David Vélez - Building Nubank and the Future of Global Banking - [Invest Like the Best, EP.495]

Summary

• Vélez’s core thesis is that fintech disruption is still in its first inning: global consumer financial services is a ~$10 trillion industry, and something like 90% still goes through branches and is owned by incumbent banks. His analogy: it’s “as if ninety percent of the world were going to Blockbuster to get VHS tapes and DVDs” while Netflix-like services held about 10%. Nubank is the proof case — close to 120M customers in Brazil, about 65% of the adult population, and the primary bank account for about 60% of its users, having fully surpassed the biggest incumbent.
• The cost math is the moat: Nubank serves 140M customers with 10,000 employees versus competitors’ 100M customers with 100,000 employees and 6,000 branches — a cost to serve of about 3% of a traditional incumbent’s. In the US, $15T of retail deposits earning about 40bps in a 4.5–5% rate world means roughly $600B a year going to banks to pay for branches rather than staying in consumers’ pockets — a profit pool Vélez says competition will eventually erode.
• On moats in the AI era, Vélez argues scaled neobanks sit in the “sweet spot”: startups have speed, but “you still can’t really get ChatGPT to get you a banking license.” Nubank’s 20,000–30,000 data points per customer across 13 years fed a proprietary foundation model for financial services and credit, developed after its Hyperplane acquisition. He says it supports fraud decisions in Mexico and Colombia and can help Nubank enter the US and other markets with its own model from day one.
• Nubank missed the Opus 4.6 moment in December ‘25 by 60 days — earlier models were “really awful” at its Clojure code — then went “code red,” mapping and reinventing 120 workflows across the company with eval infrastructure at every step. Vélez went from one direct report to 20 and the org from 10 layers to five; he calls it “a bolt of adrenaline” and says much of the productivity gain “is not even because of AI” — AI was a catalyst to re-examine everything.
• The big exogenous calls: cash goes away as payments digitize, open banking reduces data inertia, and of today’s roughly 190 currencies “maybe there will be 10” in 10–20 years, with billions banking offshore in US stablecoins, Bitcoin, and other assets. Vélez says US stablecoins could be “one of the biggest saviors” of the dollar’s reserve-currency role — the US fiscal picture is deteriorating but still better than that of 100-plus other countries, and demand for a Brazilian stablecoin is very low.
• Nubank is launching in the US and announced a stablecoin-backed account in upwards of 35 countries — its 2013 seed deck was always a global thesis, not a Latin American one. The metrics Vélez would give buy-side analysts: customer growth, NPS, cost to acquire, LTV/CAC, and ROE — Nubank charges no fees yet earns 2–3x incumbent ROE. Since some US incumbents out-earn emerging-market banks, “your margin is my opportunity.” The hardest US challenge: building a brand organically against entrenched banking brands.
• The personal arc: after a year or two asking “what else could I be doing,” Vélez concluded nothing has more impact than taking Nubank from three countries to 100 and from 100M customers toward a billion — “I’m more of a wartime general than a peacetime general.” His entrepreneurship lesson: quickly ignore imposter syndrome, because “nobody in the world cares as much about this company as I do,” and that visceral care — alongside conviction in the mission — can make the founder the right leader for a very long time.

Deep dive

1. 90% of a $10 trillion industry still runs through branches — the Blockbuster era isn’t over

• Vélez’s opening frame, answering Patrick’s “are we in the late innings?” challenge: all neobanks and fintechs combined are less than 10% of global consumer financial services, an industry of about $10 trillion; something like 90% still goes through branches and is owned by incumbent banks. “It’s still almost as if 90% of the world were going to Blockbuster to get VHS tapes and DVDs” — the Netflix/Amazon/Uber migration from incumbents to digitally native firms is “very, very early in financial services.”
• Nubank’s uniqueness as evidence: “the only neobank in the world where we’ve fully surpassed the biggest incumbent bank” — close to 120M Brazilian customers (about 65% of adults), with about 60% using it as their primary account. “We’re not a secondary wallet… We replaced the incumbent bank.”
• The growth has two legs: market-share migration from incumbents, plus market-size expansion as billions of people enter the banking system — including 3 billion who lack access to credit. Vélez calls the potential impact on GDP an underappreciated “insane boost.”

