"That is a payment decision wearing an API costume."
That line, from the five-person UK gateway startup Requesty, is the closest thing to a decoder ring for the number that broke every venture investor's brain last week.
The number: OpenRouter closed a Series B at a $1.3 billion valuation in May 2026. By August, Stripe was reportedly in exclusive talks to buy it for around $10 billion. That's, per The Information, roughly 8x in about three months.
OpenRouter went from a $1.3B Series B in May to reported ~$10B Stripe talks by August — roughly 8x in about three months.
Now do the math the skeptics did in public. OpenRouter is a routing layer — it holds no GPUs, connecting to 70+ providers and 300+ models through one interface, on roughly $50 million in annualized revenue. That's a 200x revenue multiple for software that, as one Hacker News commenter noted, you could rebuild cheaply.
So let's concede the strongest version of the bear case up front. The core product is a base-URL swap and a model list. Everyone can see it. Vercel ships one. Cloudflare ships one. Kong, Portkey, LiteLLM ship them. Databricks built its own router in June. Ramp is reportedly launching one. If the model list is the asset, this is a middleware bubble with a payments logo stapled to it.
The bear case is correct about the code and wrong about the deal.
Return to that Requesty line and read what a gateway request actually contains. A model choice. A provider choice. A price per million tokens. A budget it draws down. An identity to attribute it to.
Strip away the JSON and that is not a technical event. It's a purchase order. Someone decided to spend money, picked a vendor, and got a bill — all inside a single API call that no human ever approved.
Which explains the buyer. Not a cloud, not a lab — a payments company. Stripe isn't acquiring a router; it's acquiring standing at the exact spot where autonomous agents make spend decisions "thousands of times per minute without a human clicking approve," as Requesty put it. Whoever sits there sees the money move before the finance team does.
That's the whole thesis, and it's a coherent one. Stripe's business is growing the volume of commerce on the internet and taking a margin on it. Machine-driven inference is now a large, fast-growing line item on that ledger. OpenRouter co-founder Alex Atallah has been calling his company "an AI equivalent of Stripe" — and Stripe, tellingly, is already the thing OpenRouter uses to bill its own customers.
So the $10 billion isn't priced off routing. It's priced off the possibility that AI spend becomes payment flow, and that this layer is where you meter it.
That reframing also tells you where the moat actually lives — and it isn't the model list.
The Hacker News commenter who called the software cheap to rebuild put the real value in the same breath: the existing customers and the switching costs. Once a company's logs, budgets, and cost attribution sit in one place, leaving is painful.
Requesty, which competes here and knows it, spelled out the mechanics for its own users. If your app calls an OpenAI-compatible endpoint and passes a model string, moving is a base-URL swap. If you've spread provider-specific SDK calls and hardcoded model IDs across a dozen services, moving is a quarter of engineering work.
Then the tell: "Logs, usage records, and cost attribution are the assets you will want on the way out." The advice is to make sure you can export them — because ownership changes are exactly when data terms get rewritten.
That is the moat. Not the routing. The ledger. Whoever holds the spend history, the per-team attribution, and the budget rules holds the customer, because ripping those out is the expensive part.
Which brings us to the frenzy — and to why it looks nothing like rational, standalone dealmaking.
The moment the Stripe talks leaked, the outreach turned absurd. Requesty — five employees — fielded at least 25 approaches in two weeks about investment, acquisition, or partnership. "The optimization race has reached a level of insanity," co-founder Thibault Jaigu said.
Concentrate AI, a ten-person shop founded last year, reportedly got seven approaches in the same window, spanning fintech, cybersecurity, and hyperscalers. "The level of popularity we've experienced over the past two weeks is something I could never have imagined," said co-founder Ari Jacoby.
Sit with those ratios. Twenty-five suitors for five people. Seven for ten. Snowflake talking to Not Diamond and Martian. Baseten, Cloudflare, and Vercel all working the phones.
That is not a market pricing assets on cash flow. Concentrate AI has no revenue story that justifies seven simultaneous bidders. This is a land-grab premium, priced off scarcity and fear of being locked out.
And here's the detail that proves it: Databricks had already built its own router in June — and still talked to OpenRouter about buying it.
Read that twice. A company that has already solved the problem in-house still wants the acquisition. Not for the code it already wrote, but for the accumulated model-selection data, the specialized team, and — most of all — to keep a rival from planting a flag at the chokepoint.
One investor in Not Diamond, Dean Mai of Myriad Ventures, said the quiet part out loud: for data-infra companies and hyperscalers, "it would be unwise not to consider acquisitions at this moment," even if they already have their own routing product, because the data and engineers hold independent strategic value.
Translation: buy it so the other guy can't.
The demand underneath all this is real, to be clear. Enterprises genuinely want to hedge across models — big banks are reportedly willing to run Chinese open-weight models on their own hardware, and Meta's new open-weight Glimmer reportedly pushes more agent inference toward local, swappable models. Fragmentation is the permanent condition, and a layer that dispatches across it has a reason to exist.
But "the layer should exist" and "this specific five-person router is worth billions" are different claims. The first is a thesis. The second is a defensive reflex.
Here's the uncomfortable test for every acquirer writing these checks. By 2027, everyone will have every model — Requesty says so, and it's obviously right. The model list is table stakes. What's left is routing quality, spend control, and the ledger.
So the honest question isn't "is routing valuable." It's narrower and harder: does this asset clear on its own cash flow, or are you paying to deny a competitor the chokepoint?
For Stripe, at $159 billion and simultaneously chasing PayPal, the spend-capture logic at least connects to an existing business — own how companies buy AI, meter the flow, take the margin. Alex Konrad's framing of a coming "Ramp vs. Stripe" showdown is really a fight over whose ledger the agent spend lands in.
For most of the 25 suitors chasing five-person teams, there is no such through-line. There's a rumor, a rival, and a fear of the empty chair at the chokepoint.
That's the difference between a thesis and a scramble. Stripe may be buying a payment decision wearing an API costume. Everyone else is buying insurance against Stripe — and calling it strategy.