Stripe has finalized an agreement to acquire OpenRouter for more than $7 billion, according to Bloomberg, with TechCrunch matching the report hours later on Sunday and Fortune putting it as clinched. Neither company has confirmed it, and a Stripe spokesperson told TechCrunch it does not comment on rumors or speculation, which is roughly the sound a deal makes in the week before it gets announced.
Read the price and the logic falls out. OpenRouter closed a $113 million Series B in May at a $1.3 billion valuation. Ninety days later a payments company is reportedly paying more than five times that. Nobody pays a 5x markup in a quarter for a product. They pay it for a position.
What Stripe Is Actually Buying
OpenRouter is a gateway. Developers integrate once, and OpenRouter routes the request to whichever of its 400-plus models is cheapest or best for the job, pulling from OpenAI, Anthropic, Google, Meta and DeepSeek. It handles the selection, the failover, and the billing. TechCrunch reported in May that the company had more than doubled its valuation in a year on the back of that traffic, and it now claims roughly 8 million users and about 100 trillion tokens routed a month, a fivefold jump in half a year.
The number that matters to Stripe is none of those. It is the take rate. OpenRouter clips roughly 5% of every API call that passes through it.
That is not an AI business. That is an interchange business, and it happens to sit at the exact junction where the AI industry’s money now moves.
Seats Became Tokens, and That Changed Who Owns the Rail
Software used to be sold by the seat, which is a subscription, which is a boring recurring charge Stripe has processed a billion times. AI is sold by the token, which is metered consumption with volatile unit costs, per-model price differences, and a bill that swings with usage. Metering and settling that is a genuinely hard problem, and it is Stripe’s actual franchise.
We have watched what happens when a company gets the metering wrong. When GitHub moved Copilot to metered billing, developers hit cost surges of up to 10x and revolted. The pain there was not the model quality. It was the invoice.
So Stripe is not buying intelligence. It is buying the position between the developer and the model vendor at the precise moment that position turns into the billing rail for the industry, and it is buying the routing data that comes with it: who is spending what, on which model, at what margin, across 8 million developers. That telemetry is worth something separate from the revenue.
It also fits a direction Stripe has been telegraphing. The company has been building toward machine-to-machine commerce for a while, including its work on the x402 agent payments protocol with Coinbase and Cloudflare. An agent that can pay for things needs a meter. Stripe just bought the best-instrumented one.
The Investors Called It Before the Buyer Did
The May round was led by CapitalG, Alphabet’s independent growth fund, with NVentures, Nvidia’s venture arm, alongside ServiceNow, MongoDB, Snowflake and Databricks Ventures, plus existing backers Andreessen Horowitz and Menlo Ventures. At the time, CapitalG partner Mo Jomaa described the company by analogy.
OpenRouter, he argued, was becoming to AI what Cloudflare, Databricks and Stripe are to their own layers of the stack: unglamorous infrastructure that everything else ends up routed through.
Three months later, Stripe agreed with him literally enough to write a check for more than $7 billion. Alphabet’s growth fund marked its position up roughly fivefold in a quarter, which is a fine outcome for CapitalG and an uncomfortable data point for anyone arguing that AI infrastructure pricing has cooled off.
The Neutrality Problem Nobody Has Priced
Here is the tension the deal creates, and it is not small.
A router’s entire value proposition is that it is indifferent. Developers trust OpenRouter to send a request to the best or cheapest model because OpenRouter has no horse in the race. The moment it belongs to a company with commercial relationships across the payments stack, and with an Alphabet-affiliated fund on the cap table that just cashed a 5x, “indifferent” becomes a claim that has to be audited rather than assumed.
Enterprises are already sensitive to this. The efficiency push that has buyers optimizing token spend rather than just buying more of it depends on routing decisions being made on price and performance. If those decisions start reflecting the owner’s economics, the product’s core promise erodes quietly, and the customers who notice first will be the largest ones.
The model vendors have their own read. Anthropic, OpenAI and Google have spent two years trying to own the direct enterprise relationship. A gateway that abstracts them into interchangeable line items on someone else’s bill is the commoditization scenario they have been fighting. Handing that gateway to Stripe, with its distribution and its merchant relationships, makes the abstraction layer considerably harder to route around.
What to Watch
Watch whether Stripe confirms, and how it describes the purchase when it does. If the language is about developer tooling, this is a product acquisition. If the language is about AI-native billing and agentic commerce, Stripe is telling you it bought a toll booth and intends to run it as one.
Then watch the take rate. Five percent of every call across 400 models is either the most defensible position in applied AI or the most obvious thing for a model vendor to undercut with a free first-party router. Stripe just bet more than $7 billion that it is the first.