S03.2 · Financial Services & Capital Markets
Card networks, processors and PSPs moving $2.4T in global revenue, now facing AI agents that initiate payments directly.
Payments is the transaction-processing layer between issuing banks, card networks, merchants and, increasingly, embedded-finance platforms — roughly $2.4T in global revenue in 2023 (McKinsey Global Payments Report), forecast to exceed $3T by 2029. The structural question worth watching is agentic commerce: AI agents initiating payments on a consumer's behalf, opening a rail category between agents and merchants that did not exist as a distinct commercial relationship two years ago. Whoever owns that rail sets the terms of the next interchange debate.
Growth has run at roughly 5-6% historically and is forecast at ~5% forward to 2029, a step down from the pandemic-era pace of digital payment-method shift. The regional economics differ more than the volumes do. APAC is the largest regional revenue pool, built on China's account-based payment ecosystem rather than card rails. The US carries the highest-margin economics of any major region because of its card-centric interchange structure. Europe is being actively reshaped by PSD2/3 regulation and instant-payments mandates that compress the fee pool available to incumbents — regulatory deflation, applied directly to the take rate.
Concentration depends on which layer is being examined. At the network level it is extreme: Visa and Mastercard together hold more than 80% of global card-network volume. The processing and payment-service-provider layer beneath is fragmented across a long tail that includes Stripe, Adyen, FIS, Fiserv and Worldpay. Upstream, the value chain runs through issuing banks and the card networks themselves; downstream, it terminates in merchants, e-commerce platforms and the embedded-finance layer that increasingly sits between a brand and its payment rail. The business is transactional and volume-based, with high operating leverage once scale is reached and only moderate capital intensity. Gating — money-transmitter licensing and PCI compliance — is materially lighter than a bank charter, which is much of why non-bank entrants have crowded into the processor and PSP layer.
The adjacencies all monetize the same asset: transaction data the payments company already sees. Banking-as-a-service, embedded lending, fraud and risk data, embedded investing. Card networks building out fraud-detection and data capability are protecting interchange revenue as fraud losses and regulatory scrutiny both rise — no new dollars there, only defense of existing ones. Processors moving into banking-as-a-service are doing the opposite, converting an existing merchant or platform relationship into a second, higher-margin product line.
Agentic AI's operational footprint is already visible in fraud detection, transaction reconciliation, and chargeback and dispute handling — functions that pair high transaction volume with well-defined rules, which is exactly what automates first. The merchant-support headcount cost line that processors have historically carried compresses accordingly, and the underwriting moats built on human review start to go with it. A fraud or dispute decision that once required a trained analyst now resolves through a model trained on the same transaction history, at a fraction of the marginal cost.
The structural shift is larger than the operational one. Agentic commerce — agents transacting on a person's or business's behalf without a human clicking "buy" — creates a payment-initiation category between the agent and the merchant, distinct from the traditional consumer-initiated card-present or card-not-present transaction. Who owns the rail for agent-initiated payment, and on what economics, remains open: either the incumbent networks extend their interchange model into agentic commerce, or a new layer of infrastructure captures that value instead. And unlike most of the AI story in this segment, this is not a multi-year horizon call. Agent-initiated transactions are being processed today, even if volume remains a small fraction of total payment flow.