The Infrastructure Is Ready. Are We Using It?
For decades, building sophisticated financial services meant one thing: build everything yourself, or negotiate access with the handful of institutions that already owned the pipes. That era is ending — not with a bang, but with a quiet accumulation of infrastructure making those constraints obsolete.
Want to move money internationally? You needed a correspondent banking relationship and tolerance for a 2-to-5-day settlement window through SWIFT. Want to accept card payments? You were renting access to infrastructure that Visa and Mastercard spent 50 years and billions of dollars constructing. Want to offer clients a data-driven advisory service? You had to build the data layer, the analytics engine, and the delivery mechanism from scratch — before writing a single line of business logic.
SettlementBlockchain Replacing the Correspondent Network
SWIFT is a messaging network. It tells banks what to do; it doesn't actually move money. The settlement itself flows through a chain of correspondent relationships that introduce latency, cost, and opacity at every hop.
Blockchain-based settlement networks — whether public chains like Ethereum or permissioned infrastructure like Canton Network or Hyperledger Fabric — collapse this chain. Assets move on-chain with atomic settlement: either both legs happen simultaneously or neither does. No correspondent. No Nostro/Vostro reconciliation. No two-day wait to confirm the funds arrived.
The SIX Digital Exchange settles tokenised bonds in real time. BIS Project Mariana demonstrated FX settlement across multiple CBDCs with near-instant finality.
Instead of integrating a bank API wrapping a correspondent chain, you write to a smart contract that enforces settlement rules in code and leaves an immutable audit trail as a side effect.
The architecture exists. The question now is adoption velocity, not technical feasibility.
PaymentsMobile Infrastructure as the New Acceptance Network
Visa and Mastercard's moat was physical: tens of millions of POS terminals, each requiring certified hardware, proprietary protocols, and an acquiring bank relationship in the middle.
That moat is eroding from the edges. The smartphone is now a payment terminal — not metaphorically, but literally. Apple's Tap to Pay and its Android equivalents turn any NFC-enabled device into an acceptance point. No dedicated hardware. No certification process measured in months. No per-terminal cost structure.
A street vendor in Nairobi, a pop-up shop in Zurich, a freelance consultant in São Paulo can now accept card-equivalent payments with nothing more than a phone and an app.
More disruptive still is the direct-to-bank layer: open banking rails in Europe (PSD2), UPI in India, PIX in Brazil, and FedNow in the US are enabling account-to-account payments that bypass the card networks entirely. Real-time, low-cost, programmable. A developer building a payment flow today can route around Visa and Mastercard for a significant subset of use cases without the end user noticing.
FeaturesOpen APIs Replacing Internal Build Projects
Five years ago, if you wanted to add KYC verification to an onboarding flow, you staffed a compliance project, procured a data vendor, and spent months integrating. FX rates, document processing, sanctions screening, credit scoring — each was a six-month project in its own right.
Today, each of those capabilities is an API call. Stripe Identity, Onfido, and Jumio handle identity verification with a single SDK integration. Refinitiv streams market data with REST and WebSocket feeds. Plaid and TrueLayer surface bank account data directly. Codat normalises accounting data across dozens of SME platforms into a single schema.
The shift isn't just convenience — it's structural. Engineering effort migrates up the stack: toward product differentiation, toward the logic that is specific to your business rather than generic to your industry.
This is what "build vs buy" used to mean at the enterprise level. It now applies to individual features consumed by teams of three.
IntelligenceGenerative AI as the Integration Layer
The three layers above — programmable settlement, ubiquitous payment acceptance, and composable API features — are powerful individually. What generative AI does is make them accessible to a broader class of builder, and composable in ways that previously required significant integration work.
Consider what a practitioner can now orchestrate in a single session:
- Describe a structured product payoff in natural language and have a model translate it into pricing logic
- Ask about collateral eligibility and have a model query a smart contract, parse the result, and surface the answer in plain English
- Draft a client communication that pulls live market data, synthesises it against portfolio positions, and adapts it for a specific regulatory jurisdiction
None of these required new infrastructure. They required an intelligence layer capable of understanding intent, selecting the right tool from the available stack, and assembling the result coherently. Each infrastructure layer unlocks capability; generative AI lowers the threshold for accessing it.
The honest observation is that most financial services firms are not yet building as if this infrastructure exists. The compliance reflex is to wait for certainty. The organisational reflex is to build internally what could be consumed externally. The technology reflex is to evaluate new infrastructure against the stability standards of the old.
These are understandable positions. They are also, increasingly, competitive liabilities.
The firms building now are not waiting for SWIFT to become blockchain-native, for Visa to offer mobile-first rails, or for every API to achieve enterprise SLA status. They are building on the infrastructure that exists today, accepting its constraints, and shipping products that would have required 10× the team and 5× the time five years ago.
My conviction, having worked at the intersection of these layers across settlement, payments, and data: the infrastructure inflection already happened. The question is no longer whether the pipes exist. It is whether the organisations tasked with building on them have updated their assumptions accordingly.
The ones that have are building things that weren't possible before. The ones that haven't are explaining to their boards why the roadmap still looks like 2018.