2. Three problems — access, cost, intelligence — and the $600 billion branch tax

• On access, Vélez says credit-card penetration is 13% in countries such as Mexico and Colombia. His forecast: 10 or 20 years from now, 100% of the population will have access to banking and credit, and 100% of small businesses will have access to credit — a potentially enormous boost to GDP.
• The cost argument carries the episode’s sharpest numbers: 140M customers served by 10,000 employees versus competitors at 100M customers, 100,000 employees, and 6,000 branches — cost to serve about 3% of an incumbent’s, with efficiency returned through no fees, better experience, and better rates. “That’s pre-AI” — AI multiplies the efficiency again: “You could imagine banks that have a billion customers with 5,000 employees.”
• The hidden US tax: $15T of retail deposits paying about 40bps when rates are 4.5–5% means roughly $600B a year funding branches instead of staying with consumers — “a massive wealth tax.” Remittances on $100B of US outflows cost 5–10% per transaction and, he says, “will trend towards zero in a more competitive environment.” He also points to monthly US account fees of $5, $10, or $15.
• Intelligence is the least understood bucket: the top 1% get a private banker optimizing everything; the lower 99% get little or no such help, and financial education “never worked. People don’t sit down to take classes.” AI could give people “superintelligence” to optimize their financial lives, competing away profit pools that exist today.
• The live example: about 50M Brazilian customers get daily nudges — “your bank account or credit card at that other bank is due tomorrow… they’re going to charge you a 15% interest rate per day. Let us give you some money.” Intelligence plus open banking — for example, Plaid connecting 120M accounts — helps customers spot fees and optimize their finances. Vélez says physical space can still matter for some small-business banking, large investments, mortgages, or sophisticated products, but estimates that may be about 5% of cases; most mass-market savings accounts, credit cards, and loans don’t need it.

3. How you’d attack Nubank — and why the data moat now favors the new incumbent

• Patrick’s thought experiment: strip everything away and build a company to kill Nubank. Vélez’s answer — speed is a startup’s edge, especially “if you’re born unregulated,” but in banking “you still can’t really get ChatGPT to get you a banking license” or compress a licensing cycle of two or three years to six months.
• The data moat: 20,000–30,000 data points per customer across 13 years, across 140M customers. Vélez says Nubank built its own foundation model for financial services and credit after acquiring Hyperplane in Palo Alto, which had been struggling to sell to incumbent banks with two-year sales cycles. He says the model now supports fraud decisions in Mexico and Colombia; Nubank can also enter the US and other international markets with its model and use it to make basic decisions from day one.
• His positioning claim: companies like Nubank are “in the sweet spot” — with scale, data, brand, and licenses that startups would struggle to replicate, plus technology infrastructure that lets them benefit from AI.

4. Scale alone isn’t a moat — Blockbuster had data too

• Pushing back on his own scale argument: “Blockbuster had a huge amount of data… Did that data help them build a better search algorithm? No, because they weren’t capturing the data.” Netflix came in with no data and won through systems and culture — “Scale and data are important, but your systems and culture are equally important.”
• The incumbent contrast: fifteen different systems from decades of acquisitions, AI projects aimed at transforming COBOL and getting away from mainframes, and 10- or 15-year cloud histories still in progress. Nubank saw itself from day one as “a technology company that happens to be in financial services.”
• What the foundation model changed tactically: before AI, adding a new feature to a model could take six months, while testing a new feature took several weeks. With the foundation model, Nubank can test approaches much faster. Vélez says AI can enable decisions across the organization to be automated and made by a model rather than a human.

