Stripe Acquires OpenRouter and Moves Into the Layer That Decides Which Model Gets the Request

01 · What happened

Announced 19 August 2026. Price not disclosed by either company. Reported at $7.5 billion.

Stripe has agreed to acquire OpenRouter, the model gateway that routes requests across more than 400 models from more than 80 providers. The announcement went out from San Francisco and Dublin on 19 August 2026. It contains no price. The New York Times reported $7.5 billion the same day, citing a person with knowledge of the agreement, three days after Bloomberg reported a figure above $7 billion.

Three years ago OpenRouter did not exist. It was founded in 2023 by Alex Atallah, previously cofounder and chief technology officer of OpenSea, alongside Louis Vichy and Chris Clark. It employs roughly fifty people in New York. In May 2026 it raised $113 million led by CapitalG at a reported $1.3 billion valuation. Eighty-three days later it agreed to sell.

The strategic framing in the release is precise and worth reading literally. Stripe says it already helps businesses maximise revenue by optimising across payment methods, authorisation rates and fraud. OpenRouter optimises the other side, deciding which model handles which task, at what speed and at what price. Patrick Collison framed the combination as helping companies manage both sides of profitability in the AI era.

What is in the announcement, and what is not

The release is generous on strategy and silent on terms. It names three customers, NVIDIA, Zoom and Lovable. It states the model and provider counts. It quotes both chief executives at length on the multi-model thesis. It does not state a price, a consideration mix, a closing date, a condition, or a regulatory jurisdiction. A Stripe spokesperson told TechCrunch the company does not comment on rumours or speculation.

That places every number in this article in one of three categories. Published company figures, such as the platform fee and the model counts. Named press reporting, such as the price. Third-party estimates, such as revenue. The table below marks which is which, and the distinction carries through every chart that follows.

02 · The deal at a glance

Transaction terms Reported Public record Acquiry calculation Not disclosed

Where the reported consideration lands

One outlet has published a split of the price between founders and investors. Neither company has confirmed a figure of any kind.

Reported

The split comes from the New York Times, attributed to a single person with knowledge of the agreement. Against $164m of capital raised, $6.0bn to investors implies a strong return for a company founded in 2023. The mix between cash and stock is not disclosed, and that mix determines how much of the value is realised at close and how much rides on Stripe equity.

Reporting opened at roughly $10bn while the parties were in talks, settled above $7bn when Bloomberg reported the agreement, and arrived at a specific $7.5bn on the day of announcement. Stripe told TechCrunch it does not comment on rumours or speculation, and both companies declined to disclose the value. Every multiple on this page is built on the reported figure and inherits its uncertainty.

Acquiry calculationMultiple derivation and sensitivity

Only one of the three inputs to the headline multiple is a company figure. The price is press reporting and the revenue is a research estimate, so the multiple carries a band rather than a point value. The grid below runs both inputs across their reported range.

Across the full grid the multiple runs from 41× to 91×. The range is wide because the inputs are wide, and no amount of arithmetic narrows it. What can be said is that on every combination in the table this is a growth multiple rather than an earnings multiple, and it is priced against a revenue line that has moved by an estimated factor of 28 in twelve months.

03 · The asset

OpenRouter is one API endpoint that reaches more than four hundred models from more than eighty providers, and a routing engine that decides which one handles a given request. That sentence does more work than any product page. The single endpoint is the distribution mechanism: a developer integrates once and gains access to every model on the platform, including ones released after the integration was written. The routing engine is the retention mechanism: once an application is tuned to route on cost, latency and reliability rather than on a hardcoded model name, moving off the platform means rebuilding that logic.

The commercial signature follows from the structure. Adoption starts with a single developer and a test key. Expansion happens as applications move into production and volume grows. Removal is possible, because the API is deliberately OpenAI-compatible, but it means giving up provider failover, price arbitrage across four hundred models, and the accumulated knowledge of which model performs on which task. The asset is not the proxy. It is three years of demand data and eight million developers who already have a key.

The three-layer read

• Access layerOne OpenAI-compatible endpoint, four hundred models, eighty providers, unified billing through a credit balance. Low integration cost. This is what gets it into the codebase.
• Routing layerProvider failover, price and latency optimisation, and quality-aware routing. Published model variants let a developer request the fastest provider, the cheapest provider, or the one tuned for tool-calling reliability. This is what makes it hard to replace.
• Governance layerWorkspaces, spend management, guardrails and zero-data-retention policies. This is where enterprise contracts sit, and where the platform is youngest.

The development opportunity sits on the governance axis. OpenRouter scores highest in its category on breadth, adoption and neutrality. Enterprise controls, audit, observability and procurement-grade apparatus are where Portkey and Braintrust have invested most heavily, and where a 90-person company serving 10 million developers and companies has the clearest room to build. That headroom is a meaningful part of why a platform routing hundreds of trillions of tokens a month is estimated to earn roughly $140 million of net revenue.

04 · Unit economics

Tokens carry no mark-up. The per-token price a customer pays on the platform matches the price the provider charges directly. This is published in the company's own pricing documentation and it is the foundation of the neutrality position: there is no financial reason for the router to prefer one model over another, because it earns the same regardless of which one wins the request.

Revenue comes from a platform fee charged when a customer loads credits. 5.5 per cent on fiat, with an eighty cent minimum. 5.0 per cent on cryptocurrency. 5.0 per cent on the list-price equivalent for customers bringing their own provider keys, after the first million requests a month. Those are published figures, not estimates.

The consequence matters for every multiple in this article. A reported $140 million of annualised revenue at a blended 5.5 per cent fee implies roughly $2.5 billion of annualised inference spend flowing across the platform. Gross routed spend and net revenue differ by a factor of about eighteen, and using the wrong one produces a multiple that is wrong by the same factor.

Gross flow against net revenue Public record Indicative Acquiry calculation

The bottom three rows are the same transaction described three ways. Anyone quoting a multiple on this deal should say which line they are using. Three times gross flow sounds inexpensive; seventy-six times gross profit does not. Both are arithmetically correct.

Why routing has commercial value

Published list price per million input tokens across four widely used models.

Public record

The spread between the cheapest and the most expensive input token in this set is thirty times. That spread is the entire commercial case for a routing layer. Sending a simple classification task to a model priced at three dollars per million tokens when one priced at ten cents would do the job is the cost problem OpenRouter exists to solve, and it is the problem Stripe says it wants to solve alongside the revenue side.

05 · Buyer rationale

The buyer has an identifiable acquisition pattern: buy a capability that plugs into the existing money flow, keep the team, fold the product into the platform. Paystack brought African payments. TaxJar became Stripe Tax. Bridge brought stablecoin rails at a confirmed $1.1 billion. Metronome brought usage-based billing. Every one of those was bought to extend what Stripe already did for its customers.

This transaction fits the pattern on strategy and breaks it on scale. At a reported $7.5 billion it is roughly 6.8 times Bridge, which was the largest deal Stripe had done before it, and larger than every prior acquisition combined on any reasonable reading of the reported figures.

The four rationales, ranked by how well the evidence supports them

Strategic rationale, assessed Acquiry assessment

The transaction is underwritten principally on the first two rationales, and they reinforce each other. Stripe is buying the cost side of a customer base it already serves on the revenue side. The remaining items are optionality layered on top rather than the basis for the price.