MercadoLibre

IThe business

MercadoLibre began in 1999 as an online auction site in Argentina. It is now two businesses built to feed each other. The Marketplace lets people and companies sell goods in 18 countries, and Mercado Pago handles the money in eight.² It is a Delaware company run from Montevideo, listed on Nasdaq since 2007, and it reports in dollars while earning almost nothing in them.²

In the twelve months to 30 June 2026 revenue was $35.2B, up 46% on the twelve months before.²³ In the second quarter it passed $10B for the first time: Commerce $5.8B, Fintech $4.4B.⁴ Brazil was 54% of the quarter, Mexico 23%, Argentina 18%, and everything else 5% C (calculated).³ It sold 795M items that quarter, about 100 a second, day and night C (calculated).³

How the marketplace earns

Third-party sellers make up most of the goods sold. MercadoLibre also sells on its own account in chosen categories, under 10% of volume, which it calls 1P.² It charges a percentage of each sale plus a flat fee on cheap items, collects shipping and storage fees, sells advertising and memberships. In the second quarter Commerce revenue was 26% of goods sold, though that figure mixes fees with 1P sales booked at full price C (calculated).⁴³

Three things in that list matter more than the rest. The first is shipping. Mercado Envios is the company’s own logistics network, with more than 50 fulfillment facilities that handled 55% of shipments in the first quarter, and nearly 80% in Mexico at the end of 2025.⁶⁵ The trucks, vans and planes belong mostly to carriers. The routing, the sorting and the promise belong to MercadoLibre.² In the first half of 2026 shipping costs rose $1.8B, the largest single increase in the cost line.⁴

The second is advertising, the best margin it has. Management says it passed 10% of Latin American digital advertising in the second quarter, with revenue up 62% FX-neutral.³ FX-neutral means with exchange rates held at last year’s levels. That letter gives growth and share, not dollars. The third is first-party selling, which gave almost $5B of goods in the past year. It still loses money after corporate costs. Management says so, and adds that the path to profit will take time.⁶ Its share of Brazilian cellphones went from 12% to about three times that in three years.⁶

How Mercado Pago earns

Mercado Pago started as the checkout for the Marketplace. It now processes payments for other merchants online and in shops, runs a digital account that pays interest, and lends. Total payment volume was $343B in the past year C (calculated), and $64.1B of the second quarter’s $101.0B came from acquiring, which means merchants outside the Marketplace paying it to process their sales.²³ Assets under management reached $23B.³

Credit is now the largest single line in Fintech. Credit revenue was $2.3B in the second quarter, 22% of the whole company C (calculated), up from $1.3B a year earlier.⁴ The book was $16.4B of gross loans at 30 June: $7.7B of credit cards, $5.8B of consumer loans, $2.5B to merchants and $0.4B backed by assets, mostly cars.⁴ It grew 75% in a year, and it was $3.8B two and a half years ago.³² Management’s stated aim is to be the region’s largest digital bank.⁶ That sentence changes what kind of company this is, and I come back to it.

Why they feed each other

The company’s claim is simple. A person who uses both the Marketplace and Mercado Pago is worth far more than the sum of two people who use one. Management says those users spent 70% more on the Marketplace than marketplace-only users in the second quarter, and that cardholders are two to three times more likely to stay on both sides.³ It says profit per such user is “multiples” of the two single-sided users added up, without giving the number.³ I take the direction as true and the size as unknown.

The mechanics are easy to see. A merchant’s loan can be collected from its sales as they pass through Mercado Pago, which cuts the risk of not being paid.² A buyer’s record on the Marketplace helps score a card. A card brings the buyer back. Each is an advantage a pure shop or a pure bank has to pay for.

The decision that explains this year

In June 2025 MercadoLibre cut the order value at which shipping in Brazil is free, to R$19 from R$79.⁵² In April 2026 it added discounts for buyers paying by Pix, Brazil’s instant payment system, and lower fees for sellers in some categories.³ The results were large. Items sold in Brazil rose 56% in the second quarter against 26% a year earlier, and unique buyers in Brazil rose 32% in the first quarter, the fastest in five years.⁶³ Fewer than a third of Brazilian adults bought on the platform in the fourth quarter of 2025, so there is room.⁵

The cost is just as plain. Operating margin was 12.2% in the second quarter of 2025 and 6.7% a year later.³ Income from operations fell 17% while revenue rose 50%. Management says this is a choice, and writes that when a business grows like this the answer is “not to harvest”.⁶ It also says that unit shipping costs in Brazil fell 17% in local currency in the first quarter, and that free shipments are now profitable before fixed costs across half of the price bands from R$19 to R$79.⁶³ So half are not. The whole valuation turns on whether this spending buys a lasting position or is simply the new price of staying in the game.

