1%-secured book, the highest 90+ NPL in the peer set, and 91% single-country concentration.
5.6 Fair value
Range $9.50 – $12.50 · base ~$10.50 – $11.00
Built from the market-calibrated residual income ($9.31 at terminal ROE 21%, $12.43 at 23%), cross-checked against normalized P/E ($10.00–12.90) and tempered for the terminal-growth sensitivity Sentiment identified.
| Case | Value | Prob. | Named driver |
|---|---|---|---|
| BEAR | $5.50 – $7.50 | 25% | Risk-adj NIM compresses below 9.5%; consignado crowd-out at base rate; terminal ROE 18%; COE widens |
| BASE | $9.31 – $12.43 | 50% | Risk-adj NIM recovers to ~10.5% as guided; ROE 25.5% → 21–23%; COE_BRL 14.5% |
| BULL | $13.15 – $15.50 | 25% | Mexico inflects, efficiency holds at 17–18%, secured scales without ROE dilution; terminal ROE 24–25% |
$14.33 sits inside the bull case and ~30% above the base.
5.7 Implied return
| Price | Implied USD return | P/B | P/E on normalized FY26E |
|---|---|---|---|
| $9.00 | 11.1% | 3.48x | 11.7x |
| $10.00 | 10.3% | 3.86x | 13.0x |
| $12.00 | 8.8% | 4.63x | 15.6x |
| $14.33 | 7.3% | 5.53x | 18.6x |
| $18.00 | 5.3% | 6.95x | 23.4x |
Bogle cross-check: 0% yield + ~15% earnings growth − ~6.2%/yr multiple compression + ~0.3% buyback ≈ +9%/yr. Three independent methods converge on a 7–11% expected USD return against a 12.7% USD cost of equity.
5.8 Sum-of-the-parts — Mexico is not free
Even under the most generous Brazil calibration (45% perpetual ROE, 8% growth), the market is paying ~$20B — 28% of the cap — for Mexico, Colombia and the US. Under central assumptions it is $30–34B, or 43–49%. Mexico did ~$1.06B of FY2025 revenue, making that residual ~19x Mexican sales. The bull framing that "you get Mexico for free" is inverted.
5.9 Peer relative
| Ticker | P/B | ROE | P/B per ROE pt |
|---|---|---|---|
| NU | 5.53x | 30.1% | 0.184 |
| ITUB | 2.28x | 21.8% | 0.105 |
| BBD | 1.08x | 13.4% | 0.080 |
| INTR | 1.23x | 15.5% | 0.079 |
| KSPI (closest global analogue) | ~3.85x | 46.8% | 0.082 |
NU trades at a 75% premium to Itaú and a 125% premium to Kaspi per point of ROE — and Kaspi earns a higher ROE at less than half the relative multiple. Kaspi carries its own governance and geopolitical discount, so this is not a clean read, but the direction is unambiguous: the "EM digital bank" premium is not category-wide. It is being paid specifically for NU.
6. Conflict resolution — what the agents disagreed on
| Conflict | Resolution |
|---|---|
| Fundamentals (risk-adj NIM flat = value not being created) vs Sentiment (stock NPLs best-in-fintech, derating was governance) | Both right, different questions. Sentiment is right that the June low was a CFO event misread as credit. Fundamentals is right that flat risk-adjusted margin over two years is the durable problem. Weighted to Fundamentals — a governance overhang clears in weeks; a flat risk-adjusted margin is a statement about the business model |
| Moat (terminal ROE 20%, BRL-nominal) vs Valuation (discounted at USD COE) | Moat won outright. Valuation conceded a genuine specification error and cut its base case ~25%. Credit to both — one for catching it, one for conceding and self-reporting two more errors |
| Valuation (FV $8.90, TRIM/AVOID) vs Sentiment (FV $13.50–14.00) | Split, closer to Valuation. I rejected Valuation's own build-up COE because it implies the entire Brazilian banking sector is 30–60% overvalued — a red flag on the discount rate, which Valuation disclosed itself. I rejected Sentiment's 11–12% terminal growth as a decade-one rate. Landed at $9.50–12.50 |
| Sentiment (entry zone unreachable) vs Valuation ($8–10 entry) | Sentiment withdrew, correctly, after running the barrier math: NU's realized vol is ~50% annualized, giving P(touch ≤$11.20) ≈ 45–50% and P(touch ≤$10) ≈ 30–35% over 12 months. The zone is a legitimate resting order — but waiting for ≤$10 accepts a ~65–70% chance of never owning the asset, and that is a choice to name, not free optionality |
Remaining unresolved tension, named as a risk factor: Sentiment's terminal-growth critique is unanswered in substance. If NU's runway (7% of Brazil's pool, <1% of Mexico's, a US charter option) supports even 150bp more terminal growth than modeled, fair value rises ~30% and the verdict shifts to HOLD. This is the strongest bull argument in the file and it has not been refuted — only judged less likely.
