Paulo Passoni on what he would do with $500 million, and why an AI margin is not a SaaS margin

The J Curve is an English-language podcast and newsletter covering Latin America's technology and venture capital ecosystem, hosted by investor Olga Maslikhova. TJC Debrief is its monthly companion, in which Maslikhova and Paulo Passoni of Valor Capital take apart the month's news where Latin America meets global markets. This edition starts from a thought experiment — you have just raised $500 million, unconstrained — and runs through the AI funding boom, what replaces SaaS, why Founders Fund and Greenoaks keep showing up in Brazil, the fifty companies the secondaries market actually cares about, and stablecoins. Underneath all of it, the critical question to ask about AI companies: not what the revenue is, but what the return on the capital is.

What would Paulo Passoni do with $500 million?

Ten years of locked capital, no redemptions, and all of it in technology. He defines that last part by elimination: not industries that are not transforming, likely no railroads, and — the more interesting exclusion — probably not power generation, the obvious adjacent trade, because he expects competition between capital providers to compress returns there to 10 or 15%. Respectable, and beatable elsewhere.

The public–private split is where the reasoning is. Normally he would run near fifty-fifty. Today he would hold 15% in public markets and park the difference in cash, because committing it to private companies means he cannot sell when the entry point arrives. His discomfort is not that multiples are high but that earnings may be — semiconductors and memory in particular, where a shortage may let companies over-earn for three or four years. He will not play it, because he does not know when the music stops and the people who claim they will get out in time mostly do not.

"Every so often there is a growth scare, and a growth scare often rhymes with a higher cost of capital. Growth equity is like a long-duration bond — you change the discount rate a little and you change the present value a lot. What that means is the public markets give you entry points."

Paulo Passoni

The exception he makes for himself is small and deliberate: a slice of Nvidia exposure, perhaps through Ivana Delevska's Spear ETF, which he says is up 45% this year. Not for the return. Zero exposure means you stop paying attention, and by the time you want to load up on risk again you are out of date.

Who is the best investor at knowing when to sell?

One name: Philippe Laffont at Coatue, whom he calls the best in technology at stepping aside before trouble. The example is the second half of 2021 — Laffont sold, took damage, but far less than investors who stayed deep in technology through 2022, and reloaded into the next cycle.

The observation attached to it is the useful part for anyone reading a manager's public positioning for signal. Laffont is not selling right now. He also would not tell you if he were about to.

What is an orchestrator, and why is it Paulo's contrarian bet?

Of the large private companies his favourite is Anthropic, which he credits with dominating enterprise and, from the outside, with a culture that attracts and keeps talent. He can see roughly a trillion becoming two or three, on revenue he thinks supports it. But the position he flags as against consensus sits one layer up: the companies that stand between you and the models.

Three reasons, and the first is an analogy to cloud. Lock-in — companies that moved onto Amazon with no protection became captive and had no leverage when prices rose, until an infrastructure layer emerged to put space between them; everyone is now building directly on Anthropic. Performance — a year ago the expectation was that models would converge and become interchangeable, and instead they specialised, with the best model per subcategory changing dynamically, so blending beats picking. Cost — you do not want a routine customer call running on the most expensive model available.

His examples are Cursor for coding, Perplexity moving industry by industry into finance, healthcare and legal, and Higgsfield — a Valor investment — which combines around thirty video models, where token consumption is high enough that combining is worth even more than elsewhere. He concedes the position may not work: maybe he is a fool and is proven wrong in five years.

Is the AI funding boom a bubble?

Not yet, and his best guess at a correction is roughly three years out, offered explicitly as a guess. He puts the US at the second or third inning of adoption and the rest of the world, Latin America and Europe included, at the first. What will not repeat is how the first quarter's demand arrived — enterprises panicking they would lose ground and spending without guardrails. What he sees now is rationality, and inference costs falling several-fold at the same quality.

Which produces a prediction about narrative rather than fundamentals: optimisation will slow revenue growth at the labs, public markets extrapolate second derivatives, and the deceleration will be read as the beginning of the end. He expects a lot of those headlines in the next six months. What he will not tolerate is the costless version of the call.