Who else is in the room

The 10-K names no rival, though it says competition has intensified, mainly in Brazil and Mexico, and that barriers to entry for large technology companies are relatively low.² Its own words deserve attention, because they are in a filing that lawyers read. Section III puts names and numbers on it.

IIThe moat

A durable business keeps others from taking its profits. MercadoLibre has been tested on this for 27 years, by local rivals, by Amazon, and now by Shopee and the Chinese platforms. It has so far kept growing. The honest question is what that has cost, and whether it costs more each year.

Evidence it is widening

Volume answers first. Brazil’s goods sold grew 39% FX-neutral in the second quarter, against UBS BB’s forecast, reported by Money Times, of 24% for the Brazilian online market in 2026.³⁷ If both hold, the company is still taking share at a size where that gets hard. Engagement answers second: items per buyer rose 14% and, in Brazil, 19%. The share of users buying in three or more categories reached a record.³ Those are company numbers, but they are specific and repeated quarter after quarter.

Cost answers third. Unit shipping costs in Brazil fell 11% in the fourth quarter and 17% in the first, in local currency.⁵⁶ Falling unit cost with rising volume is what a real scale advantage looks like. Shopee can lower prices too. To match the cost it must match the density, and density is earned over years.

Evidence it is narrowing

The clearest evidence is in the margin itself. A company with an unassailable position does not give up half its operating margin in two years. Direct contribution in Brazil, the profit after local costs, was 23.8% of revenue in 2023, 20.0% in 2024, 13.7% in 2025 and 9.9% in the second quarter of 2026 C (calculated).²³ Direct contribution in dollars in Brazil was $550M in the second quarter, up 2% on revenue up 59%.³ That is what a contest looks like from inside, whoever is winning it.

The take rate is the second sign. In the second quarter MercadoLibre lowered seller fees in Brazil, and management said so in the letter, as discounts conditional on competitive pricing.³ Sellers have choices now. Their fees are how a marketplace’s power shows up in money, and for now they are going the wrong way. Management has seen this before: it says similar cuts in 2024 and 2025 strengthened engagement.³ Perhaps. But each round leaves a lower floor.

In finance the wall is lower than in commerce. Nu Holdings has 139M customers against Mercado Pago’s 88M monthly users, a $39.4B credit portfolio, and $45.3B of deposits.¹⁷³ It earned a 33% return on equity with an efficiency ratio of 19.5%.¹⁷ Whatever data MercadoLibre has, Nu is already more than twice its size in lending, with a 15-to-90-day delinquency rate of 4.8% against 7.0%, though the mix differs, and it opened as a bank in Mexico in August.¹⁷³ The 10-K itself places Mercado Pago second by monthly users among fintechs in Brazil.²

My judgement: the commerce position is real and hard to dislodge, and I think it is being strengthened by the spending, not eroded. But the price of keeping it has gone up, permanently I suspect, and part of the old margin went to competition for good. In fintech MercadoLibre has a strong distribution edge and no lasting edge over Nu on the cost of money. That is a good business with a contested wall. It is a different thing from a toll road.

IIIThe competitive advantage

Size is the first advantage. UBS BB’s estimates, reported at second hand by Money Times in March, gave Mercado Livre 42% of Brazilian online sales in 2025, Shopee 19% and Amazon 11%, in a market of about R$450B.⁷ I can’t check those against a primary source, and estimates like these vary widely by method. They agree with the order the companies themselves imply.

The last column stopped me. Nu makes more than twice MercadoLibre’s net income at a smaller market value, and trades at about 13 times. So a buyer who wants Latin American consumer finance has a much cheaper way to own it, if the credit is as good as it looks. What the MercadoLibre price adds is the marketplace, and earnings that are depressed by choice. Nu’s are not. That doesn’t settle it, but it belongs in any honest comparison.

