Sokin hits $90m run-rate as it bets on agentic finance rails

Sokin, the London and New York-based payments infrastructure company, has disclosed its financials publicly for the first time, reporting $45 million in audited net revenue for 2025, profitability, and a current run-rate of $90 million. The company is tracking to exceed $120 million by December. The disclosure arrives alongside a strategic argument that will resonate well beyond the payments sector: that the lasting value in the agentic AI era lies not with the AI agents themselves, but with the regulated, licensed infrastructure those agents depend on to move money.

Founder and CEO Vroon Modgill framed the announcement in terms that speak directly to the convergence thesis. "The moat is not the AI agent," he wrote in the letter accompanying the results. "The moat is the infrastructure the agent connects to. An AI agent is only worth something if it stands on infrastructure that can act." The company moved more than $9 billion in payment volume in 2025 and held strong cash reserves at year-end.

Infrastructure as the AI-era competitive layer

Sokin's argument is structurally interesting at a moment when most AI discourse focuses on model capability. The company holds licences across 36 countries, operates through 18 direct banking partnerships, and can settle payments in more than 70 currencies across 170-plus countries. It has built its own stablecoin infrastructure this year and owns the full settlement stack, which it says allows it to settle the fiat leg of transactions that most crypto-native providers cannot complete. Embedded finance is now its fastest-growing route to market: fintechs and financial platforms are routing money through Sokin's rails invisibly, without their end customers ever encountering the Sokin brand.

The US is the company's largest market, accounting for more than 40% of revenue. Morgan Stanley backed the company in an early round; Prysm Capital led a $50 million Series B. The investor base is majority US-based, which matters for a company that is positioning itself as global payments infrastructure at a time when dollar-denominated settlement infrastructure carries geopolitical weight.

The convergence read-across: agents need rails

The broader strategic signal here runs across sectors. The rapid expansion of agentic AI systems in enterprise settings, autonomous workflows that procure, pay, and reconcile without human approval at each step, creates an immediate demand problem: existing payment rails were not designed for machine-initiated, multi-currency, multi-jurisdiction transactions at speed. Sokin's thesis is that the regulated licence footprint is the hardest element to replicate, because it cannot be generated by a model or acquired overnight.

This framing has direct implications for enterprise software, treasury management, and the growing field of agentic workflow platforms. Companies deploying AI agents into financial operations will need a counterparty with regulatory cover in their target jurisdictions. That structural dependency is what Sokin is positioning to capture. The Model Context Protocol integration the company is building this year would allow businesses to connect Sokin directly to their own AI tools and issue payment instructions through natural language, with a human approving each transaction.

The analogy Modgill reaches for is deliberately resonant for a cross-sector audience: "Just as AWS became the infrastructure layer software runs on, Sokin will become the infrastructure layer financial operations run on." Whether that comparison survives contact with larger incumbents, including bank-backed payment networks and the major card schemes that are also building agentic finance tooling, remains to be tested. The licence footprint is real; the competitive distance from well-capitalised rivals is not yet established. For investors and enterprise buyers allocating across the fintech and AI-infrastructure stack, the disclosure nonetheless marks a credible new data point in a space where independent, profitable infrastructure players are still rare.