"Claiming there will be a bubble next year makes you sound smart, but you also have to say when the bubble is going to burst. If you keep saying bubble every year, every ten years you're going to be right more than once. It's like a lottery. […] Tell me when and why and how. That's more valuable, and nobody can do that."

Paulo Passoni

If it does come, he traces it as a chain rather than an event. Revenue does not arrive at the model companies because users are not seeing the benefit; short revenue means curtailing the physical build-out; curtailed build-out means the people constructing it discover they do not have the years of demand they assumed. And much of the financing on the physical side is leveraged, which cuts both ways.

Why is an AI margin not a SaaS margin?

Because SaaS had no capital expenditure and AI has enormous amounts of it. A 70% gross margin in SaaS converted more or less to cash; in AI, part of that margin is not profit but compensation for the capital deployed. The two numbers are not comparable even when identical.

“It might be that a 50% gross margin at the lab companies only generates an 8% return on invested capital. That's not good enough. Maybe you really need 75 or 80% to generate 20% or more."

Paulo Passoni

And the return on capital is not the finish line either, because value creation is only the spread over your cost of capital. At a blended 12 or 13%, a business reinvesting at the same rate creates nothing. He calls it shareholder value 101 — which it is, and which is why it is striking how rarely it is applied to model companies. A utility mindset rather than a software one, and the second reason he prefers orchestrators: they raise money to build an intelligence layer, not to buy chips.

Who ends up capturing the value?

The customer, as with most technology — and the worked example is Maslikhova's own. She uploaded a set of Instagram carousels that had performed well and got recognisable versions back in under ten minutes, against the five or six hours it takes her designer. Her conclusion is that Canva is in serious trouble, not because the tool is perfect today but because of the rate of improvement.

Paulo generalises it into two stages. Today you compare the marginal cost of intelligence against the cost of labour, which is why it looks like a bargain. Later you compare one intelligence tool against another, and both have to earn a return on capital.

"Then it's who is the lowest-cost producer, quality adjusted, between intelligence tool and intelligence tool. And that's going to be the price of intelligence."

Paulo Passoni

Which leaves one thing worth holding: distribution, and direct contact with the customer. CloudWalk is his example — seven million people served, an expense-optimisation bot climbing the Brazilian app store, a founder who now calls it an AI company that happens to monetise through financial products. And a warning to anyone whose revenue arrives through a lab: being distributed by Anthropic while the customer does not know your name is tremendous at first and dangerous later, because the second-order move is to squeeze you on the customer's behalf.

What replaces SaaS?

A take rate on a transaction, in place of a price per seat or per task. Valor's edge is in financial technology so he concedes the view is congenial, but the advice is portable: attach yourself to the transaction and charge a percentage. In legal technology that is the settlement of a claim; in healthcare, the settlement between patient and insurer, or hospital and insurer. Wherever money moves, sit inside the flow.

On what survives, the dividing line is not quality. Mission-critical systems persist — he uses SAP as the example precisely because nobody enjoys it and nobody rebuilds it — though evolve at zero speed and they die anyway. What worries him is the tier below: important but not mission-critical, a CRM being his example, which is exactly what becomes cheap to replicate. The cost of producing a line of code has collapsed and is heading to zero, so the price of marginal software follows it. What resists is friction — historical data, mission-critical process. Moats, as Maslikhova puts it.

Asked what he would build in Brazil today, he answers with a warning instead of a business. Distribution is the hardest thing to do, partnering your way to it does not work — he points at the business-school literature on joint ventures and their minuscule success rate — and no incumbent hands over distribution for a 30% stake. He is sceptical of anything needing millions of customers, leans B2B, flags data centres in Latin America for inference, and then says the honest thing: the cards are reshuffling and he has more questions than answers.

What is mispriced in Latin America?

The whole growth stage, and the cause is opportunity cost rather than sentiment. Capital worldwide is being pulled into enormous AI rounds, so very little flows into Latin American growth — which is why you can buy at sub-ten times revenue, in some cases companies already profitable and still scaling.

He is careful about the size of the set. At scale and still growing fast, he names three: Plata, CloudWalk, QI Tech. A tier below, growing 30 to 50% a year and building moats rather than doubling: Loft, QuintoAndar. And iFood, assembling an ecosystem across delivery and financial services by organic growth and acquisition.