Where it wins

Logistics in the region, and the combination. Shopee earned $255M of adjusted EBITDA on $38.3B of goods in the second quarter, about 0.7% of volume across its whole footprint C (calculated).¹⁸ MercadoLibre’s operating income was 3.1% of goods sold in a quarter when it says it was investing heavily C (calculated).³ The two figures are not like for like, because MercadoLibre’s includes credit. Even so, one company has more room to spend than the other. In Mexico, it reached 1.4M active payment devices, which it says equals all the incumbent banks combined.³ And it has a cheap route to customers for Fintech that a bank must buy with advertising.

Coupang is the useful mirror. It built the same kind of dense logistics in Korea and earned a 1.4% operating margin on $34.5B of revenue in 2025 C (calculated).²¹ Amazon, with AWS and advertising on top, earned 11.2% C (calculated).²² A shipping-heavy retailer without a bank doesn’t make much. That is why the credit book matters so much here. I think MercadoLibre’s margin sits above Coupang’s mainly because of fintech and ads, and I expect that to stay true, though the company does not report profit by business line.

Where it loses

Cost of funds. Nu reports deposits costing 88% of interbank rates.¹⁷ MercadoLibre funds its book with securitizations, certificates of deposit and bank loans, and $6.5B of its $10.6B of financial debt was current at 30 June.⁴ It has applied for a banking license in Argentina, filed in June 2025 and still under review when the 10-K was written.² Until it can gather cheap deposits at scale, it pays more for money than its main rival in lending.

Cheap goods is the second. Asian platforms start with China’s supply. MercadoLibre’s reply is cross-border trade, up 60% FX-neutral in the second quarter, with its first fulfillment center in China opened in December.³⁵ It calls the cross-border market in the region about $10B and says it has a small share of it.⁵ This is a late start on someone else’s ground. Mexico is the third. Management says the tax reform hurt small sellers, and a third-party transcript of the call has it adding that June and July were harder, with the World Cup and a weak economy.⁶¹⁶ Goods sold there grew 26% FX-neutral, down from 35% in the fourth quarter.³

IVManagement

Marcos Galperin founded the company in 1999 and ran it for more than 25 years. On 1 January 2026 he became executive chairman, and Ariel Szarfsztejn became chief executive.⁸ Szarfsztejn is 44, joined in 2017, and worked his way through strategy and logistics to run Commerce.⁸ So the man now in charge built the part of the business where the money is being spent. Galperin, 57, stays involved in strategy and capital allocation, according to the proxy.⁸

How they are really paid

Almost no one at the top is paid in shares. No equity awards went to named officers in 2025, and diluted shares have sat near 50.7M since 2024.⁸¹⁵ Instead the company runs the Long-Term Retention Program, or LTRP: a target cash amount paid over six years, half fixed and half moved up or down by the 60-day average share price against the price at the start.⁸ For 2025 that starting price was $1,944.47, and Galperin’s target was $14M, the others $3M to $5M.⁸ The whole company’s accrual was $303M in 2025, about 1% of revenue.² Szarfsztejn’s 2026 award is $14M, nearly three times his 2025 target, against the average close of the last 60 trading days of 2025, about $2,095 C (calculated).²³¹ So half of his award is already about 22% under water at this price, and it pays out over six years only if he stays.

That has two effects. Shareholders aren’t diluted, which is rare at a technology company and worth real money. And the cost is already in the income statement as cash, so unlike at most peers on this site, there is no stock pay to subtract from cash flow. The cost even falls when the stock does: $165M in the first half of 2026 against $188M a year before.⁴

The annual bonus is small, four months of salary at target, and is 40% revenue, 35% income from operations, 10% payment volume and 15% customer satisfaction.⁸²³ All the money measures are in constant dollars, using last year’s exchange rates. Follow what that pays for. In 2025 reported income from operations was $3,201M. The bonus counted $3,942M.⁸ In 2024 the figures were $2,631M and $4,765M.⁸ The target for 2025 was $3,436M, so the officers beat a goal that reported dollars missed. In fairness, the managers don’t set exchange rates, and the bonus is too small to steer anyone. But shareholders live in reported dollars, and the people running the company are partly paid not to.

