OpenRouter's whole reason to exist was that nobody should be locked into one AI vendor. One API key, over 400 models, a provider picked for every request, switch whenever you feel like it. This week Stripe agreed to buy it, for a reported 7.5 to 8 billion dollars, mostly in stock. The aggregator got aggregated.
I care because that box sits between my code and every model I use. And because the buyer isn't another AI lab, it's a payments company, which turns out to be the interesting part.
The numbers, quickly
The short version: rumors in late July around 10 billion, a signed deal reported mid-August at 7 to 8 billion, official confirmation on August 19 on both companies' blogs. Stripe won't confirm any figure, so every number here is reporting, not fact. The New York Times says about 7.5 billion, with 1.5 billion going to the three founders.
Two details stand out. OpenRouter raised money in May at a 1.3 billion valuation, so this is a six-fold markup in about three months. And the price came down from the July rumor, reportedly because half the industry (Ramp included) started building its own routers in the meantime. The moat is thin, and everyone at the table knew it.
The business itself is simple: about 5% on top of your inference spend, roughly 140 million in annualized revenue this summer per Sacra, on the order of ten trillion tokens routed per day. And signed is not closed. They expect to wrap up "in the coming weeks".
Stripe didn't buy a proxy, it bought the meter
Stripe has spent two years buying the plumbing of AI money: Bridge for stablecoins, Metronome for usage-based billing, an agentic commerce protocol built with OpenAI, its own blockchain for machine payments. Patrick Collison's line in the announcement: "tokens are the central currency for companies building with AI."
Put those together and the logic is clean. Metronome counts the tokens and bills them. OpenRouter decides which model gets them. Stripe now sits at the moment of selection and the moment of settlement, and picks up a market-wide view of what everyone is actually buying on the way. Eight billion for a proxy sounds insane until you notice the proxy was never the product. The meter is.
What happens to things Stripe buys
The track record is worth knowing, because it's genuinely mixed. Lemon Squeezy, bought in 2024: add-on fees crept upward, the CEO publicly owned "slower support responses and less frequent product updates", and that team is now building Stripe Managed Payments, Stripe's own product, with a plan to migrate Lemon Squeezy users onto it. TaxJar, bought in 2021: sales team gutted, customers steered to Stripe Tax. On the other side, Paystack still runs independently and is thriving, and Bridge became the core of Stripe's stablecoin stack.
The pattern is legible. When an acquisition overlaps with something Stripe wants to build itself, it gets absorbed and the original slowly starves. When it's a network Stripe can't rebuild, it gets left alone. OpenRouter, with its 400 models and millions of developers, looks like the second kind. That's the optimistic case, and it's a real one.
The neutrality question
The tension is simple. OpenRouter's pitch was neutrality, and the post announcing the deal literally promises routing stays driven by "what's best for you, the user". But Stripe has deep commercial ties to OpenAI and handles payments for most of the big US labs. Meanwhile, what's actually flowing through OpenRouter right now is cheap Chinese open-weight models: above 30% of enterprise token volume all spring per a CNBC investigation, over 60% of routed traffic per OpenRouter itself. I've written twice about those models eating the paid labs' lunch, and OpenRouter is the counter where that lunch gets served.
To be precise about facts: nobody has seen Stripe touch the routing, and both companies say nothing changes. But the incentive now exists where it didn't before, and that's worth naming even if it's never acted on.
What to watch
I use OpenRouter and this changes nothing for me today. Nothing changes for anyone at signing. But if this deal ever turns sour, it won't announce itself, it'll show up in the small print. Four places worth keeping an eye on: the fees (5.5% on credits and 5% on BYOK today), the data policy (a payments company now sits in the request path), KYC and content rules (payment processors have a history there), and the catalog (cheap Chinese models quietly losing placement would be the tell).
If you're at scale or in a regulated shop, a gateway abstraction in front of your calls is cheap insurance however this goes: LiteLLM in your own infra, the Vercel or Cloudflare gateways if you already live there, Portkey if governance is the job. The API being OpenAI-compatible means switching is a config change, which is most of the reason not to worry too much.
The honest caveats
The price comes from journalists' sources, not an official figure. The deal is signed, not closed. The revenue numbers are estimates from Sacra and The Information, not audited accounts. "Nothing changes" is a vendor claim, and Lemon Squeezy shows a day-one promise can hold for a year before the roadmap drifts. And fairness cuts both ways: Stripe owns no frontier model, which arguably makes it the least conflicted giant that could have bought this thing, and Stripe's money and fraud tooling could genuinely make OpenRouter better. Paystack fans exist for a reason.
Where this might go
The optimistic read: Stripe money makes OpenRouter faster and more reliable, and the neutral layer stays neutral because its users can leave in an afternoon. The cynical read: the meter slowly gets more expensive, the way meters do.
My hot take, held loosely: the 5% fee was never the prize. OpenRouter sees which models the entire market picks, at what price, the moment it happens. No lab has that view, and even the big clouds only see their own slice. Stripe just bought the best seat in AI for watching where the money goes, and it already owns the rails the money runs on. Watch the fees, not the press releases.
Drawn from the Stripe newsroom and OpenRouter announcement posts, reporting by Bloomberg, Axios, The New York Times, The Wall Street Journal and TechCrunch, Sacra's OpenRouter estimates, The Information's cost figures, CNBC's investigation into Chinese-model traffic, and the announcement-day Hacker News thread. A snapshot as of August 24, 2026. The price is reported, not confirmed, the deal hadn't closed at publication, and every "nothing changes" in here is the vendors talking, so check the announcements before you quote me.