Holding the Dial, Losing the Gross Margin: MercadoLibre's Only Clean Beat Is the Tax Rate

Results vs. Consensus

MercadoLibre issues no formal financial guidance, so consensus remains the only pre-print bar. On the headline lines the print cleared it comfortably. Underneath, the composition is materially weaker than the two beats suggest.

Consensus dispersion was again wide. Providers set the Q2 EPS bar anywhere between $8.65 and $8.95 and revenue between $9,660M and $9,860M. The figures above use the most widely cited pair, which is also the pair the reported surprise percentages in the trade press are computed against. Direction is unanimous on both headline lines across every provider. No line-item consensus is published for gross margin, operating income, EBITDA or free cash flow, so those rows are scored against the prior-year actual and labelled accordingly.

The one segment miss matters more than its size. Argentina came in 7.1% light on revenue, and the shortfall is concentrated in fintech, where $1,182M landed against a $1,320M bar. That is the geography carrying the highest contribution margin in the company, and it is the one nobody asked about on the call.

Year-over-Year Comparison

Sequential Comparison

The sequential view is the one management directed the market to, and for once that framing is analytically correct rather than merely convenient. Q2 2025 predates the entire investment programme, so a year-over-year margin bridge measures the decision rather than its progress. The Q1 2026 column below is derived as the reported six months less the reported second quarter, and it reconciles line for line to the Q1 2026 filing: gross profit of $3,862M less the four operating expense lines totalling $3,251M gives $611M, the reported operating income.

Two facts sit in that table and they point in opposite directions. Operating margin was effectively unchanged, down 19bps on 15% sequential revenue growth, the smallest sequential decline since the investment programme began. That is the first sign of stabilisation. But the composition inverted relative to Q1: last quarter gross margin expanded 46bps while provisions consumed 284bps, and this quarter provisions gave back 152bps while gross margin surrendered 276bps. The pressure moved from below the gross profit line to above it, which is the less reversible of the two places for it to sit.

Revenue

This is the first quarter in MercadoLibre's history above $10bn of revenue, and reported growth of 49.8% is the fastest in four years. The internal picture is more nuanced than the acceleration suggests. FX-neutral growth of 42.9% actually decelerated from 45.5% in Q1 and 46% in the letter's own quarterly series, so the reported acceleration is entirely currency. Brazil is the exception and the engine: FX-neutral revenue growth accelerated to 42% from 39%, on GMV growth of 39% and items sold growth of 56%. Mexico held at 38%. Argentina decelerated hard, to 48% FX-neutral from 65% in Q1 and 77% two quarters before, as the inflation-driven base effect continues to unwind.

The commerce and fintech split has now converged. Commerce grew 40% FX-neutral and fintech 47%, against 39% and 54% in Q1. Fintech is decelerating toward commerce rather than commerce catching up, which is the arithmetic of a credit book that is compounding more slowly than it was and a financial services line growing 31.2% in dollars against the credit line's 71.8%.

Margins

The sequential margin bridge is the analytically important one, and management laid it out with unusual precision in the letter.

"Margin was broadly stable versus 6.9% in Q1'26, although its make up was different, with operating expenses as a percentage of net revenue 2.6ppts lower QoQ broadly offset by 2.8ppts of compression in our gross margin."
— Letter to Shareholders

Our computation gives 257bps of operating expense relief and 276bps of gross margin compression, so the letter's figures are accurate. What the letter does not do is characterise the trade. Operating expense relief came almost entirely from provisions, which fell 152bps of revenue because a Q1 spike in Brazilian consumer credit provisioning normalised. That is a one-quarter reversal of a one-quarter event. Gross margin compression came from pricing decisions in Brazil, higher shipping cost per unit of revenue, and device economics in Mexican acquiring. Two of those three are ongoing by management's own description, and the third is an industry input cost.

"The QoQ gross margin compression was driven primarily by: 1) initiatives launched on our marketplace in Brazil to improve price competitiveness and supply in certain categories and ASPs, as discussed earlier, which take the form of discounts for buyers who pay with PIX and take rate discounts for sellers, both of which reduce revenue; 2) higher shipping costs that were not fully offset by higher revenue; and 3) investment in user acquisition and higher POS device costs in Acquiring, particularly in Mexico."
— Letter to Shareholders

The cost-of-revenue bridge in the filings supports that ordering. Year over year, shipping and carrier costs rose $1,038M, cost of goods sold rose $567M, collection fees rose $226M, sales taxes rose $147M, fintech funding costs rose $131M and hosting rose $127M. Revenue taxes actually fell as a share of revenue, to 5.8% from 6.6%, which means the gross margin decline is worse than it looks on an underlying basis: the one favourable input moved in the company's favour and gross margin still fell 467bps.

EPS

Diluted EPS of $9.19 fell 10.9% against $10.31, on a share count that has again not moved. Weighted average diluted shares were 50,697,301 against 50,697,375 a year ago, and the company repurchased $1M of stock across the entire first half, a change of 662 treasury shares. There is no buyback. Every dollar of the EPS decline is earnings.

The composition of the beat is where the work is. The table below holds the tax rate and the foreign exchange line at prior-period levels and re-derives EPS from the same $606M of pre-tax income.

The filings attribute the tax move to Brazil, and specifically to deferred taxes: income tax expense fell "mainly as a result of higher income tax gains in Brazil in 2026 driven by the increase in deferred tax assets in that segment." Deferred tax assets on the balance sheet rose to $2,167M from $1,541M at December 31, a 40.6% increase in six months. A deferred tax asset build is a real economic benefit when the future taxable income materialises to absorb it, and it is a timing item that flatters the current period either way. It received no mention in the shareholder letter, no mention in prepared remarks, and no question from any of the fourteen analysts on the call.

Segment Performance

MercadoLibre reports by geography. Total direct contribution grew 3.1% on revenue growth of 49.8%, so consolidated contribution margin fell 688bps to 15.2%. That is an improvement in trajectory rather than in level: last quarter direct contribution fell 3.3% in absolute dollars, and this quarter it grew.

The corporate cost pool remains the only large cost line behaving the way a scaling platform's cost lines should. Unallocated costs grew 27.9% against revenue growth of 49.8%, delivering 144bps of leverage. That is real, but it is materially less than the 264bps of leverage the same line produced in Q1, and the narrowing matters: if corporate leverage is decaying while segment contribution margin is still falling, the consolidated margin has one less offset than it had three months ago.

Brazil

Brazil delivered the quarter's clearest evidence for the investment case and, at the same time, absorbed most of its cost. Revenue grew 59.2% in dollars and 42% FX-neutral, accelerating from 39% in Q1. GMV grew 39% FX-neutral, items sold 56%, and conversion rose 1.1 percentage points year over year against a base that already included the June 2025 free-shipping threshold reduction. Active sellers grew 29%. Contribution margin of 9.9% is down 563bps year over year but up 180bps sequentially, the first sequential improvement since the threshold change.