5. End of cash, open banking, and 190 currencies becoming 10

• Three exogenous shifts he’s watching: cash going away as payments digitize through real-time systems such as Pix and UPI; open banking letting consumers share their data history and potentially monetize it by choice; and crypto and stablecoins enabling a parallel offshore banking system. Vélez says open banking will reduce data inertia and make banking less concentrated, with lower fees and costs.
• The boldest call: “Today, there are about 190 currencies around the world. My view is that 10 or 20 years from now, maybe there will be 10,” because he thinks the fiscal situation in many countries is dire and some fiat money is “basically crypto tokens.”
• On dollar reserve status: US stablecoins may be “one of the biggest saviors” of the dollar — the US fiscal picture is “not great and is deteriorating fast,” but, in his view, still better than that of 100-plus other countries. He expects the dollar to retain reserve-currency power “for a very long time, but it won’t be unique”: he anticipates five or six other fiat currencies people save in, alongside crypto assets. He’s “always been a Bitcoin fan” and sees it as part of a diversified investment portfolio.

6. The taco stand and the $20 credit line — digitalization creates creditworthiness

• The least-understood consequence of cashless economies: a taco stand in Mexico City selling for cash has real daily cash flow “nobody can underwrite on top of.” Once transactions hit a digital ledger, “that taco seller can take that information to a bank and say, ‘Hey, look at my cash flow… I can now have two taco stands, or I can have five.’” With about 70% of Brazilian employment still in small businesses, Vélez sees “trillions of dollars of credit being enabled for billions of people.”
• And yes, creditworthiness can be created, not just measured: Nubank was the first credit card for 30 million people in Latin America. Banks claimed “people don’t want credit”; Vélez says the gap was a product that matched their needs. A $20 starting credit line — “when banks saw that product, they thought it was absurd” — became $100, then $1,000, with an AI copilot coaching customers on spending, investing, and saving to improve their financial picture and creditworthiness.
• On cross-country portability: more exportable than expected. Mexico’s cash economy was the hard part — “we’re a shark in digital waters… in the offline world with banking branches, we’re much less competitive” — but Nubank is now Mexico’s second-largest bank and largest digital bank. As countries digitalize, “the future looks much more homogeneous than the past, even for financial services.”

7. The US launch: a global thesis finally executed, with ROE as the tell

• The announcement: launching in the US plus a stablecoin-backed account in upwards of 35 countries. The 2013 seed deck had “nothing specific about Brazil or Latin America” — the thesis was always global retail banking transformed by digitally native firms. Latin America was the proof: largest private financial institution on the continent, highest NPS, and highest ROE in a country like Brazil. The hardest US challenge, in his words: building an organic brand “in a way where we obviously have a lot of financial resources” but more smartly and capital-efficiently than simply spending heavily on marketing.
• His scorecard for buy-side analysts: customer growth, NPS, cost to acquire, LTV/CAC, then ROE — Nubank charges “literally no fees,” has less revenue, yet earns 2–3x incumbent ROE, which he says demonstrates a cost-structure advantage and higher-quality, lower-cost product in a commoditized market.
• The underappreciated US setup: some US incumbent banks now have higher returns on equity and are more profitable than the emerging-market banks Nubank competes with. “That means your margin is my opportunity” — the classic Amazon-style read on incumbent ROE despite the costs of providing the product.
• On Mickey Malka’s “fintech is dead”: Vélez asked him directly last week. Malka’s answer, as Vélez recounts it, was that he was doubling down on larger neobanks riding the “sweet spot” between AI transformation and scale moats. For new entrants, Vélez expects less-obvious applications on top of AI models; building a consumer brand with hundreds of millions of customers will be harder unless the product is fundamentally different.

