The code is getting cheaper and the license is becoming more valuable: the real moat of Fintech in the AI era Author: Matt Brown Compiled: Deep Tide TechFlow Deep Tide Introduction: Matrix VC Partner Matt Brown put forward a counterintuitive argument: AI makes code cheaper, but makes things that are truly difficult to replicate in Fintech - bank licenses, underwriting data accumulated from credit losses, risk control models fed by real transaction volume - more valuable than before. You can't use atmosphere programming to get a bank license, "this sentence expresses the core of the entire article. This is not just a fintech analysis, but also a map of what makes the moat harder in the AI era. The word 'Fintech' has long relied on ambiguity arbitrage in its name. 'fin' means from. Massive emails from the. gov domain, months long audits, compliance officials who are more familiar with your SAR declaration history than you are, and weekday travel to Charlotte or Washington. Tech "is a sophisticated mobile app, with a 10x user experience, and the ability to chat about investments over coffee at Blue Bottle. Fin "and" tech "have always been on the same spectrum, but the market usually rewards Fintech companies that try to be as" tech "as possible and have as little involvement in" finance "as possible. This is easy to understand. In 2021, the gross profit pool of software was approximately $0.7 trillion, enjoying a high premium. The gross profit margin of financial services is one order of magnitude larger than it, but the valuation is much more conservative. Fintech allows you to arbitrage between two ends: the economics of financial services and the valuation multiples of software companies. The difference in this profit pool also tells you where the real money is. Financial services generate the highest gross profit among all industries worldwide. The 'fin' side of Fintech is not only more defensive, but also a much larger market. Then AI came and the arbitrage space disappeared. As investors reprice how much code is worth in a world where code is getting cheaper, software valuations are compressed. Fintech companies are classified as software companies by the market, and therefore are also affected. But the market got the classification wrong. The cost of fintech, as well as its moat, is never in the code, and in the face of AI driven cost compression, they appear increasingly anti fragile. The story software with two cost structures once had one of the best business models in history: high code production costs, but once written, distribution is almost free. The difference between "expensive to build" and "free to distribute" is the profit margin. If you are a SaaS company and spend 22 to 25% of your revenue on research and development, that expenditure is also a barrier to entry for you. Competitors cannot easily replicate something that took years and millions of dollars to build. AI has compressed this gap from the top. If the code is built and distributed cheaply, the profit margin will narrow. The wall that blocks competitors has lowered, more players have entered, and pricing power has been eroded. If your business is software, this is a real problem. But Fintech expenses are not engineering expenses. Following the money, the difference quickly becomes apparent. PayPal spends 9% of its revenue on research and development, while Block spends 12%. This is not because Fintech engineering is not important - Stripe's engineering capabilities are world-class and a real competitive advantage. But most of the money doesn't flow towards the project. The flow of money is towards' finance '. Unlike research and development expenses, these costs are not just about producing products, they create moats: credit losses buy underwriting data, and before paying an engineer, 35% of income is used for credit losses and funding costs. Every bad debt loss is repayment data that competitors cannot obtain. A new entrant trains a model using synthetic data without a real benchmark. Relying solely on synthetic data cannot establish a reliable loss history. The compliance expenditure purchases regulatory licenses from Wise, which covers over 65 regulatory licenses, and invests one-third of its employees in compliance and financial crime prevention. Remittance licenses in 50 states, BSA/AML compliance programs, and bank charter requirements. These are not the advantages you have built, but the licenses you have continuously earned. You can't use atmosphere programming to get a bank license. The transaction volume bought proprietary data Toast, and the gross profit margin of its payment section is about 22%, far lower than its SaaS section's 70%, but the gross profit generated is almost twice that of the latter. Those costs were exchanged for transaction data at the merchant level, which in turn fed Toast Capital, which has accumulated over $1 billion in loans. Adyen's risk model was trained on trading patterns in over 30 markets. The profit margin of Fintech has never been high, which is why the gross profit margin of key payment companies is between 20% and 50%, rather than 80%. But a low profit margin does not necessarily mean weak business. The low profit margin of fintech is due to the large amount of costs generating compound interest advantages. Even those costs that do not generate advantages are outside the range of AI driven cost compression. AI has made every moat like this stronger. Better models lower loss rates, better fraud detection reduces chargebacks, and better compliance tools enable smaller teams to hold more licenses. AI will not replace moats, it rewards companies that choose to build in the most difficult areas of Fintech: capital flow, risk-taking, proprietary data, and regulation. So the real argument is not just 'AI helps Fintech', but AI transfers value from product surface area to proprietary data, risk-taking ability, regulatory licensing, and distribution channels embedded in real capital flows. If you build in these fields, AI is compounding in your direction. If your differentiation lies in the code, AI will compound in the opposite direction. The demand side is also continuously growing. Every atmosphere programming checkout process is a new fraud vector, and every autonomous trading AI agent is at risk of chargeback. The more things built on top of Fintech infrastructure, the more indispensable this infrastructure itself becomes. The recognition that "Fin" is the winner has begun to force smart Fintech founders to rethink their position in the "fin" and "tech" spectrum: do we bear and price the risks ourselves, or do we pass them on to our partners and let them take the profits? Do we have regulatory relationships or do we rent from people who have regulatory relationships? Is every transaction making our own risk model more accurate or training someone else's model? Is our ledger the source of real data or an incomplete mirror of someone else's ledger? This distinction divides the Fintech landscape into two parts. Companies with regulatory relationships, bearing credit losses themselves, and accumulating transaction data are building a moat that AI will deepen. The companies that rent "fin" - using partner bank licenses, BaaS provider ledgers, and others' risk models with better interfaces - face exactly the same problems as SaaS companies. Their differentiation lies in the code, which has just become cheaper. The old arbitrage of applying software valuation multiples to financial services economics is dead. The new arbitrage is simpler: having 'fin'.