7. Verdict
WATCH · Conviction 6.5 · Fair value $9.50 – $12.50
This is a genuinely good business — better than most things in the portfolio on business quality — trading roughly 30% above a defensible fair value, with a mechanical earnings-quality repricing 13 days away.
What earns the 6.5 (the bull case, honestly stated): - ROA 2.5–3.1x Brazilian incumbents, surviving full tax and leverage normalization, with 44% less balance-sheet risk - Profitable share gain: +36% card book vs industry +12.8%, at a 19% post-loss margin while Mercado Pago bought share at −6/−7% - Per-share discipline that is genuinely rare: 0.77% annual dilution, SBC falling in absolute dollars, a first buyback - Over-reserved, not under-reserved: 249% coverage, allowance at a series-high 16.2% of book, CAR 16.6% vs 10.5% required - The competitive field is thinning: Will Bank liquidated, 339 fintechs out of capital compliance, Santander exiting the demographic, Mercado Pago deferring its license - Mexico is real optionality that just turned profitable, with a full bank charter live Aug 6 - Printing 30% ROE into a deteriorating credit cycle, not a benign one
What caps it there: - Risk-adjusted NIM flat for two years while headline NIM hit a record — the moat is being expressed as volume in a market that is ~68% penetrated - ~41% of net income in the one book consignado is refinancing away, where NU holds 0.4% share vs Itaú's 18%, and whose only defense (secured lending) is ROE-dilutive by construction - No funding moat — deposit costs flat at 88% of interbank for two years, with incumbents structurally cheaper, while LDR climbs 9.2pp/quarter - BACEN is an attacker NU cannot defend against, and its agenda is now pointed at NU - Single-country: 91.3% Brazil. No country P&L disclosed - Insiders sold into a 41% drawdown and bought nothing
Key risks, ranked
- 🚩 Tax repricing (near-certain, dated). Consensus FY26E appears built on an 8.69% rate management has disowned. Q2 is the first full quarter under the guided 15–20% rate and the higher CSLL. A downgrade cycle with nothing to do with credit.
- 🚩 Consignado crowd-out (structural, not cyclical). −5.3pp ROE at base migration. Not fixable in a quarter.
- 🚩 Banco Porto Real / BCB approval vs the 2026-11-28 naming deadline (binary, apparently unpriced). Failure forces a rebrand of the most valuable intangible NU owns across 115M customers.
- Risk-adjusted NIM fails to recover to the guided 10.5% — −3.2pp ROE, and the base rate is unfavorable.
- Funding constraint binds in 2–3 years as LDR climbs toward incumbent levels.
- Regulatory action on interchange or rates — −2.4 to −4.3pp ROE.
- BRL reversal. BRL appreciated 9.5% over 12 months; a reversal hits USD holders directly and is not in the reported numbers.
Action
| Zone | Price | Note |
|---|---|---|
| Genuinely attractive | ≤ $9.00 | ~11%+ implied USD return, base case intact |
| Accumulate | $9.00 – $11.00 | Straddles base. ~30–50% probability of being touched within 12 months |
| Hold / no action | $11.00 – $14.00 | Fair to rich |
| Do not initiate | current $14.33 | Inside the bull case, ~30% above base, with a negative catalyst pending |
| Trim | ≥ $15.00 | Above base under every calibration |
| Hard trim | ≥ $17.50 | Requires terminal ROE above the moat ceiling |
Do not initiate before the Aug 13 print, and both the Valuation and Sentiment analysts reached that independently. The asymmetry is negative: the SoFi precedent (Q2'26, reported 2026-07-29 — every credit ratio improved, guidance raised, stock fell 8.4% because absolute charge-offs rose with the book) says the market is currently penalizing fast-growing digital lenders on provision flow regardless of stock quality. NU will report absolute provisions up on a book growing 40%. A clean quarter may get partial credit; a soft one gets punished twice, on credit and on tax.