"Before the threshold change, conversion gains were incremental. Lowering it delivered a step-change – and twelve months on, that gain has proven sustainable, with conversion in Q2'26 up 1.1ppts YoY, even as we lapped the June 2025 threshold reduction."
— Letter to Shareholders

Lapping is the important word. A conversion gain that persists through the anniversary of the change that produced it is a permanent shift in the platform's attractiveness rather than a promotional spike, and the cohort evidence supports it: post-change buyer cohorts purchase more items across more categories with higher retention, and ecosystemic user growth in Brazil accelerated from 35% before the change to almost 50% now. The sequential contribution margin improvement is what the Q1 note said had to happen for the investment to be provable, and it happened.

Assessment: Brazil is the strongest single argument that this is investment rather than erosion, and the 180bps sequential contribution margin recovery is the first hard evidence that the curve bends. The offset is that the improvement came from the credit line normalising rather than from commerce economics, and management layered a fresh, unsized round of PIX and take-rate discounts on top in April. Brazil is 54.4% of company revenue, so the consolidated margin cannot inflect until Brazil's commerce economics inflect, and that has not yet happened.

Mexico

Mexico was Q1's quiet standout and is Q2's problem. Revenue grew 55.2% in dollars and 38% FX-neutral, roughly flat with Q1, but contribution margin fell 432bps sequentially to 13.1% after holding at 17.4% last quarter. Management attributed just under half the decline to the acquiring business and split that into two causes: an upfront investment in acquiring new merchants, and higher point-of-sale device costs driven by memory chip prices.

"This pace of expansion is creating some near-term margin pressure, driven by higher upfront investment in acquiring new users, where we continue to see healthy payback periods, and by higher device costs, reflecting the increase in memory chip prices."
— Letter to Shareholders

The call added a third element that the letter omits: a one-off charge from restocking device inventory in Mexico, because the devices are sold at a loss and the loss is booked when inventory is taken on. Management said that reverses across the balance of the year without sizing it. On the demand side, GMV grew 26% FX-neutral, decelerating from 28% in Q1, and management named three drags: the tax reform flagged last quarter, a weaker macro backdrop, and lower consumption during the World Cup, which it said was deeper in June and July.

Assessment: The margin damage is credibly explained and partly mechanical, but the demand commentary is the part that should carry forward. Management volunteered that consumption weakness was deeper in June and July than in the quarter as a whole, which is a negative pre-announcement about the first month of Q3 delivered without being labelled as one. Mexico now runs 1.4mn active acquiring devices, equal by management's count to all incumbent banks combined, so the device cost line scales with the strategy rather than against it.

Argentina

Argentina is the quarter's genuine disappointment and the least-discussed part of the print. Revenue grew 20.4% in dollars, the slowest in the company, and missed the Street's $1,980M by 7.1%. Fintech revenue of $1,182M missed a $1,320M bar by 10.5%. Contribution fell 5.7% in absolute dollars to $623M and margin fell 941bps year over year to 33.9%, the largest year-over-year decline of any geography for the second consecutive quarter.

"In Argentina, growth decelerated to 38% YoY in FX-neutral GMV and 22% YoY in sold items, against a challenging consumption environment."
— Letter to Shareholders

Three forces are compounding. The FX tailwind that inflated Argentine dollar revenue through 2025 is now a headwind, with the currency a material drag on dollar-reported revenue in the quarter. Consumption is weak. And the credit card launched in roughly September 2025 is still in its dilutive cohort phase, with the first cohorts only nine months old. Management was confident on credit quality specifically, noting that repayment behaviour is in line with underwriting expectations even as other Argentine financial institutions worry about the cycle.

Assessment: A 33.9% contribution margin remains extraordinary and Argentina still generates more absolute contribution than Brazil on a third of the revenue. But this segment has now given up 941bps of margin while decelerating to the slowest growth in the company and missing consensus on both the segment line and its fintech component, and none of that was raised on the call. Argentina is where the next negative surprise is most likely to originate, and it is currently the least-examined part of the story.

Other Countries

Revenue grew 63.0% to $463M with contribution up 132.1% to $65M and margin up 418bps to 14.0%, though down 208bps sequentially. Cross-border trade is the driver: FX-neutral CBT GMV grew 60% overall, with triple-digit growth in Brazil, Argentina and the smaller markets, and CBT now represents almost twice the share of GMV in smaller countries as it did a year ago. Volume through the Chinese fulfilment centre grew 170% sequentially.

Assessment: At 4.6% of revenue this cannot move consolidated results, but it is the cleanest read on whether the playbook travels, and it is the only geography where contribution margin expanded year over year. Growing contribution 132% while investing is a materially better outcome than any of the three large markets managed.

Revenue Composition: Commerce vs. Fintech

The geographic segments are the reported structure, but the product disaggregation in the filings explains the margin. It shows a mix shifting toward the two lines that dilute gross margin.

Commerce

Commerce grew 50.1% in dollars, with first-party product sales growing 80.1% and reaching 14.4% of total company revenue against 11.9% a year ago and 12.6% in Q1. Commerce services, which is the take-rate, shipping-fee and advertising line, grew 42.0%, so the mix inside commerce continues to rotate toward the lower-gross-margin inventory business. Advertising had the standout quarter within services: management reported crossing 10% share of the Latin American digital advertising market for the first time and 62% FX-neutral net revenue growth, and the CFO cited 73% growth in dollars on the call.

Assessment: The two lines diluting commerce gross margin, 1P and free-shipped low-ticket orders, are both growing faster than the take-rate business that carried the historical margin structure. Advertising is the credible structural offset and is compounding at more than the company average, but MercadoLibre still does not disclose it in dollars, so its contribution cannot be sized from outside. A quarter that crosses a named market-share threshold and still withholds the revenue number is choosing not to be modelled on that line.

Fintech

Fintech grew 49.3%, and the internal composition again matters more than the total. Credit revenues grew 71.8% to $2,278M and are 22.4% of total company revenue, while financial services and income, the lower-risk transactional line, grew only 31.2%. Fintech monthly active users reached 88mn, up 29.4%, with assets under management of $23bn, up 68%, and AUM per user of $264, up 29%.

Assessment: Fintech revenue growth is still principally credit growth, and credit revenue is booked gross of a provision charge. The good news relative to Q1 is that the ratio improved: provisions consumed 56.0% of credit revenue against 61.8% last quarter, although still above the 52.0% of a year ago. The deceleration in financial services and income to 31.2% is the number to watch, because that is the line that would carry the fintech business if credit growth slowed, and it is now growing at roughly half the credit line's rate.

