Stripe has finalised an agreement to buy OpenRouter, the gateway that sits between applications and the models they call, for more than $7 billion, according to a Bloomberg report carried by TechCrunch on 16 August at 20:57 UTC. If it closes, a payments company will have paid AI-infrastructure prices for a toll booth.

The markup

OpenRouter's previous mark was $1.3bn, set in a $113m Series B announced in May 2026. The investor list on that round was Sequoia, Andreessen Horowitz, Menlo Ventures and Alphabet's CapitalG. A $7bn price is roughly 5.4 times that figure, reached inside a single quarter.

What the company actually does

OpenRouter is a router, not a lab. It exposes one API surface across more than 400 models to a claimed 8 million users, and its chief executive has described it as "the equivalent of Stripe for AI, because it provides customers with a single access point for different systems and prevents lock-in". That framing is presumably part of why Stripe is interested; it is also the company's own marketing.

What the common framing gets wrong

Much of the coverage has already changed the verb. Headlines now read that Stripe acquires, seals or completes the deal. Nothing has been announced by either party. The strongest available claim is a signed agreement — not a closed one, and not a confirmed one. Stripe told TechCrunch it "does not comment on rumors or speculation", and OpenRouter said nothing at all. No regulatory clearance, no closing date and no consideration mix between cash and Stripe stock is public.

The second error is arithmetic dressed as insight. The $1.3bn Series B valuation was itself reported rather than company-confirmed, so the widely repeated "5x markup" compares two unconfirmed numbers. The third is the one that matters most for anyone valuing the business: OpenRouter's revenue is disclosed nowhere. It is a pass-through router taking a thin margin on tokens it forwards. The gross inference spend flowing through it, and its 8 million users, are not its top line.

The layer being priced

Strip out the unconfirmed parts and one fact remains interesting: someone is willing to discuss $7bn for the abstraction layer between applications and models. That is more than most model labs carried at their own Series B. The routing tier — unglamorous, thin-margin, invisible to end users — is being valued as though it will be the durable position once model choice becomes a commodity.