On the public side MercadoLibre remains his favourite; monetisation is shifting to financial services, which markets kind of don't love as much as operating revenue, thinking it more unstable and subject to cycles. Maslikhova's view is that the stock is underpriced no matter the scale. Nubank he treats differently — he sees competition arriving and says it now trades as a bet on the Brazilian election.

Why is renting a motorcycle a better business than selling one?

Three effects stack, and Mottu — which rents motorcycles by subscription to delivery riders in Brazil — is where he demonstrates them. Cost of capital: an individual at the base of the pyramid with no data history funds themselves at 50 to 100% a year; a company managing the risk at scale funds at mid-to-high teens. Maintenance: a fleet serviced at scale beats an individual paying retail. Insurance: the individual buys through expensive channels, the company self-insures. Put together, cars and motorcycles in these markets should be sold as a service — a market under 10% penetrated, which he sizes at $2 to $3 billion in Brazil.

The part worth stealing is where he puts the moat. The business looks financial, and the financial engineering is the replicable half.

"Even though it has a financial nature, the moat comes from operations. The financial part a lot of people can replicate. The operations, few can."

Paulo Passoni

What he means is unglamorous and specific: Mottu sources parts from Asia and assembles in Brazil to reduce import tariffs, and designs the machine so it is hard to strip and sell into the grey parts market — a problem he says Honda has badly. Making the stolen asset harder to liquefy is the moat.

What are Founders Fund and Greenoaks underwriting in Brazil?

The sheer volume of labour tied up in one category of problem. For Enter, the millions of lawyers processing claims in what he calls the world capital of lawsuits, where large companies routinely carry legal staff of 50 to 100 people before you count the cost of the claims. For Pax, security, where he says losses are a heavy burden on the economy and the path runs through government first — and where he adds, without softening it, that he is not certain how it plays out.

The second half is the people: second-time founders and the best of the best, who are also very good at fundraising, have Silicon Valley networks, and speak the language. Which lets them, in his phrase, almost bypass the local venture ecosystem.

He then raises the uncomfortable question himself, unprompted, about his own market — is there now adverse selection for local funds? — and answers no. Sometimes the best founders are not the best at fundraising, at least early. Oftentimes the best fundraisers cannot deliver. On whether those firms are backing regional champions or global category creators he is blunt: most likely regional, with optionality beyond it, but nobody is underwriting a Brazilian company on the theory that it one day takes the United States.

If they are regional, where is the exit?

This is the bind he identifies and does not resolve. A billion-dollar-plus valuation requires a US listing, which requires not one billion of revenue but a path toward tens of billions. Very few make it. M&A then — but he is clear it is not where success lives, and his precedents are one-offs: Okta buying Auth0 out of Argentina, Visa buying Pismo for a billion. What does travel is the company whose talent is in Latin America and whose customers are global, which he likens to the Israeli model.

Local strategic buyers mostly do not step in, on a high cost of capital and a conservative posture — and two stories shape how they now behave. Itaú bought a piece of XP, which gave XP the credibility to list, after which Itaú sold and, he suspects, concluded it should have bought the whole thing. And Nubank's acquisition of Easynvest, which he thinks the company probably regards as a mistake.

"Once you get control, what's the incentive for the people there? Very little."

Paulo Passoni

On Enter, he thinks whoever invested at the last round does believe Brazil alone reaches two to five billion of revenue. The company's own people put it at $500 million worst case and $5 billion best, with a $2 billion base, and he says early traction supports it. What he praises is not the projection but the behaviour — a management team keeping its head down and not giving interviews. Drink your own Kool-Aid and become an ambassador for your company, he warns, and the probability you execute drops.

What does the US business do to CloudWalk's valuation?

Below roughly $100 million of US revenue, investors price the American business at nothing — five million reads as evidence you will never succeed there. Cross it while still growing fast and the option switches on. He says CloudWalk could cross by the end of this year and puts the potential value of the US business at around $30 billion: not what the market would pay today, but what the option is eventually worth. In Brazil, meanwhile, a reclassification is under way from a company benchmarked against Stone and PagSeguro to an AI company that monetises through financial services.

His description of how that repricing actually happens is the most deflating passage in the episode, and the most accurate.

