Code Is Getting Cheaper, Licenses Are Getting More Valuable: The Real Moat of Fintech in the AI Era

Author: Matt Brown

Translation: Deep Tide TechFlow

Deep Tide Introduction: Matrix VC Partner Matt Brown presents a counterintuitive argument: AI makes coding cheaper, but it also makes the truly inimitable assets in Fintech—bank licenses, underwriting data for credit loss accumulation, and risk control models fed by real transaction volumes—more valuable than ever before.

“You can’t program your way to a bank license,” captures the core of this article.

This is not just an analysis of Fintech; it’s a map of “what moats are more solid” in the AI era.

The full text is as follows:

The term “Fintech” has long relied on the ambiguity in its name for arbitrage.

“fin” means “coming from.” A large volume of emails from .gov domains, months-long audits, compliance officers more familiar with your SAR filing history than you are, and business trips to Charlotte or Washington—these are the “fin” side. “tech” refers to a sleek mobile app, tenfold user experience, and chatting over coffee at Blue Bottle about investments.

“fin” and “tech” are always part of the same spectrum, but the market typically rewards those Fintech companies that resemble “tech” as much as possible and minimize “fin.”

This is easy to understand. In 2021, the gross profit pool for software was about $0.7 trillion, enjoying a high premium. The gross profit pool for financial services is an order of magnitude larger, but valuations are much more conservative. Fintech allows arbitrage on both ends: the economics of financial services combined with the valuation multiples of software companies.

This profit gap also reveals where the real money is. Financial services generate the most gross profit across industries worldwide. On the “fin” side of Fintech, not only is it more defensive, but it’s also a much larger market.

Then AI arrived, and the arbitrage space disappeared. As investors reprice “how much software is worth in a world where code gets cheaper and cheaper,” software valuations are compressed. The wall that kept competitors out has lowered, allowing more players in and eroding pricing power.

If your business is software, this is a real problem. But Fintech’s expenses are not engineering costs. Follow the money, and the difference quickly becomes clear.

PayPal spends 9% of revenue on R&D, Block spends 12%. This isn’t because Fintech engineering isn’t important—Stripe’s engineering capabilities are world-class and a real competitive advantage. But most of the money doesn’t flow into engineering.

The money flows into “fin.” Unlike R&D expenses, these costs are not just about producing a product—they build moats:

Credit loss costs buy underwriting data

Before paying an engineer, Affirm spends 35% of revenue on credit losses and capital costs. Each bad debt loss is a repayment data point that competitors cannot access. A new entrant training models on synthetic data has no real benchmark. Synthetic data alone cannot establish a reliable loss history.

Compliance expenses buy regulatory licenses

Wise operates within over 65 regulatory licenses, dedicating one-third of its staff to compliance and anti-financial crime measures. Licenses across 50 states, BSA/AML compliance programs, bank charter requirements—these are not built advantages but licenses continuously earned. You can’t program your way to a bank license.

Transaction volume buys proprietary data

Toast’s payment segment has a gross profit margin of about 22%, much lower than its 70% SaaS segment, but the gross profit generated is nearly twice as much. These costs produce merchant-level transaction data, which feeds back into Toast Capital, having issued over $1 billion in loans. Adyen’s risk models are trained on transaction patterns across more than 30 markets.

Fintech’s profit margins have never been high, and that’s the key

Payment companies’ gross margins range from 20% to 50%, not 80%. But low margins do not mean weak business. The low margins in Fintech are because many costs generate compound advantages. Even costs that do not create advantages are outside the scope of AI-driven cost compression.

AI makes each of these moats stronger. Better models lower loss rates, improved fraud detection reduces chargebacks, and better compliance tools enable smaller teams to hold more licenses. AI will not replace moats; it rewards companies that build in the hardest areas of Fintech: cash flow, risk-taking, proprietary data, and regulatory licenses.

So the real argument is not just “AI helps Fintech,” but that AI shifts value from surface-level products to proprietary data, risk capacity, regulatory licenses, and distribution channels embedded with real cash flow. Building in these areas allows AI to compound your advantage. If your differentiation is in code, AI works against you.

Demand side is also growing continuously. Every atmosphere-programmed checkout process is a new fraud vector, and every autonomous trading AI agent is a chargeback risk. The more infrastructure built on Fintech foundations, the more indispensable that infrastructure becomes.

“Fin” is the true winner

This realization has already begun to make smart Fintech founders rethink their position on the “fin” and “tech” spectrum:

Are we taking on and pricing risk ourselves, or passing it to partners and letting them profit?

Do we have regulatory relationships, or are we renting them from those who do?

Is each transaction making our risk models more accurate, or training others’ models?

Is our ledger a source of real data, or an incomplete mirror of others’ ledgers?

This distinction divides the Fintech landscape into two camps. Companies with regulatory relationships, who bear and accumulate credit losses, and who build transaction data moats, are creating moats that AI will deepen. Those renting “fin”—using partner banks’ licenses, BaaS providers’ ledgers, others’ risk models with better interfaces—face the same problems as SaaS companies. Their differentiation is in code, and code has just become cheaper.

Applying old arbitrage based on software valuation multiples to financial service economics is dead. The new arbitrage is simpler: owning “fin.”