Key Operating KPIs

Unit Economics Underneath the KPIs

Two rows changed the picture from last quarter. Average selling price stopped falling: down 0.6% year over year but up 5.1% sequentially, which says the low-ticket mix shift that free shipping produced has found its level rather than continuing to dilute. And capex intensity reversed. In Q1 capital expenditure grew 5.9% against revenue growth of 49.0% and intensity fell 120bps, which supported the argument that this is an income-statement investment cycle rather than a capital one, and therefore more reversible. This quarter capex grew 53.7% year over year and 62.7% sequentially, and intensity rose to 4.34%. That is only one quarter, and the absolute level is unremarkable for a company building logistics capacity into 44% volume growth. But the "easily reversible" characterisation is now weaker than it was.

Because KPI counts are reported in whole millions, the derived per-buyer ratios above carry rounding of up to roughly one percent. Management's own disclosed figure for consolidated items per buyer growth is 14%, with Brazil at 19%.

The Credit Book

Credit deserves separate treatment because it is where the quarter's two opposing signals both live. The good news is real and management led with it. The portfolio reached $16,375M gross, up 12.5% sequentially and 75% year over year, and every profitability measure improved.

The asset-quality table underneath tells a more divided story.

The 15-90 day bucket improved 97bps to 7.0%, which is the figure management cited three times across the letter and the call, and it is genuinely good. The 90-360 day bucket, which is where losses crystallise because the company writes a loan off at 360 days past due, rose 112bps to 18.7%. That is faster than last quarter's 79bps deterioration, and it happened while the book grew 12.5%, meaning the numerator grew 19.7% while the denominator grew 12.5%. Allowance coverage rose 50bps to 26.7%, which is management's own model reserving for exactly that.

The difference from last quarter is that this time somebody asked. The question and the answer are covered in the Analyst Q&A section below; the substance of the answer was that the metric mixes products, that it moves with origination pace, and that a late-2025 vintage defaulted somewhat above expectation and was subsequently curtailed. The first two points are correct mechanics. The third is a disclosure that had not previously been made, and it means at least one identified cohort underperformed underwriting.

Key Topics & Management Commentary

Overall Management Tone: Confident and noticeably more relaxed than the pre-emptive, argue-the-case posture of the prior quarter, with the format changed to match: brief opening remarks and then straight into questions, and a chief executive closing the call on strategy rather than a chief financial officer closing on numbers. Answers on commerce and engagement were evidence-rich and specific, and answers on credit asset quality were the least quantitative part of the call, leaning on the one delinquency bucket that improved while characterising the one that deteriorated as small. Management is no longer defending the margin decision; it is treating it as settled and moving the conversation to what the spending buys.

1. The Dial Held, But the Composition Rotated Above the Gross Profit Line

The single most important fact of the quarter is that the margin decline nearly stopped. At 6.7% against 6.9%, this is the smallest sequential deterioration since the investment programme began, and it was achieved on 15% sequential revenue growth. Management framed this as the correct lens.

"Income from operations of $683mn declined 17% YoY, with a margin of 6.7%, broadly stable QoQ, as we continue to set the dial in a way that prioritizes long-term value creation over short-term profitability – a trade-off we make carefully, and only where we see clear evidence of the economics that will follow."
— Letter to Shareholders

What the framing omits is that the stability is composed of a reversing item and a persisting one. The 152bps of provision relief reverses a Q1 spike that management itself described as temporary. The 276bps of gross margin compression comes from pricing decisions still in force, shipping costs still rising faster than shipping revenue, and device costs management says are industry-wide and outside its control. If provisions have now normalised, that relief cannot repeat, while the gross margin drag can.

Assessment: Stability at 6.7% is worth something, and it is the first evidence that the floor is a floor. But a stable margin composed of a one-off credit reversal offsetting a structural gross margin decline is not the same as a stabilising margin. The composition is the reason we are not upgrading on a flat sequential print.

2. The Brazil Price Investments Landed Exactly Where Management Said, Still Unsized

Last quarter management volunteered that the Brazilian take-rate reductions were implemented late in Q1 and would flow through the P&L in Q2. They did, and they are named as the first of three drivers of sequential gross margin compression. The April addition of PIX discounts for buyers, on top of the seller take-rate discounts, widened the programme further. Management reported the operational return: a meaningful improvement in the price competitiveness index and active seller growth of 29%.

The magnitude is still not disclosed. A quarter after being told the headwind was coming, and a quarter after it arrived and was identified as the largest single driver of a 276bps sequential gross margin decline, the market still cannot separate the take-rate investment from the shipping cost increase from the Mexican device charge. All three sit inside one number.

Assessment: The conditionality of the discounts remains the sophisticated part: a take-rate reduction contingent on the seller maintaining competitive listing prices converts a margin giveback into an enforced price-competitiveness mechanism. But an unsized headwind that is now the acknowledged largest driver of the company's most-watched margin line is a modelling problem no analyst can solve from outside, and it has now persisted across two consecutive quarters of explicit questioning.

3. Gross Margin Broke the One Signpost the Bull Case Rested On

Our Q1 note identified two consecutive quarters of sequential gross margin expansion as the specific evidence that would move this name toward Outperform. Q1 delivered 46bps of expansion. Q2 delivered 276bps of compression, to 40.9%, the lowest gross margin the company has reported in this cycle. The filings' own explanation adds a detail the letter does not: revenue taxes fell to 5.8% of revenue from 6.6%, so the underlying deterioration is worse than the reported one.

The letter's stated causes are all revenue-side or input-cost items rather than logistics efficiency failures. Price competitiveness initiatives reduce revenue at unchanged cost. Higher shipping costs not fully offset by higher revenue is a pass-through failure rather than a productivity failure. Memory chip inflation in POS devices is an input shock. None of the three is evidence that the logistics cost curve stopped bending, but none is evidence it is still bending either, because the metric that would show it was not disclosed.

Assessment: This is the quarter's clearest negative and it is the direct reason the rating does not move up. The gross margin was the line that had to inflect for the investment thesis to become a valuation thesis, and it went the wrong way by a wide margin. Two quarters of data now point in opposite directions, which is the definition of an unresolved trend.

4. Credit Asset Quality: the Bucket They Cite Improved and the Bucket They Do Not Cite Did Not

Management's asset-quality case is anchored on one metric, stated three times.

"in Q2'26, the 15-90 day NPL was 7.0% for the portfolio as a whole, and 4.6% for the credit card specifically – both near their historic lows."
— Letter to Shareholders

That is accurate and it improved 97bps sequentially. In the same filed table, the 90-360 day bucket rose to 18.7% from 17.6%. Total past due rose 56bps to 28.3%. The allowance rose 50bps to 26.7% of the book, which is the company's own expected-loss model registering the same deterioration. Write-offs of $712M grew 84.5% year over year against a book that grew 75%.

