Stripe’s OpenRouter deal puts routing beside billing
Stripe’s reported OpenRouter acquisition, travel referrals near conversion parity, and gaps in AI review and inference infrastructure.
By Saeed Ezzati6 min read
The audio edition
Listen to this newsletter
0:003:44
Read transcript
Stripe is acquiring OpenRouter, bringing an AI model-routing layer into the same company that already handles its payments and token billing. The terms are not confirmed: published estimates range from seven-point-five billion dollars to more than eight billion, with closing expected within weeks. That uncertainty matters, especially because OpenRouter was reportedly valued at one-point-three billion dollars in May. OpenRouter gives developers one endpoint for more than 400 models from more than 80 providers. Its system weighs complexity, price, and speed, then routes each request to what it determines is the best-fit model. Developers can change providers without changing their code. OpenRouter says more than 10 million developers and companies use the platform, processing more than 10 trillion tokens a day. Nvidia, Zoom, and Lovable are among its customers. SiliconANGLE reported that OpenRouter retains about 5 percent of inference spending processed through the platform, with revenue near a 50-million-dollar annualized rate in March. The strategic connection predates the acquisition. Stripe has supplied OpenRouter’s payments infrastructure since at least January, and the companies built token billing that automatically meters and prices model use. Stripe also recently acquired usage-based billing company Metronome, after completing its Bridge acquisition and later buying crypto-wallet developer Privy. In other words, Stripe is assembling infrastructure around how software usage is measured, paid for, and now potentially routed. OpenRouter says it will remain independent, preserve its customer-first approach, and keep routing calls based on what is best for the customer. Those are stated commitments, not disclosed deal terms, and the formal structure remains unknown. Stripe co-founder and chief executive Patrick Collison described the rationale as using scarce compute efficiently and helping businesses spend tokens effectively. OpenRouter chief executive Alex Atallah has argued that AI will be multi-model and needs a neutral orchestration layer. The practical test after closing is whether that neutrality holds while Stripe gains a deeper position around the spending those routing decisions create. That same infrastructure story is appearing one layer earlier, in how people find travel products. Adobe says AI referrals to U.S. travel sites rose 119 percent year over year in July, while those visits converted at a rate just 1 percent below non-AI traffic. A year earlier, the gap was 47 percent. The visitors also showed 21 percent higher engagement, spent 67 percent more time on site, and had a 42 percent lower bounce rate. The caveat is that travel has reached near-parity, not leadership. Across U.S. retail sites, Adobe said AI traffic rose 62 percent and converted 60 percent higher than non-AI traffic, marking the 11th consecutive month in which AI-directed retail conversion exceeded the non-AI rate. Adobe’s survey of 5,000 U.S. consumers found that 84 percent said AI improved their travel-planning experience, though that measures sentiment rather than purchasing behavior. The weaker link is discoverability. Adobe’s Content Visibility Checker gave airline homepages an average score of 42 percent, compared with 71 percent for hotels, 70 percent for cruise lines, and 64 percent for car rentals. Adobe says rich visual content may inspire travelers, while detailed text is easier for language models to read. The useful question for airlines is whether fares, routes, flight details, and service information are machine-readable. Across both stories, watch the control points: who gets routed to the right model, and who gets surfaced as the right answer. For operators, the practical work is to measure those paths separately and make the underlying usage, product, and content data legible enough to compete.
