Stelleo Tolda on beating Amazon in Brazil, and how MercadoLibre stopped being asset-light

The J Curve is an English-language podcast and newsletter covering Latin America's technology and venture capital ecosystem, hosted by investor Olga Maslikhova. In this episode: Nigel Morris, who co-founded Capital One inside a small Virginia bank in 1994 and grew it from roughly $1 billion to nearly $20 billion in market value, then co-founded QED Investors and backed Nubank before it existed — after David Vélez pitched him "a Capital One in Brazil" over breakfast in Mexico City. He explains why Nubank, Revolut, Klarna and Monzo are all now aiming at the same market, why banks structurally cannot innovate, why a company growing 40% with an NPS of 89 currently cannot raise growth equity, and what he thinks genuinely changed in Latin America — which is not capital, regulation, or technology.

What did Latin American e-commerce actually look like in 1999?

Nothing existed. Internet penetration in Brazil was two to three per cent, on dial-up modems, on desktop computers most families saw no reason to own. Nobody had bought anything online, so the product needed evangelising before it needed selling — store owners had no concept of the internet as a sales channel, and buyers were unwilling to pay for something they couldn't see. Infrastructure was missing across the whole region, not just Brazil.

And on top of that, the classic marketplace problem. Tolda names it plainly: every marketplace starts at day zero with zero sellers and zero buyers, the way Uber started with no drivers and no riders. What made it worse was that the technology to run an auction had become commoditised — you could launch by building a front end — which is why there were eighty of them.

Why did MercadoLibre survive when eighty competitors didn't?

Three things, in Tolda's account, and the first two are linked. Obsessive focus on the user experience, and a decision to build their own technology early because they wanted the experience to be better. Most competitors believed the game was marketing — that it was about eyeballs, and eyeballs were bought with advertising. MercadoLibre played some of that game too, at their investors' urging, until March 2000 made efficiency the only thing that mattered.

The third was duration. The founding team believed they were democratising commerce in the region and were in it for the long run. When the funding stopped, most of the competition wasn't.

“At the time we said it was B2C and B2B — back to consulting and back to banking. Where we stayed on.“

Stelleo Tolda

What happens when the bubble bursts two months into your round?

The Nasdaq began falling in March 2000. MercadoLibre closed its second round in May, straight into what became the internet winter. Some investors wanted out — not a down round, but to shut the company and get their money back. One investor, Susan Segal, then at Chase Capital Partners, was decisive in closing it. Tolda notes she still sits on the board today, twenty-six years later.

The money came with a condition attached that shaped the next decade:

"This is the last money that you'll have to make this company viable.”

The caveat from MercadoLibre’s investors, May 2000

It took seven years to reach the IPO. Half that money was still in the bank when they got there. Everything superfluous went — including a Miami office opened on the theory that they might serve US Hispanics, a market where they had no advantage over eBay. Marketing became portfolio management, run as maths: test channels, measure return per dollar. They became one of the first significant advertisers on Google in the region, contracting out of Mountain View before Google had Latin American offices, buying keywords at almost no competition and therefore almost no cost.

Who blinked first on charging fees?

The best story in the episode, and a clean piece of game theory. For its first twelve months MercadoLibre was entirely free in every market, living on investment dollars. Everyone in the industry knew the model was transaction commissions. Nobody wanted to move first.

MercadoLibre went first, in Brazil, with Tolda running it — the first market in the company and, he says, in Latin America to charge at all. The fee was 2.9%. Two weeks later DeRemate, the strong Argentine regional player, started charging 2.7% in Brazil. A week after that Lokau, a local Brazilian player, came in at 2.5%. And the fourth, iBazar — a French company running France, Spain, Italy and the Netherlands, whose Brazilian operation had been started by one Frenchman living there who volunteered — charged nothing at all, planning to monetise through banner advertising against traffic.

MercadoLibre had the highest fees of the four. It eventually bought all three of them. Tolda's reading: it was never about the fees. It was about experience, technology and trust.

Why did MercadoPago exist before it was a payments company?

It was built to solve the trust problem, not the payments problem. In the original model the buyer paid the seller in advance and hoped — which meant fraud risk, broken goods, and no recourse. MercadoPago was created to sit between them: the buyer pays MercadoPago, MercadoPago tells the seller to ship, the buyer confirms receipt, and the funds are released to the seller net of fees. Argentina in 2003, Brazil in 2004 — eight years before Amazon arrived.