What the plan doesn’t reward says as much. Nothing pays for free cash flow, return on capital or credit losses across a cycle. Revenue and payment volume together are half the bonus. A lender that is paid for growth in a good year has a known way of ending up in trouble. The LTRP is the counterweight, because its six-year tail is forfeited on leaving, and that is the part I like.⁸

Their own money

Here the numbers are less comfortable. The new chief executive personally owns 76 shares, about $123,000 at this price. The chief financial officer owns 410. All 17 directors and officers together report 129,974 shares, 0.26%, and 97,933 of those are Stelleo Tolda’s, of which 20,000 are pledged in a prepaid forward sale.⁸ These people have large exposure to the share price through the LTRP. It is the kind that can’t lose them money they already had.

The exception is the family. The Galperin Trust and related entities report 3,400,136 shares, 6.7%, worth about $5.5B.⁹ Form 144 filings say Marcos and Karina Galperin settled the trust. Marcos Galperin himself reports 35 shares directly, and the filings don’t name the beneficiaries.²⁴⁸ The holding was 3,650,136 in May 2024 and 3,550,136 that October.¹⁰²⁵ In between came a Form 144 notice for 100,000 shares in August 2024, then notices for 150,000 in February 2025, at about $2,198, and 32,014 in May 2025, at about $2,551 C (calculated).²⁴²⁶²⁷ A Form 144 is a notice of intent, and the reported drop since May 2024 is 250,000 shares. The trust’s filings describe the wider restructuring as estate planning. Selling 7% of a holding near the highs isn’t a verdict on anything. It is still a fact, and the buyers at those prices are down about a quarter or more.

What they did with capital over a full cycle

The record is good on the things that count most. In March 2019, with the business losing money, the company sold $1.15B of stock at $480 and $750M to PayPal at $436.10.¹⁴ In November 2021, near a peak, it sold 1M shares at $1,550 for $1.52B net.¹³ It raised money when money was cheap and put it into logistics and credit, and revenue is now about 15 times the 2019 figure C (calculated). It has paid no dividend since 2018, and says it has better uses.² Buybacks of about $1.0B from 2020 to 2023 were mixed: $486M went out in 2021, nearer the high, and $504M in 2022 and 2023, at lower prices.¹⁵¹ Almost nothing has been spent on acquisitions: goodwill is $168M.³⁴ It built what it owns, and that is rare.

The one point against: today’s price is only 5% above what it charged buyers of the 2021 offering, five years on, while revenue is about five times as large C (calculated).¹³¹ That is mostly the market’s doing and not management’s. But margin and credit choices now rest on a new chief executive who has not run the company through a bad year. In his first two quarters operating margin was the lowest since 2021 C (calculated).³¹⁵ I can’t yet judge him on the only test that matters, which is what he does when a good idea stops working.

VFinancials

The ten-year record is simple in outline. Revenue rose about 34 times in nine years. The share count rose 15%, almost all from the 2019 and 2021 offerings. Operating margin went from 21.4% in 2016 to losses in 2018 and 2019, when free shipping first went in, then up to 14.6% in 2023, and has been coming down since C (calculated).²¹⁵ That dip is the most useful thing on the page: the company has done this before, and margin came back. But I think it came back partly because credit and Argentine float grew around it, and I can’t assume the same help twice.

What this year’s margin is made of

Between the second quarter of 2025 and 2026 operating margin fell 5.5 points. Gross margin fell 4.7 points, on shipping, 1P and fee cuts C (calculated).³ Provisions rose 2.3 points, from 10.2% of revenue to 12.5%.⁴ Product and technology fell 1.2 points as engineering headcount stopped growing and AI did more of the work.³ That last line is real: code submissions doubled while spending on it grew slower than revenue, even after about $80M more a quarter of AI costs.³

Provisions are the odd item. Under current accounting a lender books its expected lifetime loss on the day it lends. So a book growing 75% shows heavy costs at the start and pays them back later. Management makes that case, and it is correct as far as it goes.³ In the past twelve months the charge on loans was $4.2B while write-offs were $2.45B C (calculated).²⁴ The gap, $1.8B, was the reserve being built. If growth slows that gap closes and profit rises by itself. If losses arrive as predicted, the reserve was right. If they exceed it, the gain never comes.