“Out of fifty investors, maybe twenty are going to double-click. Of those, maybe three or four agree it's interesting. And of those three or four, maybe one ends up investing."

Paulo Passoni

Why does the secondary market only care about 50 companies?

Because a continuation fund only works if the portfolio holds something people want to own — you cannot do one with dogs, and nobody will provide the exit for the prior investors. Maslikhova sets the question up with the market data: secondaries tracking toward $250 billion in 2026, GP-led continuation vehicles already over $115 billion in 2025, a tool once seen as distressed now a mainstream liquidity channel.

He expects the market to grow manyfold because companies keep delaying listings, then narrows it to what decides whether that continues: what happens when SpaceX's lockup expires. If the settled price lands well below what secondary buyers paid, people think twice and companies are pushed to list earlier. If those buyers make money, listings get delayed further and the funds get bigger. All eyes, he says, on the Anthropic, OpenAI and SpaceX listings over the next two years.

"I think for secondary markets, there are only fifty companies in the world that matter."

Paulo Passoni

The work is not underwriting two or five thousand companies. It is knowing which fifty are on the list — it moves — and transacting privately for stakes held inside funds, SPVs and family offices. If you manage a fund holding none of them, the question is whether you can make the case that one is about to join. He thinks around ten Brazilian companies have a shot and two or three might make it over the next decade.

Is the IPO window really open?

Open in the US, and selectively. What holds it open is the performance of the big names — if they falter, the party stops. Maslikhova puts the first quarter's data to him, fewer listings but more dollars raised; his read is one word, concentrating.

On buying listings he is dismissive in a way worth quoting to anyone tempted. Two or three out of ten IPOs are worth more than their offer price three to five years later, which is why plenty of public-market investors do not participate at all. Klarna is his illustration: buy at the offer and you have fewer dollars than you started with — true, he immediately adds, of IPOs in every cycle.

"IPO is a poor batting average game. […] If you equal-weight and put a hundred dollars in ten IPOs, you're probably going to lose money.”

Paulo Passoni

He is honest that he used to enjoy the game, because he rated his ability to pick the two out of ten — and that most participants are not investing at all but flipping. Renters, not owners.

What would Paulo hold for ten years?

SpaceX, but not yet — he expects a better entry once the supply of stock clears. Google, still, on data and moat. ByteDance, which he calls cheap, with a great deal of data, TikTok's strength in emerging markets and Seedance among the best video models — where the risk is not the business but whether the Chinese government lets you exit, which is precisely why it is cheap. PicPay, trading now at three or four times 2027 earnings and repricing to ten times the same number, with the Brazilian election as the acknowledged risk.

One clean pass: Cerebras, declined on the grounds that he does not know the company well enough to have a view either way, which is a more useful answer than most. Klarna he does answer, and calls tricky — buy-now-pay-later in the US is fiercely competitive, the European position better, the profit and loss structure not pretty. He credits Sebastian Siemiatkowski as an interesting founder who will have to dig into his conviction and keep building.

Why is Tether the only company that matters in crypto?

Because it is the only one operating at scale, and the reason has nothing to do with technology. He had just come from a Zurich conference where the world's regulators met the crypto industry, and says Tether gave the most interesting presentation there: over 650 million users, highly profitable, an instrument that has quietly exported the dollar. He offers one statistic he does not source — that last year Tether bought half of every treasury bond China sold — and his gloss on it is not celebratory. Almost too good for its own good.

Then the arithmetic, which is the most striking business model in the episode. Roughly $200 billion of assets at 4% is about $8 billion of income. No taxes, because it is offshore. It drops to the bottom line. Some years they beat it and some miss, because around 30% of reserves sit in assets they consider safe and others might not. If $200 billion becomes a trillion, the arithmetic multiplies with it.

"It's a massive cash machine. With 150 people."

Paulo Passoni

The catch he names himself: the company does not need money and does not trade, so nobody outside can answer what it is really worth or what it is doing with the cash.

Can a consortium stablecoin work?

He will not call it either way, and the reason is governance rather than technology: consortium solutions have the potential to work and to fail badly, because the members never agree on what should happen. The coin is Open USD, backed by Visa, Mastercard, Coinbase, BlackRock, Shopify, BBVA and a long list of others. He thinks it threatens Circle more than Tether.