The mix explanation management offered is real. The credit card is now 46.9% of the portfolio against roughly a fifth four years ago, and the card's 15-90 day delinquency of 4.6% sits below the blended 7.0%. A shift toward the card should mechanically lower the blended 15-90 ratio. But the same mix shift does not explain a rising 90-360 ratio, because a growing, lower-risk share of the book should pull both buckets down.

Assessment: The profitability metrics genuinely improved and the cost of risk fell 380bps, which is the strongest credit evidence in four quarters. The late-stage delinquency trend did not improve, deteriorated faster than last quarter, and is now in its third consecutive quarter of increase. Both things are true; the disclosure emphasises only the first.

5. The Credit Card: Deliberate Dilution at 2.6mn Issuances a Quarter

Card NIMAL went from breakeven a year ago to negative 2.5%, and management was direct about why.

"Credit card NIMAL compressed from breakeven in Q2'25 to -2.5% in Q2'26. This is not because of a deterioration in asset quality; it simply reflects the significant step-up in issuance – 2.6mn new cards in Q2'26 versus 1.6mn in Q2'25 – as new cohorts are naturally dilutive."
— Letter to Shareholders

On the call, the fintech president added the cohort curve: Brazilian card cohorts reach NIMAL breakeven after 12 to 18 months, consistently, and Mexican paybacks are better than Brazilian ones because the industry economics are better. He also volunteered the reason the blended card NIMAL will not inflect soon: cohorts older than three years are a small part of the portfolio, and the majority of the book was issued in the last two years, so accelerating issuance keeps lowering average portfolio maturity.

Assessment: This is the clearest explanation of card economics management has given, and the 12-to-18-month cohort breakeven is a usable modelling input. The implication is uncomfortable for anyone underwriting a near-term margin recovery: the blended card NIMAL improves only when issuance decelerates, and management has committed to the opposite. The strategic argument for the card, that holders are two to three times more likely to be ecosystemic, is the real justification, and it is a customer-lifetime argument rather than a credit-spread argument.

6. Provisions and the Net-of-Provisions Defence

Management pre-empted the provision question in the letter with a specific analytical claim.

"if you net credit revenue against provisions, as a share of consolidated revenue, the result was slightly higher in Q2'26 than in Q2’25 – meaning the incremental revenue this business generates continues to more than keep pace with the incremental provisions required to support it."
— Letter to Shareholders

The arithmetic checks. Credit revenue less provisions was $1,002M, or 9.9% of consolidated revenue, against $636M and 9.4% a year ago. The claim is true and it is the right way to look at a lending business in an origination-growth phase. It is also narrower than it sounds, because it says the credit business is contributing more absolute margin dollars, not that the consolidated margin is improving, and the ratio it improves against is a year-ago base that predates the acceleration.

Assessment: A well-constructed defence that survives checking, and a meaningful improvement on Q1, when provisions consumed 61.8% of credit revenue against 56.0% this quarter. The residual concern is not whether credit is profitable, which it plainly is at a 58.9% gross yield against a 33.0% cost of risk. It is whether the reserve is adequate for a book whose late-stage delinquency is rising and whose fastest-growing exposure sits off the balance sheet.

7. Mexico: an Investment, a One-Off and an Input Shock in the Same Line

The 432bps sequential contribution margin decline in Mexico decomposes into three items management named but did not size: upfront merchant acquisition costs, a device restock charge booked upfront because devices are sold at a loss, and memory chip inflation raising unit device cost. Management said the restock charge reverses over the balance of the year, and that it deliberately did not raise device prices because paybacks remain in line and competitors did not raise theirs either.

The demand side is the more forward-looking disclosure. Mexican FX-neutral GMV growth of 26% decelerated from 28%, and management attributed the softness to the tax reform flagged in Q1, a weaker macro backdrop, and lower consumption during the World Cup, adding that the weakness was deeper in June and July.

Assessment: The margin damage is credible and partly self-reversing, and holding price into an industry-wide input shock while gaining share is the right competitive call. The June and July demand comment is the item to carry forward, because July sits inside Q3 and the comment was volunteered rather than extracted. Consensus has Q3 revenue at $10,566M, which requires roughly 42.6% year-over-year growth with one of the three large markets decelerating and a second one still unwinding a base effect.

8. AI: $80M a Quarter, an Engineering Headcount That Stopped Growing, and a Disclosed Return

MercadoLibre gave more AI detail than most companies of its size, and the detail was mostly about return rather than spend.

"AI investment grew roughly $80mn YoY in Q2'26, split between cost of goods sold and Product Development. Despite this, Product Development expenses fell from 8.4% of net revenue in Q2'25 to 7.2% in Q2'26, as the productivity gains described above allow us to grow without adding engineers at the pace we once would have."
— Letter to Shareholders

The operational evidence is specific: code submissions up 110% year over year, merged code up more than 100%, deployments up nearly 75%, and rollbacks down; agents independently reviewed more than half a million code submissions and migrated roughly 9,000 services over the past year; customer service headcount fell from 10,000 to 7,000 over four years while the business tripled, with 90% of interactions resolved without a human. The AI-powered search architecture completed rollout across the five largest sites, and management stated the conversion and advertising click-through uplift more than covers the third-party model cost even where the more expensive models are used.

The chief executive drew the boundary of the claim carefully, and so did the chief financial officer: the 118bps of product development leverage is mostly a multi-year efficiency trend rather than an AI effect, and the AI cost is inside that line rather than beside it.

Assessment: The Q1 commitment that engineering headcount stays broadly flat through 2026 was restated on this call, hedged as "probably", and product development delivered 118bps of leverage against 136bps in Q1, on a revenue base 15% larger. This is the most credible cost-side offset in the business and the only one management is willing to quantify. It is also, at 118bps against 467bps of gross margin decline, nowhere near large enough to change the consolidated picture on its own.

9. The Ecosystem Argument Replaced the Margin Argument

The letter's central section is an argument about ecosystemic users, meaning users who transact on both the marketplace and Mercado Pago. Management reported that this cohort grew 37% year over year, has been the fastest-growing user segment since Q4 2023, generated 70% more GMV and 55% more items sold per user than marketplace-only users, and produced almost 90% higher TPV and more than twice the AUM per user than fintech-only users.

"Most importantly, this deeper engagement results in improved profitability – contribution profit per ecosystemic user is multiples of the sum of the profit of a marketplace-only user and a fintech-only user. Profitability doesn't simply add when a user becomes ecosystemic – it multiplies."
— Letter to Shareholders

Asked directly what proportion of the base is ecosystemic, management declined to disclose it. Without the share and without the base rate, the multiple cannot be converted into a consolidated margin path, which is precisely the conversion an investor needs to underwrite the current spending.