Reported earnings against what an owner can take out

Operating cash flow looks wonderful and means little. It was $12.1B in 2025, six times net income, because $5.3B of customer money and $1.6B of card payables pass through it, and loans to customers sit elsewhere, in investing.² Most of that cash belongs to someone else: $13.1B of restricted cash and the $16.0B payable to customers roughly mirror each other.³

The company’s adjusted free cash flow removes the customer money, then takes off capital spending and loan growth, and adds back the borrowing that funds the loans. It was $1,481M in 2025, and in the past twelve months $1,127M, or 40% below net income of $1,863M C (calculated).²³ That is the cash left after everything management chose to do. At this price it is a 1.4% yield.

Then ask what an owner could take out if the company stopped growing and just stood still. No stock pay to subtract. Capital spending of $1.5B would fall to about depreciation, $1.0B, which is a guess at upkeep E (estimate). The loan book would need no new equity. Provisions would fall toward write-offs, adding about $1.3B after tax. That gives about $3.1B, or 3.8% of the market value, about 69% above reported earnings E (estimate). So the true owner number today lies somewhere from $1.1B to $3.1B, depending on how much of today’s spending is growth and how much is simply what it takes to stay where it is. In a price war that split is the whole question, and it can’t be read from any filing. Even the higher number needs a business that could stop growing and keep its customers. This one probably couldn’t.

The balance sheet

Total assets were $51.4B at 30 June and equity $7.8B, so this now looks like a finance company.³ The company’s net debt figure is $6.4B, but it includes leases and funding for loans.³ For valuing the business I use a narrower one. Available cash and investments were $6.75B, and the corporate notes due in 2031 and 2033 are $1.3B, leaving $5.5B, about $108 a share C (calculated).³⁴ In September it added $1.0B of 5.85% notes due 2036, rated BBB-, Baa3 and BBB-, which add equal cash and debt.³⁰ It is now investment grade with all three agencies. That is a real change for a company with most of its income in countries that are not.

Taxes need a word. The effective rate has swung from 21.4% to 36.6% in three years.² Brazil is the reason for some of it, Argentina the rest. There is also a $632M provision in Brazil for withholding tax on payments to Argentina, almost entirely covered by $593M of court deposits, and the company lost an appeal in July.⁴ Reasonably possible legal losses not accrued are up to $622M.⁴ None of that moves my values much. It does show how much of this company’s life is spent in court with tax authorities.

VIValuation

First, how this goes badly wrong

Before any numbers, I asked what would most likely ruin this investment, and worked back. Three things can.

1. The fight never ends. The danger is that MercadoLibre wins and still can’t raise prices. Shopee has its own investors, Amazon has AWS, and Chinese platforms have factories behind them. Each can run Brazil at a low return for years. If free shipping at R$19 and lower seller fees are simply the new terms, margin never recovers, and today’s 44 times earnings is a multiple of about the right earnings, not depressed ones.
2. The loan book meets a bad year. It has grown more than fourfold in two and a half years. $4.6B of it, 28%, is past due, of which $3.9B is more than 30 days late, against an allowance of $4.4B C (calculated).⁴ Unused card limits are another $14.0B.⁴ Loans turn quickly, which helps, and management says delinquency is near its lows.³ But the newest loans have not seen a bad year, many customers were overlooked by banks before, and $6.5B of financing falls due within a year.⁴ A lender can be right about its average borrower and still run out of cheap money.
3. The profits sit in the wrong currency. Argentina is rated Caa1, and it imposed capital controls in 2019 that it began to ease only in 2025.³¹⁴ It was 21% of 2025 revenue and 42% of direct contribution, and 40% in the second quarter C (calculated).²³ The peso went from about 16 to the dollar in 2017 to 1,482 at 30 June.²⁹⁴ Foreign exchange losses were $615M in 2023 and $337M in 2025.¹⁵ Brazil’s real has gone from 3.25 to 5.43 over the decade.³² Owners are paid in dollars and earned in these.

None of these needs a surprise. Two of them have already happened to this company, and the first is under way. They shape my bear, my rate and the wide range. A fourth sits outside the model: a regulator could change what a payments firm may do with $23B of customer assets. Brazil’s central bank already requires $11.2B to sit in guaranteed accounts.⁴

Two different questions

How good is the business? Very good, I think. It leads a large market that is still moving online, it builds what it owns, it does not dilute, and it reinvests at high returns: return on average equity was about 36% in 2025 C (calculated), though credit and float flatter that number. How good is the price? That is a separate matter. At 44 times trailing earnings it doesn’t look cheap, and on owner cash of 1.4% it looks expensive. It looks fair to modestly low only if the margin comes back. So this looks like a good business at a price that asks you to believe one thing, and that thing can’t be proved yet.