The market underneath it is concentrated — Tether about half, Circle about 30%, everything else currently unimportant — and roughly $300 billion in size against an expectation of $2 to $5 trillion within five years, which implies about 50% compounding. Growth in the first half ran at 30%, and he says plainly that he does not know why. His explanation for the shortfall is that the killer application is still just one thing: remittances, where the cost falls from around 90 basis points to around 10.

His base case — she offers him bull and bear, and he answers with neither — is that more than half of global B2B payments settle on stablecoin rails within five years. The reasons reduce to two words: economics, and 24/7.

Bank rails or stablecoin rails?

Both, chosen per function rather than per ideology. Banking an immigrant in the US who sends money home regularly: stablecoins, obviously. But the flow is not the business — the unit economics of moving money are thin, and serving it is customer acquisition. The monetisation is lending to that person, whose income is seasonal, in the month you can see they are short. And that loan needs traditional rails, because the law requires a licence or a bank-as-a-service partner. Do it offshore in crypto and you have no legal recourse if they default.

What did Paulo learn from selling Circle?

Two lessons, and the first arrives through a hiring exercise that has aged into something more interesting than he intended. Recruiting for the US growth side, he set roughly fifty applicants a case study: Valor owns a position in Circle at $100 a share — hold or sell? The overwhelming majority said hold. The part to keep is why: because that is what Anthropic told them.

Valor sold, having made their return, having watched the stock trade at $60 and at $300, and unable to identify their own edge. He thought it looked expensive on earnings but carried strategic value, in that someone might acquire it — which is where the second lesson comes from.

"You never buy a stock hoping somebody buys it from you strategically. M&A has a 20 or 30% probability."”

Paulo Passoni

What happened afterwards was the consortium launching Open USD — the same coin, as he confirms — challenging Circle's position more directly than Tether's. Though he notes volume begets volume, and previous attempts at similar coins went nowhere.

Lightning round: the miss Paulo keeps

Most overrated Latin American trend. He declines; he does not think there is a clear one right now. One category to short globally. No answer either — he sits with it and says he has to think about it.

Most underrated category to fund in Brazil. Vertical lending, still. Few people like it, because it is risky and demands expertise. Who captures the value in AI. Short term, chips and memory. Long term, whoever owns distribution.

The most under-discussed structural change in venture. Mega funds. The industry was built on the premise that you are a great stock picker — average returns roughly zero, top quartile enormous. Raise tens of billions and you must deploy tens of billions, which drags you toward the average. His preferred alternative is a dedicated vehicle for a company you genuinely have access to and understand, rather than a giant fund in which it is a small position.

A private company he wishes he owned. Maslikhova rules out the obvious answer in advance. He gives it anyway, twice over: Anthropic and OpenAI, both missed.

"We completely did not understand what was happening. We thought the valuations were insane, and the capital required was too high, and we couldn't see a company ramping up as fast as Anthropic did — and then we felt that we missed it. […] We could have tried. We didn't even try."

Paulo Passoni

What he takes from it is a discipline rather than a regret, and the right note to end an episode about mispricing on: what happened is sunk. Work with marginal information, always. Ignore what people say publicly, do the work, analyse forward from the moment you have the facts.

COMPANIES AND FUNDS - GLOBAL

A monthly companion to The J Curve, in which Olga Maslikhova and Paulo take apart the month's news at the intersection of Latin America and global markets. Where the interview series is built around one person's history, Debrief is built around what happened since the last one — and it is where the archive's standing positions get revisited, revised and occasionally reversed on the record.

Co-founder and Managing Partner, QED Investors. Co-founded Capital One with Rich Fairbank, spinning the credit card business out of Signet Bank in 1994 and serving as President and COO until 2004. QED has made more than 200 fintech investments in 20 years, including Nubank, Credit Karma, Remitly, Flywire, AvidXchange, Creditas and Bitso. Also an investor in Swansea City. Previously on the boards of The Economist, Brookings, National Geographic and London Business School.

Paulo — why speed is the last moat, and the CloudWalk–Ramp · Brian Requarth, Latitud — why ~50% of LatAm founders now go global. · Hernán Kazah, Kaszek — the power law and whether LatAm has an exit problem

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