8. Missing Opus 4.6 by 60 days, then code red: 120 workflows, 5 layers, 20 direct reports

• The confession: Nubank’s earlier programming models were “really awful” at Clojure through 2025 — “our most experienced engineer said, ‘This is not gonna work.’” Then Opus 4.6 arrived in December ‘25. Nubank missed the switch, catching it only 60 days later after meetings with early-stage founders in California: “Holy shit, this is actually working out. It can do Clojure very, very well.”
• The response was total: a company-wide code red, 120 workflows mapped and reinvented natively using AI tools — product design, KYC, internal audit, compliance, and regulatory capital management — with eval infrastructure measuring each step. His caveat: much of the gain “is not even because of AI… there is always an opportunity to actively reinvent the way you do things”; AI augmented that opportunity.
• Structurally, Vélez went from one direct report to 20, and the organization from 10 layers to five. Echoing Patrick’s reference to Tobi Lütke and the “dictator’s dilemma,” he credits a proactive “company brain” and a culture of going directly to people doing the work: “I’ll go to whoever has access to the best information.”
• The underdiscussed frontier is HR: semiannual manager feedback with 30-day recency bias is “the type of feedback you would give an Olympic athlete. That would be crazy.” Nubank is building daily AI coaching — pull-request feedback for engineers and instrumentation giving managers real-time visibility instead of relying only on “very low-information-density” one-on-ones.

9. Agents as a customer segment, and why customer obsession is the anti-AI-bro edge

• On agentic commerce: Vélez says agents become a customer segment as important as consumers. Nubank has a dedicated team treating agents like its under-18 or small-business segments, including SEO for agents. But in banking, trust stays human: “the human will want to know where their life savings are deposited. You’ve got to do both.”
• His influences: Amazon for consumer obsession, Disney for magic — “for something as boring as sending money, how do you create a magical experience out of that?” — down to the US welcome kit. His contrarian read on the moment: in the AI debate, “you don’t hear that much conversation about customers”; people are focused on efficiency, while “the AI bro, the tech bro, is much more enamored today with the technology versus the consumer experience itself.”
• A counterintuitive reversal since the earlier conversation: from remote to hybrid and significantly more time in the office, because AI makes the case for in-person collaboration stronger. “The individual at home will be competing with AIs, and we’ll lose that battle.”

10. Founder versus CEO, imposter syndrome, and the bear/bull spectrum

• After one or two years asking what else he should do, Vélez concluded nothing beats taking Nubank “from three countries to 100 countries, or from 100 million customers to a billion.” His self-diagnosis: “I’m more of a wartime general than a peacetime general” — disruption energizes him. On the founder/CEO split: the founder is impatient and an individual contributor at heart; the CEO builds systems and patience. Vélez says you can redesign the roles, but “if you have to choose, you choose more of the founder than the CEO,” especially amid big technology shifts.
• The hardest part of the long haul was imposter syndrome — “are you the right driver of the car as it becomes more complex and goes faster and faster?” He hired experienced external managers, then “quickly realized that actually wasn’t what the organization needed” — it needed more of his clear leadership, motivation, and willingness to dream big and push. “Yes, I’m pushing enough… the organization will push back if needed.”
• On leadership, Vélez emphasizes having a differentiated vision, inspiring people to follow it, and listening enough to recognize when you’re wrong. Good leaders navigate between rejecting all consensus and being consumed by it: understand which parts of the status quo persist for a reason, and push to change the parts that should change. He describes that balance as requiring conviction, active listening, and humility.
• Most emotional moments: passing Brazilian incumbent banks after a decade of local experts saying Nubank had no shot, and launching in Colombia — the country Vélez left at eight for Costa Rica because of the security situation — at an event with the president and 100,000 people connected to the videocast.
• The bear case: saturating the current three-country footprint, “very soon we’re going to run out of Brazilians,” and failing to internationalize or take advantage of AI. The bull case: riding real-time payments, open finance, and AI as market share moves from incumbents to digital-native companies over the next two or three decades; scaling through a multi-country platform toward 100-plus countries. Nubank’s pre-AI cost structure, he says, is 50 times better than incumbents’, enabling higher yields, lower fees, a better experience, digital services, and advice integrated into products.
• His closing entrepreneurship lesson: quickly ignore imposter syndrome — visceral care for the company can make the founder the right leader for a long time. A mission you truly care about makes the hard work easier to sustain. He says AI makes mission-finding harder because launching a company is easier, and that some of the best founders are “going for the impossible, truly”: hardware, space, and problems they come to believe only they can solve.