What the price implies

At $1,624.53 the market values MercadoLibre at $82.4B. That is 2.3 times trailing revenue, 44 times trailing earnings and 29 times Nasdaq’s 2027 consensus of $55.19 C (calculated).¹¹¹ The stock is 38% below its June 2025 high, while revenue is about 46% higher. The market has marked down what a dollar of revenue will earn.

Put simply, with my base growth the price needs margin to settle near 13.4%. The company made 12.7% in 2024 and 14.6% in 2023 C (calculated).² So the price doesn’t need a new record. It needs a return to about where the business was before this round of spending, on a far larger base, with far more credit in the mix. Or, at my margin, it needs dollar growth 1.3 points a year slower than my base. That is not a demanding bar, and it is why I come out above the price. The reason to doubt it is the first item above.

What I can’t judge

Two inputs matter more than all the rest, and I can’t know either. The first is the settled margin. Nobody has run a marketplace, a logistics network and an unsecured lender together at this size in these countries, so there is no good precedent. MercadoLibre says “significantly higher margins over time”, and gives no number.⁶ Coupang at 1.4% and Amazon at 11.2% bracket it badly.²¹²² Here is the range.

The second is the currency. I model in dollars, and I assume local growth stays three points above dollar growth. The real lost about 5% a year against the dollar over the decade, which at its share of revenue is close to that, and the Mexican peso held up.³² In the second quarter FX actually helped, with 50% in dollars against 43% FX-neutral, and that won’t last.³ The peso’s path is not a forecast anyone can make, and it does more damage than any operating risk because it hits everything at once.

I also can’t say how much of the $5.5B of corporate net cash is truly free. If a growing lender needs it as liquidity, take $108 off my values. And I can’t see profit by business line, because the company reports by country. Everything I say about credit’s share of profit is inference.

The discount rate

I use 12.0%. The capital asset pricing model starts with the 4.94% ten-year Treasury yield in Q4 2026 and Damodaran’s 4.20% implied equity risk premium of 1 October.³³³⁴ My ten-year beta on month-end returns is 1.43 and the five-year 1.41, which give about 10.9% C (calculated).¹³⁵ That can’t be the whole answer for a dollar investor in reais and pesos. Damodaran’s country premiums are 3.24% for Brazil, 2.46% for Mexico and 9.71% for Argentina, or about 4.3 points weighted by revenue, and adding all of that gives 15.2% C (calculated).³¹ That is too much, because a Nasdaq-listed beta already carries some of it and the company’s own ten-year dollar notes pay only 5.85%.³⁰ I add about a quarter of the premium. That is higher than the 10.6% on Uber and the 10.5% on Amazon. On TSMC I add no country premium because sales are in dollars, and on Tencent I use a 10.0% hurdle. MercadoLibre earns in reais and pesos, so it gets some. Above about 12.8% my base falls below the price.

Three cases

I keep the model simple. I set two numbers a year, dollar revenue growth and operating margin, and take cash from them. From operating income I subtract a recurring foreign exchange loss, 0.6% of revenue in my base, because it has been there most years. Tax is 28%; the last three years were 36.6%, 21.4% and 29.7%.² Capital spending is 3.6% of revenue against depreciation of 2.9%; it was 4.6% in 2025 and 3.7% in the first half C (calculated).²³ I run fifteen years, as on my Visa and Mastercard pages, because growth is still high at year ten.

The one unusual line is the loan book. Each year I grow it with revenue, to 55% of revenue, from about 47% today C (calculated). I count 72% of it as net of the allowance, and I charge owners 50% of the growth in that net book, about the recent record: 47% in 2025 and 54% in the first half C (calculated).²³ The rest is funded by fintech borrowing. Deposits or a bank license could lower it, but I can’t count on one yet. Because that borrowing funds the loans, I don’t subtract it again as debt, and I leave customer money out altogether. Every extra ten points of own funding costs about $77 a share. Base 2026 revenue of $42.0B is six reported months plus 42% growth on last year’s second half E (estimate). I start the count of cash with 2027, and leave the second half of 2026 out.

