Embedded finance

The Synapse collapse defined the post-2024 regulatory baseline

Synapse Financial Technologies' April 2024 Chapter 11 was the structural event for the category. The intermediary held ledgers between fintech clients (Yotta, Juno, others) and partner banks (Evolve Bank & Trust, Lineage, AMG, American). When the company collapsed, $265M of end-user funds were frozen and a $65-95M reconciliation shortfall emerged from pooled-custody bookkeeping that could not be reconstructed. The regulatory response set the industry baseline. The FDIC's September 2024 'Synapse Rule' proposal requires banks to verify custodial-account beneficial-owner records daily, not periodically. The Federal Reserve issued a cease-and-desist to Evolve Bank in June 2024 finding it had inadequate risk management for fintech partnerships. Blue Ridge Bank, Five Star Bank, and Metropolitan Commercial Bank exited BaaS entirely. The CFPB allocated $46M toward Synapse/Evolve victims - the first fintech-specific consumer-protection allocation at scale. The structural read is that BaaS has bifurcated: well-capitalised orchestrators with compliance-led operating models (Synctera, Unit, Treasury Prime) survive and consolidate share; the long tail of middleware providers without the same risk-management footprint exits or is acquired.

State of the art (2026)

Embedded finance in 2026 is a regulated industrial layer, not a frontier. The durable economics sit with issuing and payments: Stripe reported $1.9 trillion in 2025 total payment volume (up 34%) and Marqeta processed a record $383B for full-year 2025 (+31%). The BaaS orchestration tier bifurcated after the April 2024 Synapse Chapter 11 – compliance-led survivors (Synctera, Unit, Treasury Prime, Column) consolidate share while thinner middleware exits. The FDIC custodial-account recordkeeping rule that defined the post-Synapse baseline remains unfinalised, and 2025 FDIC leadership signalled rescission of several proposals, leaving the compliance floor uncertain. The live demand layer is agentic commerce: the Stripe-OpenAI Agentic Commerce Protocol and ChatGPT Instant Checkout shipped in September 2025, with Visa Intelligent Commerce and Mastercard Agent Pay routing agent payments through existing card rails.

Issuing infrastructure and payments incumbents capture the durable economics

The durable revenue layer in embedded finance is issuing and payments, not orchestration. Stripe processed $1.9 trillion in total payment volume in 2025 (up 34%, roughly 1.6% of global GDP); Marqeta processed a record $383B for full-year 2025 (+31%, led by European expansion). Card-issuing infrastructure - Marqeta, Lithic, Highnote, Galileo, i2c, Thredd - has commoditised the underlying card-network connectivity but kept the embedded-card and embedded-wallet vertical-SaaS layer. Adyen extends the same pattern in Europe with deeper acquiring footprint. Block and PayPal hold the consumer-facing tier. The BaaS orchestrators (Synctera, Unit, Treasury Prime, Column) sit above this layer offering compliance-and-orchestration but earn lower take rates and are more exposed to partner-bank concentration risk. The structural read is that the issuing-and-payments tier compounds while the orchestration tier consolidates - a 2-3-player BaaS oligopoly emerges by 2027.

AI agentic commerce is the next demand expansion layer

Embedded finance through 2024 was about putting payments, lending, and accounts inside non-financial software at the human-user level. The next layer is autonomous AI agents that execute payment, lending, treasury, and insurance actions on behalf of merchants and consumers - agentic commerce. This layer is now live, not theoretical: the Stripe-OpenAI Agentic Commerce Protocol and ChatGPT Instant Checkout shipped in September 2025, Visa launched Intelligent Commerce in 2025 (with Intelligent Commerce Connect following in April 2026), and Mastercard Agent Pay binds tokenised credentials to specific agents. Stripe Issuing, Adyen platform APIs, Marqeta program APIs, and the BaaS orchestration layer are all positioning to support agent-initiated transactions. The open question through 2026-2027 is whether agent-led volume becomes a material share of embedded-payments GMV and whether it routes through existing issuing-and-payments rails (most likely) or whether new agent-native settlement layers emerge that bypass existing card-network economics.