Stripe has, according to Bloomberg, finalised a deal to buy OpenRouter for more than $7bn. The payments giant declined to comment, and the sum is not officially confirmed, but the reported number tells its own story about where the smart money in AI now thinks the durable profit lies.
OpenRouter is what the industry calls an AI gateway, or a model router. In plain terms, it gives developers a single doorway to more than 400 models, and lets them switch between providers such as OpenAI, Anthropic and Google according to price, speed or mood, without rewriting their code or marrying any one lab.
Roughly 8 million users now pass through it, drawn by the promise of never being held hostage by a single vendor’s pricing or downtime. The pitch is neatly captured by chief executive Alex Atallah, who has called OpenRouter “the equivalent of Stripe for AI”, a comparison that reads rather differently now that Stripe has apparently agreed with him.
Which is presumably why Stripe agreed with him and bought it. The logic fits a company that made its fortune not by inventing money but by metering its movement, taking a modest slice every time value changes hands online.
Apply that instinct to a fragmenting model market, where developers juggle a dozen suppliers and hate being locked in, and the appeal of a neutral billing-and-routing layer becomes obvious.
It is the same wager that other infrastructure players are placing on the plumbing rather than the product, as Baseten’s $1.5bn raise on cheap inference made plain earlier this year.
The price is the part that should make you sit up. The Wall Street Journal reported the two sides were in talks back in July; Bloomberg now says the deal is done.
OpenRouter closed a $113m Series B only in May 2026, at a $1.3bn valuation, backed by Sequoia, Andreessen Horowitz, Menlo Ventures and Alphabet’s CapitalG. Three months later, Stripe is reportedly paying more than five times that figure. Markups like that are not paid for revenue. They are paid for position.
And the position here is genuinely strategic. As frontier labs commoditise one another and prices keep sliding, the model itself starts to look like the least defensible link in the chain.
The switching layer, by contrast, gets more valuable the more the market splinters, because someone has to meter, route and bill usage across all of it.
That is the thesis behind bets like IBM’s $240m push into cheap open-source inference, and it is a nervy one for anyone hoping a single model will command lasting rents.
That nervousness is not abstract. The same relentless drive towards the cheapest adequate option, sometimes called thrift-maxxing, is exactly the behaviour OpenRouter industrialises.
Every time a developer swaps a pricey flagship for a cheaper alternative to shave a bill, the router wins and the model loses a little pricing power. Stripe, in other words, is buying the machine that helps customers shop around, which is a rather elegant place to stand.
There are caveats worth keeping, and this being AI, plenty of them. The deal is reported rather than announced, Stripe is saying nothing at all, and regulators on both sides of the Atlantic tend to take a keen interest when a payments behemoth swallows a chokepoint.
Neutrality, too, is a fragile asset; a router owned by one giant may not feel quite as impartial to the labs whose traffic it meters, and some may think twice about routing through a rival’s cash register.
But if the reported figure holds, the message is hard to miss. In the AI gold rush, Stripe has decided the safest bet is not the gold. It is the tolls on the road to the mine.
Get the TNW newsletter
Get the most important tech news in your inbox each week.