Assessment: A genuinely differentiated structural argument, and the credit card evidence supporting it, that cardholders are two to three times more likely to remain ecosystemic, is the strongest link between the credit spending and the commerce business. But an unfalsifiable profitability multiple on an undisclosed share of an undisclosed base is a narrative rather than a model input. The company chose to make engagement the centrepiece of the quarter while withholding the one number that would let the market size it.

10. Unit Shipping Economics: the Metric That Disappeared

Last quarter management disclosed that Brazilian unit shipping cost fell 17% year over year in local currency, accelerating from an 11% decline in Q4, and warned that further gains would be non-linear. This quarter the metric was not disclosed at all, in the letter or on the call. What replaced it is a qualitative statement that free and slow shipments are now variable-contribution positive across half the price bands between R$19 and R$79, and, on the negative side, that higher shipping costs were not fully offset by higher revenue.

The qualitative statement is an improvement on Q1's disclosure that some price brackets had reached breakeven, so the direction of travel on that specific narrow measure is positive. But the headline unit-cost series, which was the single number that let the market judge whether the logistics investment was paying back, is gone in the quarter when gross margin fell 276bps.

Assessment: Management warned in Q1 that the pace of unit-cost improvement would slow. Withdrawing the metric in the quarter that would have shown the slowdown is the least favourable disclosure decision of the print. It is also the reason the shipping component of the gross margin decline cannot be separated from the pricing component, and it converts what should have been a testable claim into an assertion.

11. Cash Flow Turned Positive, and the Balance Sheet Kept Levering

Adjusted free cash flow was $214M, against negative $56M in Q1, and management volunteered it in prepared remarks rather than leaving it in the appendix as it did last quarter.

"Adjusted free cash flow was $214mn, even after $441mn of capex and $2.1bn of investment in the expansion of our credit portfolio (partially offset by $560mn of fintech funding)."
— Letter to Shareholders

The comparison the letter avoids is the year-over-year one: Q2 2025 adjusted free cash flow was $454M, so this quarter is down 52.9%. Net debt rose 11.8% sequentially to $6,425M and 37.2% since December 31. Operating cash flow for the six months of $5,737M is strong, but $4,069M of it went straight into loans receivable and $2,023M into restricted customer-fund balances.

Assessment: Positive free cash flow while funding a book growing 75% is a legitimately good outcome and materially better than Q1. The half-year picture is less flattering: $158M of adjusted free cash flow against $512M in the prior-year first half, on a business 49% larger. The credit book is consuming the cash generation of the commerce business, which is a choice rather than a problem, and it is the answer to why there is no buyback.

12. The Call Changed Shape

The format was different. Where prior quarters opened with a recorded video of quarterly highlights and a scripted chief financial officer message, this call opened with brief remarks and went straight to questions, and it closed with the chief executive on strategy rather than the chief financial officer on results. Fourteen analysts asked questions across roughly the same duration.

The content shifted with the format. Six of the fourteen questions went to credit, card economics or the loan book, against a more evenly distributed Q1. Two centred on AI economics. Two addressed margin, the opening question on the sequential gross margin and a later one on sustainability into the second half, and each was answered once rather than being pressed three separate ways as it was last quarter.

Assessment: The market has accepted the margin reset and moved on to the question of whether the credit book is safe. That is a rational reallocation of attention and it is where the residual risk actually sits. It also means the gross margin decline, which is the quarter's largest single negative, received one question and no follow-up.

Guidance & Outlook

MercadoLibre issues no formal financial guidance and none was introduced. What the company provides is a set of directional forward statements, which function as de facto guidance. The table below collects the ones made this quarter alongside the Q1 statements they update.

Street positioning after the print: Post-print consensus carries FY26 EPS of $38.17 on revenue of $41,469M, with Q3 at $9.42 and $10,566M.

Implied ramp required: FY26 EPS of $38.17 on 50.6968M shares implies net income of roughly $1,935M, a 4.67% net margin. The first half delivered $883M at 4.64%. The second half therefore needs about $1,052M on roughly $22,455M of revenue, a 4.69% net margin. That is a 5bps recovery rather than the 199bps the Street required after Q1.

Street at: Consensus has fully absorbed management's message. Hold the first half's realised 4.64% net margin flat across the year and FY26 EPS is $37.99 against a published $38.17, a gap of less than half a percent. Three months ago that gap was roughly 25%.

Guidance style: MercadoLibre has never guided and this quarter it declined a further disclosure, the ecosystemic share of its user base, when asked directly. The pattern is consistent: the company discloses growth rates and ratios generously and absolute levels selectively, and the levels it withholds are always the ones that would let an outsider size an investment or a headwind.

Analyst Q&A Highlights

Sequential Gross Margin Compression and the Ability to Pass Costs Through

The opening question went straight to the sequential gross margin, noted that two of the three causes disclosed in the letter were new to the market, and asked both whether the cost pressures could be passed through to customers and whether these headwinds had been contemplated when management said in May that Q1's margin was the level to expect. The answer decomposed the sequential move in order of importance and, in doing so, disclosed a Mexican device restock charge that appears nowhere in the letter.

Q: "There was a bit more quarter-over-quarter gross margin compression than we had perhaps expected. So thank you for also laying out the sources that drove this in the shareholder letter. Now when I think through each of those, the first one, I think, was well mapped. The second and the third one felt a little bit more incrementally new."
— Irma Sgarz, Goldman Sachs

A: "And then there was a one-off charge because we -- in Mexico, we restocked a significant volume in terms of devices. And that's because we are growing very fast in Mexico. We see a big opportunity to continue growing. And as you know, as we sell those devices at a loss, we need to book the losses upfront. So when we increase the stock, there's a one-off compression on margin. So that will compensate throughout the rest of the year."
— Martin de los Santos, CFO

Assessment: A more complete answer than the equivalent exchange last quarter, and it produced new information rather than a restatement. The second half of the question, whether these headwinds were known in May when the margin level was pre-committed, was not addressed. That matters, because the answer determines whether the May statement was a forecast that held or a forecast that was rescued by an unrelated provision reversal.

Whether the Margin Level Holds Through the Second Half

A follow-up asked directly whether margins are sustainable at these levels into the second half, and whether management would accept trade-offs if the operating environment deteriorated. The response identified the sequential drivers, described the reinvestment logic, and set out the framework by which investment intensity is governed without committing to any margin level.