Is 166B of revenue in 2036 reasonable? It is 14.7% a year in dollars for ten years after 2026, from a company that averaged 48% over the past nine C (calculated). It is less than a quarter of Amazon’s 2025 revenue, for a region management says has half the online share of the US.²²⁵ I think about half of it would come from lending and payments, and I can’t test that. Against the history it isn’t a stretch, and my earnings sit close to the Nasdaq consensus to 2028.

• Price$1,624.53Close, Q4 2026 [1]
• Base value$1,822My estimate E (estimate)
• Margin of safety below$824My bear is three quarters of this E (estimate)

My margin-of-safety price is $824. I set it so that if the bear case proves true, a buyer at that price loses no more than a quarter. It is 55% under my base, and it is far below today because the range is wide. Weighted 30%, 55% and 15% for bear, base and bull, my value is $1,640, about 1% above the price. I put more on the bear than the bull because the bull needs more things at once, as Section VII shows. At $1,624.53, my bear value is 62% lower and my bull 86% higher, so the range is not comfortable. A base value above the price is not the same as a margin of safety, and today there isn’t one on my numbers.

Why the market may be right

The market isn’t stupid, and I may be the one who is wrong. Earnings per share fell in both quarters of 2026 while revenue grew about 50%.⁶³ There is no guidance, so nobody knows when the spending stops. The letter says the dial is set to favor long-term value over short-term profit, and gives no date.³ A new chief executive has room to spend early in his term. And lending this fast, in these countries, has ended badly for others. At 13%, a point above my discount rate, my base is $1,579, below the price. It is a thin edge on a good story.

The next best use of the same dollar

Treasuries pay 4.94%.³³ MercadoLibre’s own 2036 notes carry a 5.85% coupon and sit ahead of the shares.³⁰ For the stock to beat them by a fair amount over ten years, owner cash has to rise from about $1.1B to about $13B, which is my base. A bondholder needs none of that.

The best comparable on this site is Uber, another local network with logistics, where the $65.74 on that page is 24% below my $86 base, at a lower discount rate and without a loan book. On those published figures Uber offers a wider gap with less to go wrong. Amazon is the template for the same playbook, and my page has its price 47% above base. Outside the site, Nu offers Latin American credit at 13 times earnings with a 33% return on equity.¹⁷²⁰ It doesn’t have the marketplace. What you pay extra for at MercadoLibre is Commerce, and the option that margin returns. For the credit alone, this is the dearer way to own it.

Where this could be in 5 and 10 years

Each value is my model’s value at that date: the present value, in that year’s dollars, of the owner cash that comes after it, at 12.0%. It uses today’s 50.7M shares and keeps net cash at today’s $5.5B. Dates are year ends, counted from Q4 2026.C (calculated)

Dollars per share, rounded to $10. Owner cash is summed and not discounted. The return is the 2036 value plus that cash, against $1,624.53, over ten years.C (calculated) Values are my estimates.

For the base to stand at five years, revenue has to reach $103B in 2031 at about a 12.5% operating margin, with owner cash near $5.5B that year. At ten years it is $166B at 14.5%, with owner cash near $13B. Value climbs because the cash gets closer, and the base pays out about $1,280 a share on the way.E (estimate)

The cases differ most on the settled margin, 9.0%, 14.5% and 17.5%. Then come the currency, where the bear loses five points of growth a year and the bull two, and the loan book, where owners fund 60%, 50% or 40% of its growth. About two thirds of the base’s 2036 value is cash after 2041, so the rate matters too: at 13% it is $3,290.E (estimate)

The most likely way the base is wrong is margin. If it sits 2.5 points lower in every year and settles near 12.0%, the 2036 value is about $2,880, not $3,660, and the 2031 value about $2,220.C (calculated)

These are estimates of value, not forecasts of the share price. The market price can sit above or below value for years.

VIIThe case for the price

What most people miss

Three things cut in the company’s favour and rarely make the headlines. Part of today’s margin loss is accounting and reverses by itself when growth slows: in the past twelve months the loan charge ran $1.8B above write-offs C (calculated).²⁴ There is no share dilution at all, because management is paid in cash; the count was 50.70M in 2024 and is 50.70M now.¹⁵³ And almost none of the assets were bought. Goodwill is $168M on a company with $51.4B of assets, so what you see in the numbers is what it built.⁴³

The bull case

In plain words: Latin America has seven online purchases a person a year against 41 in the US, and MercadoLibre’s buyers already make 11.⁶ More than half of Mexicans rely on informal credit, and 85% use cash for most small purchases.⁶ Management writes of a “once-in-a-generation opportunity”, and of today’s investments giving way to much higher margins.⁶ It has done this before: free shipping began in 2016, operating losses followed in 2018 and 2019, and then came the leading position.⁶¹⁵ Its case is that the same thing is happening now in Brazil, in cards, in Mexico and across the border.

