OpenRouter does not build models. It sits in front of them. One API key, hundreds of proprietary and open-weight models behind it, and the ability to compare price, latency and quality and switch between them without rewriting anything. Founded in 2023, it is the boring plumbing of the inference economy — and according to reporting from The Information on Thursday, matched by The Wall Street Journal on Friday, Stripe is in talks to buy it for close to $10 billion.
Read that against its own funding history. In June 2025, OpenRouter raised a $40 million Series A led by Andreessen Horowitz and Menlo Ventures at a $547 million post-money valuation. In May 2026 it raised a $113 million Series B led by CapitalG — with NVentures, ServiceNow Ventures, MongoDB Ventures, Snowflake Ventures and Databricks Ventures alongside — at roughly $1.3 billion. Ten billion is about eight times that mark, set roughly two months ago. Menlo has put the platform’s throughput at more than 100 trillion tokens a month, a run rate north of a quadrillion tokens a year, across some 8 million users.
Nothing is signed. The Information reports a deal could be announced within a month, and also that talks could collapse. OpenRouter is reported to have held earlier sale conversations with Databricks — already an investor — and several other large tech companies are said to have evaluated bids.
Stripe keeps buying the toll booth, not the road
Stripe’s entire business is being the layer companies decline to build themselves: metering, billing, failover, fraud, reconciliation. OpenRouter is that exact shape for inference. If AI agents are going to spend money on compute the way software spends money on bandwidth, somebody has to meter the tokens, price them, switch providers when one goes down and hand the customer a single invoice. That is a payments problem wearing an AI costume — and it lands the same month Stripe and Advent bid $53 billion for PayPal. Stripe is not buying AI. It is buying the meter.
Our take: This is the clearest price yet on a thesis this site has tracked all year: value is draining out of individual models and pooling at the switching layer. When up to 46% of U.S. tokens can shift onto Chinese open-weight models on price alone, and when production workloads run open while frontier models get reserved for the hard 5%, the router is the only participant that wins on every outcome. Every price war between model vendors makes it more valuable, not less, because it gets paid on volume regardless of who is cheapest that week. A $10 billion tag on a company with no model, no chips and no data center is the market saying the commodity is the intelligence and the franchise is the switchboard. The uncomfortable corollary: labs burning $100 billion in capex may be building the thing that gets arbitraged.
What to watch
- Whether it closes at all. Two outlets, one figure, no signed agreement. Competing suitors have circled before, and $10 billion for a two-month-old $1.3 billion valuation invites a counterbid.
- The neutrality problem. A router’s value is that it is indifferent between vendors. Its cap table already includes Google’s CapitalG, Nvidia’s venture arm, Snowflake, MongoDB, ServiceNow and Databricks. Ownership by any one strategic buyer is a reason for developers to ask who the routing rules now favor.
- Take rate, not token count. Quadrillions of tokens is a vanity metric until you know the margin on each one. The multiple only works if the router keeps a real cut rather than passing through at cost to win volume.
- The read-across. If routing is worth $10 billion, every middle-layer company — gateways, eval tooling, observability, spend management — just got repriced by comparison. Expect that to show up in funding rounds before it shows up in revenue.