Q: "EBIT margin came in pretty much as you anticipated sequentially in 2Q. It does seem like some of the investments were offset by improvements in NIMAL on the consumer credit side. Can you comment about the sustainability of margins at these levels in the second half? Are you committed to making trade-off if potentially operating environment somewhat changes and becomes unfavorable?"
— Deepak Mathivanan, Cantor Fitzgerald

A: "So that philosophy will continue to be the same going forward. We invest in a very disciplined manner. We have engagement and growth targets for each initiative that we take on. And more important than that, we have a clear path to profitability to those initiatives and we measure against those, and we lever and we graduate the intensity of investments based on those results."
— Martin de los Santos, CFO

Assessment: The same question that was asked three separate ways last quarter was asked once and answered once, which is itself a signal that the market has stopped expecting a numerical answer. The framework response is more informative than a level would be in one respect: it commits management to an internal test for each initiative, which means the spending is not open-ended in principle. It remains open-ended in practice, because none of the tests or their thresholds are disclosed.

The 90-Plus Day Past-Due Bucket

The sharpest credit question of the call did what no analyst did last quarter: it separated the improving short-dated delinquency metric from the deteriorating late-stage one, noted that short duration makes the short bucket a weak signal, and asked whether the cost-of-risk improvement was seasonal. The answer characterised the 90-day metric as mixed across products, mechanically sensitive to origination pace, and small, then disclosed a vintage that underperformed.

Q: "We saw definitely an improvement in provisioning levels this quarter with the cost of risk reducing Q-on-Q. We saw good trends on the short-term NPLs as well. But on the other side, as your duration is short, it's quite tricky to look only to the short-term NPL in our view. So actually, at the same time, the 90-day NPL had quite a meaningful deterioration this quarter. So my question is, how should we read that?"
— Kaio Penso Da Prato, UBS

A: "And so part of what is happening is that there was a change at the pace we were issuing at the end of last year and some early years borrowers defaulted a little bit higher than we expected. But then there was a slowdown at the pace we were issuing that product because of that deterioration. And so you see a little bit of a jump in the over 90 days. But it's not something that I would be concerned about. Again, that is a very small number."
— Osvaldo Giménez, President, Fintech

Assessment: The most consequential exchange of the call. The mechanical points are correct: a fast-growing book does dilute delinquency ratios, and short-duration products do concentrate defaulters in the tail because good payers exit quickly. But the characterisation of the bucket as very small does not survive the filings, where it is 18.7% of a $16,375M portfolio and $3,060M in absolute terms. The genuinely new information is the admission that a specific late-2025 vintage defaulted above expectation and that issuance of that product was subsequently curtailed. That is the first acknowledged underwriting miss in this cycle, disclosed only under direct questioning.

Brazilian Credit Cycle Risk Into the Second Half

A separate line of questioning put investor concern about a Brazilian credit deterioration in the second half and 2027 directly to management, and asked what protective actions were available for a cycle the company cannot control. The response denied any current deterioration and pointed to two prior episodes as evidence of the playbook.

Q: "Investors are quite concerned about the potential deterioration during the second half of the year and 2027. So just trying to understand here 2 parts of my question. So number one, if you understand that so far, so good. [...] And the second one is, what can you guys proactively do to protect yourselves from an eventual deterioration in the credit cycle, something that you cannot control."
— Rodrigo Gastim, Itaú BBA

A: "And with regards to what we expect for the future, I would say that we have been through a downgrade cycle in the past in Brazil, and we have been through -- we are right now about adverse macro conditions in Argentina. And in both cases, we have been cautious. And when we thought that we had to curtail the number of available lines or the lines of credit, we have done so."
— Osvaldo Giménez, President, Fintech

Assessment: A reasonable answer with a concrete precedent, and the claim that Brazilian NPLs are near cycle lows is supported by the 15-90 day series. The response is about willingness to curtail rather than about early-warning indicators, and willingness is the easier half. The company runs a book where expected losses are provisioned at origination, so curtailment stops the future charge but does nothing for the $14,047M of committed lines outstanding, against which only $52M of allowance is carried.

Card Portfolio Profitability and When the Blended NIMAL Inflects

Questioning on card economics probed whether a book above $7bn is approaching a profitability inflection given the spread the product should earn. The answer explained why the inflection is a function of issuance pace rather than portfolio size, and effectively told the market not to expect it while growth continues.

Q: "But given the level where it is right now, above $7 billion, and of course, the yields, the spreads on this product should be extremely robust right now. So is it fair to say that we are reaching a level where we could expect a profitability inflection?"
— Joao Pedro Soares, Citi

A: "We have been accelerating the pace of issuance. So if you were to look at cohorts that are older than 3 years, those are rather small part of the portfolio. And the majority of the portfolio has been issued, I would say, in the last 2 years. And that's why we need to need more time for the average of the portfolio to mature. As we are confident of how we are issuing, we have been able to accelerate the pace of issuance and that in some way slows down the average maturity of the portfolio."
— Osvaldo Giménez, President, Fintech

Assessment: An honest answer that closes off a bull argument. Anyone modelling a card-driven margin recovery inside the next twelve months is modelling against management's own description of the cohort mathematics. The blended card NIMAL is a function of issuance velocity, and issuance velocity is a strategy choice management has committed to keeping high. The card's contribution to the investment case is therefore ecosystem retention, not spread.

AI Economics and the Product Development Line

A question observed that the disclosed AI cost increase and the product development leverage together look strongly return-positive, and asked whether the trend continues. The answer was notable for correcting the premise rather than accepting the flattering version of it.

Q: "So I think you said the cost of tokens was up $80 million year-over-year, but you're saving, I think, almost 1% maybe of revenues in terms of product development. So it certainly seems like if you just look at those 2 dimensions that it's already quite ROI positive. So I'm just kind of curious, like as we look forward, could this be a trend that continues and maybe provides kind of more ROI and benefit to the P&L?"
— Josh Beck, Raymond James

A: "Let me just clarify the product development scale that you're seeing that I mentioned before, I think it was 1 point year-on-year. Most of that doesn't come from AI, just to be clear, okay? We have been scaling product development for many years now, and that trend continues as we become more productive and more efficient in the way we manage our product development. What I tried to explain is that even though we include the majority of the AI cost within that line, that line continues to scale, just to be clear."
— Martin de los Santos, CFO

Assessment: Management declined an easy narrative win, which is a governance positive and rare on this topic. The correction has a real modelling consequence: the product development leverage is a long-running operational trend that would have continued without AI, so it should not be extrapolated as an AI dividend that accelerates. The genuine AI returns cited elsewhere, in customer service headcount and in search conversion, are real but sit in cost of revenue rather than in the line the question was about.