If entering later, stage it. A single $9 limit accepts a ~2-in-3 chance of never owning the asset. Three tranches — roughly $12.00 / $10.50 / $9.00 — capture most of the discipline without the all-or-nothing outcome.
Kill criteria at Aug 13 — any one moves the bear case to base: - 15–90 NPL fails to reverse from 5.0% - Risk-adjusted NIM stays ≤9.5% (a third consecutive flat print) - Gross margin declines sequentially again after 40.4% → 35.3% - Management walks back the seasonality decomposition
Free leading reads before NU prints: Itaú Aug 4, Bradesco Aug 5, MELI Aug 5, BB Aug 12. Itaú's individuals 15–90 (3.0%) and Bradesco's PF 90+ (5.4%) are the cleanest incumbent proxies; MELI's NIMAL (17.8%) is the closest unsecured-fintech read. ⚠️ BTG has called Q2'26 "provavelmente o período mais difícil do ano em termos de provisões."
A portfolio-specific passage was removed from the public build.
9. Sources
Primary (agent-browser / direct download — international.nubank.com.br, BusinessWire and SEC all 403 normal fetches):
Q1'26 Results Spreadsheet (XLSX — restated managerial series back to 2021) · Q1'26 press release · FY2025 Form 20-F · Q1'26 6-K interim financials · Mexico Operations Authorization 6-K · Q1'26 and Q4'25 earnings-call transcripts (roic.ai)
Corporate actions: CFO appointment · Banco Porto Real / Brazil license · $1.0B buyback · Reuters — false liquidation message
Macro / regulatory: BCB Focus 2026-07-24 · IBGE IPCA June 2026 · Agência Brasil — BCB credit stats 2026-07-30
Competitive: Exame — Mercado Pago defers Brazil license · BofA via euqueroinvestir — consignado origination shares · Rio Times — downgrade cascade · Brand Finance · SoFi Q2'26 8-K · Berkshire Q1'26 13F
Vendor (used, cross-checked, and found wanting — see §1): Yahoo Finance MCP, roic.ai MCP, .mcp/fin.py
10. Open items — flagged, not guessed
- No country-level P&L or ROE. Brazil standalone ROE last disclosed Q1'24 (">40%"). Consolidated 29% understates Brazil by an unknown but large margin.
- No CET1 ratio for Brazil — total-capital CAR only. Colombia's capital ratio undisclosed.
- Mexico/Colombia quarterly P&L never published. The Mexico break-even claim rests on a three-point chart.
- Q1'26 revenue-based taxes jumped +63% QoQ to −$217.9M against +7% revenue — unexplained in the release. Worth an IR question.
- Next-earnings-date discrepancy: Yahoo's
earningsTimestampshows 2026-11-11; two independent agent paths and the quarterly cadence (Q1 reported May 14) both give 2026-08-13. Report uses Aug 13; verify before acting. - Cost of equity band could not be tightened — Brazil NTN-B real yield and sovereign CDS were estimated from Selic rather than pulled live. Every valuation output is presented as a function of COE and shown robust across 14.5–17.5% BRL.
- Q2 2026 13Fs unfiled (~Aug 14). All institutional data is as-of 2026-03-31 and predates the collapse it describes.
- Berkshire exit date unverified — absence at Q1'26 is confirmed; when and at what price is not.
- Tencent, Sequoia, SoftBank positions and lock-up overhang — unverified; search budget exhausted.
- NU's UK redomiciliation — last confirmed reporting January 2025; no 2026 update either way.
- The "Nu Pagamentos 11% card share" figure circulating in July 2026 traces to 2023 data — do not use it. No 2026-vintage issuer-level Brazilian card-share table exists in reachable sources.