Subscribe to get high-signal insights on how modern fintech is built.

analysis

What Stripe Gets from OpenRouter

Stripe paid $7 billion for a model router. The price makes no sense until you see what it turns Stripe into.

By Alex Kugell ·

Stripe just paid roughly $7 billion for OpenRouter, an AI model routing service with about $140 million in annual revenue. That's a 50x revenue multiple for a company that raised its Series B at a $1.3 billion valuation three months earlier.

The coverage has focused on whether Stripe overpaid. But the price only looks absurd until you see what Stripe becomes once it owns the routing layer for AI model calls.

The position Stripe is buying

Every company running AI in production faces the same operational problem: which model do you call, when, and how much does it cost?

OpenRouter sits between the application and the model providers, routing each request to the right model based on price, latency, and capability. Over 400 models from every major provider, with millions of developers already integrated.

Stripe already processes payments for 88% of the Forbes AI 50. When an AI company charges a customer, Stripe collects 2.5% to 3%. But when that same company pays for the model inference that generated the response, Stripe saw none of it.

Now Stripe sees both sides. Money in through payments. Money out through model routing. The take rate on an AI transaction jumps from roughly 2.5% to something closer to 7.5% when you combine both.

Why this couldn't be built

Stripe builds most of its infrastructure. It built Atlas, Billing, Radar, Issuing. So why buy a model router at 50x revenue?

Network effects. A model router is only useful if it has relationships with every model provider and usage data from millions of developers. OpenRouter had both.

Building a router from scratch means starting with zero routing data, zero provider relationships, and zero developer trust while trying to convince the same companies you already process payments for to also let you manage their AI spend. That's a cold start problem that money alone doesn't solve.

Ramp understood this immediately. Hours after the deal leaked, Ramp launched Router.com, promising 40% cost savings on inference. The race to own the model selection layer started the moment Stripe signaled it was valuable.

The structural play

Corporate AI spend has grown 20.7x since June 2025, according to Ramp's data. For most companies, model inference is the fastest-growing line item on the P&L. Whoever controls the routing layer controls the purchasing decision: which model, which provider, at what price.

Stripe already sits at the center of how AI companies make money. Metronome (acquired earlier) handles usage-based billing. Stripe handles payment processing. OpenRouter now handles model procurement.

That's the full financial loop: meter usage, route the model call, collect the payment.

The closest analogy is a general contractor who already handles your invoicing and now also controls which suppliers get the materials orders. They see your costs, your margins, and your vendor relationships. The information advantage compounds faster than the transaction fees.

What this means for the rest of payments

OpenRouter also built something called the Machine Payments Protocol, a framework for AI agents to pay each other using stablecoins without human involvement.

Stripe hasn't said what it plans to do with this. But if agents start initiating and settling transactions programmatically, the company that owns both the model routing and the payment processing layer is positioned to be the default rail.

Visa and Mastercard are still figuring out how to authenticate an agent. Stripe just bought the company that routes the agent's brain.

Sources

Alex Kugell
Alex KugellAI/ML Engineer, Fintech Infrastructure

Writes Ledger Drift. Engineer building AI systems and financial infrastructure. Open source contributor to x402, ACK, and MPP.

Frequently Asked Questions

Why would Stripe pay 50x revenue for a model routing company?
Stripe processes payments for 88% of the Forbes AI 50, so it already captures roughly 2.5% when an AI company charges a customer. Owning the routing layer lets Stripe also capture the model procurement side, lifting the take rate per AI transaction closer to 7.5%. Combined with Metronome for usage-based billing, Stripe now touches every financial event in an AI company's inference loop.
Why couldn't Stripe build a model router instead of buying one?
A router needs relationships with every model provider and behavioral data from millions of developers to route effectively. Starting from zero on both while asking existing payment clients to also hand over their AI spend is a cold start problem that capital alone doesn't solve. Ramp launched a competing product within hours of the deal leaking, which tells you how fast the window for owning this layer was closing.
How does the OpenRouter deal change Stripe's position in agent-to-agent payments?
OpenRouter built a protocol for AI agents to settle transactions with each other using stablecoins, without human approval loops. If agents start initiating and completing payments programmatically, the company that owns both the model routing layer and the payment processing layer becomes the default rail. Visa and Mastercard are still working on agent authentication while Stripe now controls where the agent's inference requests go.
What information advantage does Stripe gain from owning the routing layer?
Stripe sees which models each customer calls, how much they spend on inference, which providers they rely on, and how their usage patterns change over time. That data arrives before it shows up in public benchmarks or provider earnings. Paired with payment data from the revenue side, Stripe now sees both the cost structure and the revenue of its AI clients simultaneously.
Why is model inference the fastest-growing corporate expense?
Corporate AI spend grew over 20x in twelve months according to expense management platforms, and for most companies, inference is now the fastest-growing line item. Each AI feature a company ships adds to the compute bill, and the cost per request varies based on model choice and task complexity. Controlling which model gets called on each request means controlling the largest variable cost in a growing number of companies' P&L.

Built by Trio, a fintech-native engineering partner helping teams build the next generation of financial technology and infrastructure.

Subscribe to Ledger Drift for high-signal insights into how modern fintech is built, from systems to code to teams.

Keep reading

ai & mlRAG for Compliance Docs: What Actually WorksMost RAG pipelines are tuned for clean prose, but compliance documents have cross-references, nested definitions, and tables that break every default.
analysisAfrican Fintech: The LeapfrogAfrica's fintech leapfrogs are real where they follow market structure and overhyped where they import assumptions from somewhere else.
analysisThe OCC's AI Risk Tiers and What They Mean for EngineersThe OCC gave examiners a four-tier rubric for scoring AI in banks, and your architecture determines which tier you land in.
View more ›