Mexico Contribution Margin and Whether the Tax Headwind Is Fully in the Run-Rate

Questioning on Mexico split the contribution margin decline into its acquiring and non-acquiring components and asked how much of the acquiring damage is a passing input-cost shock against a deliberate customer-acquisition investment, then asked whether the tax reform headwind is now fully reflected. The answers separated cleanly, and the demand answer went further than the question required.

Q: "In terms of direct contribution margin compressing for Mexico, roughly half of that was from acquiring. How much of that was the device cost pressure related to memory chip inflation that will eventually grow over versus deliberate customer acquisition cost investment? And second, in terms of GMV decelerating with the tax reform headwind, is that headwind fully in the run rate now in -- was it fully in the run rate in second quarter?"
— Craig Maurer, FT Partners

A: "So we remain optimistic with the market, although we do see near-term challenges, which were a bit deeper in June and July with the World Cup. There's nothing that is actually impacting the huge secular trend of growth and opportunity that we see in Mexico and more importantly, the long-term earnings power that we find in that market."
— Martin de los Santos, CFO

Assessment: The most forward-looking disclosure of the call, and it was volunteered rather than extracted. Naming July as part of a deeper demand weakness is a comment about a month that sits inside the current quarter, from a company that gives no guidance. The acquiring answer was clear that the compression is device inventory and chip cost rather than a pricing decision, which is the better of the two possible explanations, but it also means the input cost persists until chip prices normalise.

What They're NOT Saying

1. Unit shipping costs, at all. Q1 disclosed a 17% year-over-year decline in Brazilian cost per shipment, accelerating from 11% in Q4, and warned that further gains would be non-linear. Q2 disclosed no unit-cost figure in the letter or on the call, in the quarter when gross margin fell 276bps and higher shipping costs were named as a driver. The series that would settle whether the logistics investment is paying back was withdrawn at the moment it became most informative.
2. The size of the Brazil price investments. The PIX buyer discounts and seller take-rate reductions are now identified as the largest single driver of the sequential gross margin decline. Two quarters after they were first flagged, they remain unquantified, so the pricing, shipping and device components of a 276bps move cannot be separated.
3. The effective tax rate, and what happens when the deferred tax build stops. The rate fell to 23.2% from 28.0% in Q1 on a Brazilian deferred tax asset that grew 40.6% in six months. It is the entire earnings beat. It appears in the filings, it is absent from the shareholder letter, and none of the fourteen analysts raised it.
4. Advertising revenue in dollars. The company reported crossing 10% share of the Latin American digital advertising market for the first time, with 62% FX-neutral and 73% dollar growth. It still does not disclose the absolute figure. This is the most credible structural offset to gross margin compression and it cannot be sized from outside.
5. The ecosystemic share of the user base. Management built the quarter's central strategic argument on ecosystemic users and their profitability multiple, was asked directly what share of the base they represent, and declined. Without the share, the multiple is not a model input.
6. Off-balance-sheet credit card commitments, now $14,047M. Unused agreed card lines grew 18.1% in the quarter against a drawn card book of $7,676M. Expected losses on committed lines are provisioned at issuance, so this balance is the forward pipeline of the provision charge. Second consecutive quarter with no mention in the letter and no analyst question.
7. Brazilian payroll loans. Flagged in Q1 as launching soon with government integration already complete. No mention in the letter or on the call, and no analyst asked. Either the launch slipped or it was immaterial; neither was stated.
8. The Mexican device restock charge, in dollars. Disclosed only in response to a question, described as compensating through the rest of the year, and never sized. It is one of three unsized components of a 432bps sequential contribution margin decline in the company's second-largest market.
9. The Argentine revenue miss. Argentina missed consensus by 7.1% on revenue and 10.5% on fintech revenue, decelerated to the slowest growth in the company, and gave up 941bps of contribution margin. It generated one question on the call, about the credit card ramp, and no discussion of the shortfall.
10. MELI+ subscribers. Subscriber growth of 72% year over year was disclosed for a loyalty programme presented as central to the engagement strategy. No absolute number, no revenue contribution, no penetration rate.
11. Any capital return. $1M of stock was repurchased across the entire first half, a change of 662 treasury shares. Capital allocation was not discussed on the call and no analyst asked. With $4,069M of the six-month cash generation absorbed by growth in loans receivable, the answer is implicit.

Market Reaction

• Pre-print setup: Closed at $1,922.57 on August 5 ahead of the after-market release. Down 4.6% year to date against an S&P 500 up 12.8%, down 20.2% over trailing twelve months, but up 6.5% over the prior thirty days. The 52-week closing range entering the print was $1,546.81 to $2,510.97, so the shares came in 23.4% below their 52-week closing high and 24.3% above the low.
• After-hours move: Down between 3% and 4.3% on the print evening, with the release at approximately 4:05 p.m. ET and the call at 5:00 p.m. ET.
• Next-day session: Opened at $1,825.51, a 5.0% gap down, traded a range of $1,766.05 to $1,833.21, and closed at $1,830.00, down 4.8% or $92.57.
• Volume: 1.2M shares against a 30-day average of 0.4M, roughly 3.1 times normal.
• Market context: The S&P 500 fell 0.2% on the reaction session, so essentially the entire move was company-specific.

The shape of this move is the informative contrast with last quarter. In Q1 a 7.2% after-hours decline widened to a 9.9% opening gap and a 12.7% close, because the numbers were available at 4:05 p.m. and their implication for the rest of the year only became clear during the call. This time the after-hours print of roughly $1,840 sat within 0.6% of the eventual $1,830.00 settle. The market priced the release correctly on first read and the call changed nothing. Volume of 3.1 times normal against 6.0 times in Q1 says the same thing: this was a repricing, not a repositioning.

What repriced was one quarter's margin composition, not the earnings base. The earnings base was already reset in May, and consensus has since moved to it. A 4.8% decline on a quarter that beat both headline lines, held operating margin sequentially and improved every credit profitability ratio is a narrow, specific reaction to one number: the 276bps of sequential gross margin compression that the first analyst question opened on. That is a proportionate response to the quarter's actual negative.

The setup made it a manageable move rather than a violent one. The shares came into the print 23.4% below the 52-week closing high, having rallied 6.5% over thirty days but still down 20.2% over twelve months. Expectations were not elevated, which is why a disappointing margin composition produced a single-digit decline rather than a gap. The residual question the tape does not answer is why a stock trading at 49.8x trailing earnings on a business whose gross margin is still falling absorbed the news this calmly.

Street Perspective

Debate: Is the Sequential Margin Stabilisation the Turn, or a Provision Reversal in Disguise?

