One interpretation of the AI infrastructure race suggests foundation model companies build the rails and everybody else builds the trains. Anthropic just made clear it has other plans.
The company has launched ten new AI agents built specifically for the financial sector, covering pitch deck drafting, financial statement review, KYC screening, credit memo generation and compliance escalation. These are production-ready tools targeting the core workflows that fintech and financial services startups have spent the last three years building products around.
Why Now?
The launch was a logical next step – according to Anthropic’s own slide deck shared at a recent finance-focused event, financial services is now the company’s second-largest industry by enterprise revenue, trailing only technology.
Around 40% of Anthropic’s top 50 enterprise customers already exist in the financial sector. Claude has been embedded in banks, insurers and asset managers for long enough that Anthropic knows exactly which workflows those institutions use it for and exactly which agent templates would sell.
The revenue trend reinforces the strategic logic. According to estimates from Sacra, Anthropic’s annualised revenue reached approximately $30 billion in March 2026, up around 1,400% year-on-year from $9 billion at the end of 2025. Around 300,000 business customers now account for roughly 80% of that revenue – with a large share concentrated in finance, legal and professional services.
The agent launch is Anthropic moving to capture more of the value those customers represent, rather than sharing it with the startups building on top of its API.
What This Compresses For Startups
The core structural hurdle for app-layer startups is obvious. Many fintechs and SaaS companies built their defensibility around AI-wrapped workflows: pitch deck tools, financial statement analysers, KYC triage dashboards, compliance automation. Anthropic now offers ready-to-run templates for exactly those tasks, often integrated with Microsoft 365 and core office infrastructure. The competitive advantage these startups thought they had was really only a head start.
The competitive asymmetry goes further than product overlap. Because Anthropic controls both the model and the agent stack, it can optimise latency, tooling and pricing in ways that pure application-layer startups struggle to match without comparable infrastructure or capital. Claude Opus 4.7 scores 64.37% on the Vals AI Finance Agent benchmark, according to Anthropic, outperforming other models on vertical-specific tasks.
A startup selling a finance AI tool is now, in effect, competing with the company whose model it was likely built on.
The Bigger Pattern
The same compression is happening across verticals as foundation model companies follow the money into application layers. OpenAI’s move into health AI earlier this year followed the same logic: identify the highest-value workflows, build native agents, and let the API revenue from startups continue while capturing the application margin directly.
The finance vertical is particularly exposed because the workflows are well-defined, the buyers are big-ticket players with premium valuation appetites. Early data on AI agent usage cited by analysts shows finance domains still under 5% penetration of production tool calls, suggesting Anthropic ample room to advance.
For startups in this space, the threat of competition from foundation model companies is no longer a question; it’s a reality
The Road Ahead
A realistic assessment suggests that the most exposed startups are those whose entire product is a wrapper around a single workflow that a large model provider can replicate in a template. Startups with deep domain expertise, proprietary data integrations, regulatory relationships or workflow complexity that goes beyond what a general-purpose agent can handle are in a different position.
The HSBC Chief AI Officer appointment and the deep integration of AI in financial services both point to an enterprise market that is still early in deployment and still willing to pay for the right specialisation. The opportunity to build unique, defensible vertical AI in finance is still open, but the window is closing faster than anticipated.