The lesson that prompted it came from watching eBay. eBay had its own payments product and its users chose PayPal anyway, an external tool, because it was better. Tolda's conclusion was that the best product wins even inside someone else's platform, and that payments were therefore not something they could afford to leave to others. Only later, as more payment methods were connected, did it become an engine worth selling to third parties — one integration giving a regional player Brazil, Argentina and Mexico at once.

He frames the whole ecosystem this way, and rejects the idea it was ever a master plan. The concept he uses is adjacencies: marketplace to payments to credit to shipping, each step taken because the experience was incomplete without it.

Why throw away eleven years of working code?

Because by 2010 the codebase was what they called the monolith — everything on one code base, spaghetti code, the CRM written on the same code as the front end and the back end. Which meant that when the site went down, customer service went down with it, so users couldn't even report that the site was down.

They rebuilt from scratch over about two years: APIs, web services, the first apps, third parties able to connect. Tolda is unusually direct about how much rode on it.

It was a very courageous move at the time to throw away all the technology and build everything from scratch. That was a game changer for the history of MercadoLibre. We probably wouldn't be speaking here if we hadn't decided to do that, and if we hadn't executed well.

The timing is the point. All of it happened before Amazon arrived in Brazil.

What did MercadoLibre do when Amazon arrived in Brazil?

Amazon entered Brazil in 2012 and was aggressive. Tolda's response was a war room that ran for a couple of years — a multidisciplinary standing team drawn from technology, product, marketing and sales, meeting weekly against explicit deliverables.

"I remember when we heard the news. I gathered my team and we started a war room that lasted for a couple of years. We met once a week with deliverables that were very clear. The plan was to beat Amazon.”

Stelleo Tolda

He pushes back on the popular version of the story — that MercadoLibre only survived because Amazon forced it to integrate vertically. The vertical integration came first. Fixed price replaced auctions before eBay made the same move. MercadoPago shipped in 2004. The full technology rebuild landed in 2010. By 2012 it was a different company, and the war room was about deploying an advantage it already had.

On why Amazon was slow, he declines to speculate about their internal reasoning and offers a structural answer instead: Brazil was small relative to Europe and Asia, which were the priorities. Latin America is still quote-unquote small to most global players — and that is precisely what made it winnable. Brazil was a must-win market for MercadoLibre and a rounding error for Amazon, and the two companies allocated talent accordingly.

What was the year-one playbook in Brazil?

Traction, friction and granularity. Remove whatever stops people transacting; run teams by category; and manage supply and demand at the most granular level the data allows. It wasn't big data — it was 1999 — but they read it obsessively, category by category, asking whether there were enough sellers for the buyers or too few buyers for the sellers, then acting on the answer that same day: marketing to pull buyers in, business development to recruit sellers.

His four principles for executing in a complex market: clarity of strategy, access to the information needed to decide, a bias to action he describes as a certain aggressiveness, and a genuine willingness to take risks — which he says became a formal cultural principle. Plus the team, which he keeps returning to.

Why give up being asset-light?

Because asset-light was a story the market loved and the customer didn't. As a public company MercadoLibre marketed itself on requiring little capital — connecting people over bits and bytes while somebody else did the dirty work of moving boxes, off-platform and out of their control. The problem was that the last mile was the experience, and they didn't own it.

"We decided to kill a sacred cow.”

Stelleo Tolda, on abandoning the asset-light model

The trigger point was 2017 and Mexico — investing aggressively in free shipping as a way to double down before, as he puts it, their friends from the north came south. Free shipping drove volume; the capex built the network that could carry it. Warehouses bought or leased, equipment installed, trucks leased, and eventually planes.

The starting position was almost comically far back: "We knew nothing. Until 2013 or 2014, we had never seen a package." So they planned first, visited companies around the world to learn, and — for the first time in the company's history — hired consultants, who Tolda says were genuinely useful in modelling densities and deciding where facilities should sit. Then they hired operators from the market for skills they didn't have.

How do you build a delivery network from zero?

You start with the part that isn't warehousing. The first thing MercadoLibre launched was collections from its largest sellers — routes that drove past sellers big enough to justify a pickup, brought product to a warehouse, and shipped out from there without ever storing it. Cross-docking, in the industry's term.