Bull view: The bull case on the Street is that Q2 is the inflection quarter. The sequential margin decline nearly stopped, at 19bps, on 15% sequential revenue growth, while the company absorbed a fresh round of Brazilian price investment, an industry-wide device cost shock and a one-off inventory charge. Strip the transitory items and the underlying margin already turned. Direct contribution grew in absolute dollars after falling last quarter, Brazil's contribution margin inflected 180bps, and the corporate cost line is still delivering leverage. The investment cycle peaks in 2026 and operating leverage shows through from here.

Bear view: The bear camp reads the same table and sees a swap. The 152bps of provision relief is the reversal of a Q1 spike management itself called temporary, so it cannot repeat. The 276bps of gross margin compression comes from pricing decisions still in force and shipping costs still outrunning shipping revenue. A margin held flat by a non-repeating credit reversal offsetting a persisting gross margin decline is not a floor; it is the last quarter before the floor is tested without help.

Our take: The bear reading of the composition is correct and the bull reading of the direction is not yet wrong. Both facts are in the sequential table and neither is disputed. The decisive point is that the gross margin, not the operating margin, is the line that determines whether this business earns a premium multiple in 2027, and it went backwards by more than the operating margin held. Q3 resolves it cleanly: with provisions normalised there is no reversal left to spend, so the operating margin in Q3 is a direct read on whether gross margin has stabilised. That is a single quarter away, and it is worth waiting for.

Debate: Has the Credit Book Turned the Corner or Simply Grown Past Its Delinquency?

Bull view: Every profitability metric that matters improved. Cost of risk fell 380bps to 33.0%, NIMAL recovered 290bps to 20.7% with gains in the three largest markets, provisions fell to 56.0% of credit revenue from 61.8%, the 15-90 day NPL improved 97bps to 7.0%, card delinquency is 4.6% and near all-time lows, and renegotiated loans fell to 1.5% of the portfolio. Book growth decelerated from 87% to 75% while revenue growth held at 50%, so the gap that mechanically drives the provision charge is closing. Credit revenue net of provisions rose as a share of consolidated revenue.

Bear view: The improvement is in the flow and the deterioration is in the stock. Late-stage delinquency rose 112bps to 18.7% of the book, its third consecutive increase and faster than last quarter's 79bps. Total past due rose to 28.3%, allowance coverage rose to 26.7%, and write-offs grew 84.5% year over year. Off-balance-sheet card commitments reached $14,047M against a $7,676M drawn card book. Management characterised the deteriorating bucket as a very small number when it is $3,060M, and disclosed only under questioning that a late-2025 vintage defaulted above expectation.

Our take: The bulls have the better of the profitability argument and the bears have the better of the disclosure argument. Cost of risk falling 380bps while the book grows 75% is a genuinely strong result and it is the single biggest positive in the print. But the 90-360 day bucket is now three quarters into a deterioration that management has never volunteered, and the eight-month Brazilian consumer vintages written under the widened credit box described last quarter still have not completed a loss curve. The right posture is that credit is more profitable than the bears think and less transparent than the bulls need it to be, and that the resolution arrives in the second half when those vintages season.

Debate: Does 47.9x Forward Earnings Compensate for an Unresolved Gross Margin?

Bull view: The estimate risk is gone. FY26 consensus has been cut 20% to $38.17 and now sits within half a percent of a flat-margin outcome, so the denominator is finally honest and the residual risk is upside rather than downside. Enterprise value of $99.2B against FY26 revenue of $41.5B is 2.39 times sales for the dominant platform in a region where the digital transition has decades to run, growing revenue 50% with accelerating volumes and a credit business compounding at 75%. Buy the franchise while the earnings are artificially suppressed by spending management can stop whenever it chooses.

Bear view: The multiple got worse, not better. The shares are at 47.9x forward and 49.8x trailing against roughly 45x and 43.1x three months ago, because the price rose 12.1% while the estimate fell 20.2%. Investors are paying more for a business whose gross margin declined 276bps sequentially, whose free cash flow halved year over year, whose net debt rose 37% in six months, and whose entire quarterly earnings beat came from a tax rate nobody discussed. There is no valuation support at these levels for a business earning a 4.6% net margin.

Our take: The bears are right on the arithmetic and the bulls are right that the arithmetic is not the whole case. What has genuinely changed since May is that the estimate is now credible, which removes the specific reason we were unwilling to underwrite the multiple. What has not changed is that the multiple is higher than it was and the gross margin trend is unresolved. A 47.9x forward multiple on a business at trough margins is defensible if the trough is behind; it is not defensible if the gross margin has another leg down. That question gets answered in Q3, and paying up ahead of the answer, after a 12% rally, is the wrong side of the risk.

Model Update Needed

The comparison below is against post-print Street positioning and against the framework set out in our Q1 note.

Valuation impact: At the reaction close of $1,830.00 the shares carry a market capitalisation of $92,775M and an enterprise value of $99,200M against net debt of $6,425M. On our FY26 EPS range of $36 to $40 the stock trades at 46x to 51x, against 49.8x trailing twelve-month earnings of $36.77 and 2.39 times FY26 consensus revenue. Three months ago we said there was no valuation support for a constructive stance until consensus migrated toward the high $30s. Consensus has migrated, and the share price migrated with it in the opposite direction, so the arithmetic that produced our Hold is intact at a higher price on a lower number. We were right that estimates had to fall and wrong that they would fall before the stock rallied: the shares returned 12.1% against 4.2% for the S&P 500 since our initiation, which is a cost of waiting that has to be stated rather than buried.

Thesis Scorecard Post-Earnings

Overall: Unchanged in aggregate, with two offsetting moves. The estimate risk that was the binding constraint in May has been discharged: consensus is now an honest number and the gap between published FY26 EPS and a flat-margin outcome has closed from roughly 25% to under 1%. Against that, the specific operating evidence we said would justify an upgrade went the wrong way, and by more than the operating margin held. Credit improved on every profitability measure and deteriorated on the one asset-quality measure management does not cite. The thesis is not stronger or weaker than it was; the reasons for holding it have rotated from the estimate to the gross margin.

Action: Maintain Hold. The upgrade condition we set in May, two consecutive quarters of sequential gross margin expansion, was not merely unmet but reversed, and the second condition, a similar share price on a marked-down estimate, was met only on the estimate half while the shares rallied 12.1%. The stock is now more expensive on both trailing and forward earnings than it was when we initiated. The downgrade watch item, the 90-360 day past-due bucket, deteriorated faster than last quarter, but the profitability of the book improved enough that a negative rating is not supportable. Q3 is the clean test: with the Q1 provision spike fully reversed there is no offset left, so the operating margin will read directly on whether the gross margin has found its level. Revisit toward Outperform on a Q3 gross margin at or above 41.5% with the 90-360 day bucket flat, and toward Underperform if gross margin takes another leg down while late-stage delinquency continues to build into the seasoning of the eight-month Brazilian consumer vintages.