I looked for named views from the investors in this series, and found little I can verify. Altimeter, Atreides and Berkshire Hathaway held no shares at 30 June.³⁶³⁷³⁸ Duquesne sold its 2,766 shares in the second quarter.³⁹ The holders I can show are growth funds. At 30 June Baillie Gifford had 3.04M shares, down from 3.23M; Tiger Global 153,126, up from 135,006; D1 277,475, down from 373,675; Sands Capital 378,208, down from 475,455; ARK 22,044, down from 23,988.^4041424344 Those filings are four months old, and they show more selling than buying as the price fell. They don’t say why.

Against reality

Today’s numbers support the growth half of the case. Revenue up 43% FX-neutral at this size is rare.³ Credit quality looks fine so far: the 15-to-90-day delinquency rate was 7.0%, near historic lows, and net interest margin after losses was 20.7%, down two points in a year on the shift to cards.³

The profit half is unproved. If they keep unit shipping costs falling, if card cohorts break even in 12 to 18 months, as a third-party transcript has them saying, and if Brazil’s contribution margin recovers from 9.9% to the mid-teens, then a settled margin of about 13% is possible, and the price with it.¹⁶³ Today operating margin is 6.7%, so it has to about double. Card net interest margin went the wrong way, to −2.5% from about breakeven, though management puts that down to 2.6M new cards in the quarter.³

It helps to multiply out what my bull needs. Margin above any year since 2016, 35%. Dollar growth of 17% a year for ten years, 40%. No serious credit loss across the cycle, 70%. Currency drag of two points or less, 50%. Those are my guesses and they aren’t independent, since a strong region helps most of them. Multiplied, they give about 5% E (estimate). With the overlap, perhaps 15%, the weight I give the bull. Each step is believable. All four together are not likely.

What a 10x would take. Ten times $1,624.53 is $16,245, about $824B of market value. At 25 times earnings that needs about $33B of net income. At a 12% net margin, against 9.2% in its best recent year, that is $275B of revenue, or 21% a year in dollars to 2036. My bull has $211B and about $26B of net income, which at 25 times is about eight times today E (estimate). MercadoLibre did rise 9.7 times in the past ten years, from $168.01, but that began at about $7B of value, not $82B.¹ NVIDIA has gone about five times since May 2023 from near $1T, in the strongest demand cycle anyone has seen.⁴⁵ So tenfold in a decade would take my bull case, plus more, plus a rich multiple on a company by then growing about 10%. For most companies this size it is close to impossible. Here it is merely very unlikely, and I would not build a case on it.

Where I come down

The volumes moved me, along with the falling unit shipping cost: that is what a real scale edge looks like. The letters did not, because every company in a price war says it is investing. A year in which revenue growth holds above 30% and Brazil’s contribution margin rises would raise my values most.

VIIIRisks, and what would change my view

The three ways this goes badly wrong are at the start of Valuation. These are the rest.

• Regulation of payments and credit. Interest caps, card fee rules and mandatory reserves can move overnight. Brazil’s central bank already requires a large part of customer money to sit in guaranteed form.⁴
• Tax. Brazil’s consumption tax changes begin in 2026 and run to 2033, with a split-payment system.² Mexico’s reform is already biting.⁶ Pillar Two adds a minimum tax in some countries.²
• Funding costs. Interest on the funding of the book is in cost of revenue, so higher local rates cut margin directly. The transcript also has management saying Marketplace margin moves with rates.¹⁶
• Founder. Galperin remains active as chairman, and a full exit would remove the person who made the 2016 and 2025 shipping decisions.⁸
• AI shopping. If an assistant from another company becomes the place people start shopping, the advertising line suffers. The company’s AI search rollout is early, and its results are its own.³
• Fraud and security. Buyer protection costs rose $155M in the first half.⁴ A large breach at a company that also holds deposits would be worse than one at a shop.
• Supply costs. Memory chips raised the cost of payment devices this year, a small sign of how far outside shocks reach.³