And the unit economics only work at density, which means the early version has to lose money on purpose. A new route might carry twenty parcels, which is uneconomic. Inject volume and it becomes two hundred, at which point the cost per parcel collapses and the route turns profitable. Repeat that n times and you have a network that is cheap, fast and very hard to copy.

Underneath all of it was the product and technology team, which Tolda calls the spinal cord of the whole effort. They evaluated the warehouse management systems on the market, didn't like any of them, and built their own.

How did Wall Street react to the margins collapsing?

This is the passage most worth reading if you cover MercadoLibre as a listed equity, and it's a useful case for any company contemplating the same trade. Before the logistics investment, revenue was growing around 10% against EBITDA margins of 30–40%. After it, growth ran at 30–40% and EBITDA went to zero or negative.

The reaction split. Some investors concluded the model had never been as profitable as it looked. Others, in Tolda's telling, said finally — they had been staring at a company under-investing against its own opportunity, and were happy to trade margin for growth. Some shareholders left; others arrived. It was understood over several quarters rather than immediately.

He attributes the tolerance to reputation built deliberately over years: MercadoLibre never gave guidance, but was unusually open with analysts and investors about its plans, showed up at conferences, hosted people in Buenos Aires and São Paulo, and — critically — had a track record of executing what it said it would. And they flagged the shift to the market in advance rather than letting it show up in a print.

What gets centralised and what gets localised across 18 markets?

An early decision that turned out to matter: they put all their technology infrastructure in the United States. Locally hosted servers would have been fastest for one market and terrible for every other; a central location in the US meant some latency everywhere and coordination in one place. Development, too, ran almost entirely out of a single building in Buenos Aires.

That eventually broke on local reality — Pix in Brazil is meaningless in other markets and needs a team in the country that can build against it. Tolda describes the resulting arrangement as a positive tension rather than a formula, and says the product looks broadly the same everywhere. What changes is sequencing. They used to test in small markets before launching in large ones; now they lead with the large markets for impact. Credit cards went Brazil first, as the largest and most financially stable, then Mexico, and only recently Argentina — because the market is finally ready.

Why doesn't MercadoLibre buy companies?

It does, occasionally — it bought the three competitors it had out-priced in 2001 — but as a rule it doesn't. In twenty-seven years it has made very few acquisitions.

“We build versus buy. We love to build. We don't like to buy very much.“

Stelleo Tolda

Olga puts to him that several parts of the ecosystem — logistics most obviously — could be standalone businesses. He confirms it's a live debate inside the company, and gives an answer that is more revealing than a yes or no:

"We're so much better than everyone else that we want to keep it for ourselves. For now.”

Stelleo Tolda, on spinning out logistics

What is AI actually doing inside MercadoLibre?

Tolda's framing is comparative rather than breathless: it feels like the early commercial internet, including the parts that ended badly. Effervescence, excitement, some hubris, valuations that are extremely high, and a great deal of infrastructure being built — much of which, he notes, took several years to be used to capacity last time, with a blip in the middle.

Underneath the cycle, the tools are simply better. Credit models, fraud prevention and personalisation were already machine learning problems; what has changed is speed, accuracy and interface. They have shipped a conversational agent inside MercadoPago in Brazil that can pay a boleto, surface what's due in the next fifteen days, or flag a better home for your money. The next step he describes is the interesting one — moving from responding to prompts to initiating them, doing the work in the background and presenting the decision.

The efficiency effect is already visible in headcount policy. Customer experience was capped some years ago even as the business kept growing, because a rising share of cases now resolve with zero response time and no human contact. Technology hiring is capped too.

On what the proprietary data buys: credit. MercadoLibre uses bureaus and public sources like everyone else, but the transactional history that decides whether someone merits a thousand reais or five thousand is data nobody outside the company can see. He calls it a benefit of scale, a benefit of big data, and a benefit of having a team that can actually use it.

How do you keep a culture at 100,000 people?

You write it down late, and you don't write it yourself. For most of the company's life the culture was transmitted by example — what Galperin and Kazah were doing in the Buenos Aires garage, what Tolda was doing in a small São Paulo office. Twelve or thirteen years ago, watching the company grow and hire from other cultures, they decided to define it.

The method is the part worth copying. They hired a consultant, then assembled a group of leaders who were deliberately not the founders and not the C-level, and had them draft the proposal. The C-level reviewed and sent it back a couple of times. What came out was recognisable as what already existed in practice, which is why it held.

The animating metaphor is a high-performance team — trust, common purpose, shared values, an understanding of what each person brings.

"We're Latin Americans, we love soccer. It's like we're playing the Champions League now. We started out playing locally, we grew into the first division of our countries, and it's been many years since we started playing against the best in the world. We're competitive against the best in the world.”

Stelleo Tolda

The other principle he names is permanent incompleteness: "We have this idea that we're always in continuous beta. Our company is in beta version. We're never ready, because we're always improving."

How do you hand over a company you spent 23 years building?

Slowly, and with the succession designed years before it happens. Tolda stepped out of the day-to-day in 2022, handing his role to Ariel Szarfsztejn, who had led the logistics build. Marcos Galperin made the same transition, with the CEO succession taking formal effect in January 2026. Both remain on the board.

What they optimised for in choosing a successor is notable: alongside being an excellent operator, Szarfsztejn was chosen substantially for being a cultural ambassador — someone who embodied what the written culture described. That was the point of writing it down.

Tolda is candid about the personal half, which is the part most founders don't discuss. Identity gets wrapped up in the job, and leaving abruptly can be very hard. It wasn't, for him, because he was ready — and because he could see people younger than the founding team with real talent wanting to occupy the space and prove themselves. That, he says, was the signal that the company might actually become perpetual, which was the original ambition: build something world-class that outlives its founders.

What makes a board useful rather than a drag?

Tolda has now seen governance from three sides — operator answerable to a private board, operator answerable to a public board, and independent director. His route to the third was three years as an advisor first, which served both as a handover mechanism and as the cooling-off period the SEC requires before a former executive can be independent. During it he mentored Szarfsztejn and much of the leadership, until they needed him less, which he describes as the desired outcome.

His test for a good board member: they challenge you, and they do it from a caring position — they genuinely want you to be better. Some skin in the game, but not too much, because large investors can carry incentives that diverge from founders and executives, particularly at the private stage. He treats aligning those incentives as work both sides owe each other.

And the failure mode he names is board members who operate — a mistake he says happens in private and smaller companies but is not confined to them.

One unexpected benefit of leaving: "During 23 years I was so immersed in MercadoLibre that you have no time to see what's happening in the outside world." Boards, investing and mentoring have given him a perspective he says he simply couldn't have had inside.

Rapid fire: six questions

The hiring method he trusts most. Cases — but not consulting cases. He uses them to observe behaviour and cultural fit, asking how a candidate would act under genuine uncertainty. And he tells them plainly that if they aren't comfortable with uncertainty, MercadoLibre isn't the right place, and that this is fine.

The biggest waste of time in most companies. Meetings held to convey information. Read the deck.

What he learned outside that he couldn't have learned inside. That there is life beyond it, that he can be interesting, and that learning happens elsewhere — after 23 years in which nearly all of it happened in one environment.

The startup wisdom he thinks is wrong. The worship of speed at the expense of quality. Quality trumps speed.

He then complicates his own answer, which is the better part. eBay once told MercadoLibre it wasn't growing fast enough, and he now thinks that advice was right for the moment: their customer acquisition cost was extraordinarily low, so investing to accelerate was correct — but having raised a second round with no more behind it, they were fixated on reaching profitability quickly. Deferring profit to become a much bigger company, and profitable at a much larger scale, was the right call.

Overhyped or underhyped in Latin America. AI — overhyped globally, arguably underhyped in the region, because Latin American founders are used to building with fewer resources.

"The worst thing for an entrepreneur is abundance. The best thing is scarcity.”

Stelleo Tolda

Starting again in 1999, knowing what he knows. "I wouldn't have wasted money on a TV commercial."

COMPANIES AND PRODUCTS

⠀PEOPLE

⠀IDEAS AND FRAMEWORKS

Co-founder of MercadoLibre, which he started in 1999 with Marcos Galperin and Hernán Kazah after Stanford. Ran Brazil from the beginning, and served as Chief Operating Officer until stepping out of the day-to-day in 2022 after 23 years. Now an independent director on MercadoLibre's board, having served three years as an advisor first, and active across the regional ecosystem as an investor and mentor including through Endeavor.

Hernán Kazah, Kaszek — building Latin America's $100B+ companies · Nigel Morris, QED — Capital One, Nubank and what actually changed in Latin America · Brian Requarth, Latitud — why ~50% of LatAm founders now go global

🎙 The J Curve  is where LATAM's boldest founders & investors come to talk real strategy, opportunity and leadership.