Stripe and Ramp Are Betting on AI Traffic Controllers

Stripe is known for processing payments. Ramp is known for managing corporate expenses. But this week, the two companies made almost identical moves: building the layer that decides which artificial intelligence (AI) model gets used.

Ramp launched Router.com on Wednesday (Aug. 19), opening up a tool it had run internally for three years. A company can connect an application programming interface (API) to Router instead of directly at OpenAI or Anthropic, and Router picks which model handles each request based on cost and whether that model is good enough for the job, Ramp said in its announcement. Internally, Router has cut Ramp’s own AI costs by about 30%, and the company said early customers are cutting costs by an average of 40%.

Stripe made a bigger version of the same bet, agreeing to buy OpenRouter, an independent startup, in a deal reported to be worth at least $7.5 billion, according to a Wednesday report by Silicon Angle. While Ramp built its router to manage its own workflow and is now selling it, OpenRouter was built as a public marketplace: developers connect once and reach more than 400 models from over 80 providers, instead of setting up a separate connection to each AI company. OpenRouter processes more than 10 trillion tokens a day and serves more than 10 million developers, Silicon Angle reported.

A Router Handles What No Single Model Can

The logic behind both deals is the same. The cost of solving the same task, writing a line of code or addressing a support ticket can differ enormously depending on which AI model handles it. As companies run AI across millions of requests, the the processing costs of those requests escalate.

A router sits between the app and the AI companies to fix that. Instead of an app being locked into one model, Router.com and OpenRouter each check every request and send it to the cheapest model that can still do the job well, and both switch to a backup model if a provider goes down, according to Ramp’s own product page and reporting on OpenRouter’s platform. Ramp built its own testing system, using real engineering work at the company, because public rankings of AI models didn’t answer the specific cost questions Ramp needed answered, according to Ramp’s own engineering blog.

Both Companies Are Building a Cost-Control Layer for AI

What Stripe and Ramp are building resemble the discipline companies already use to control cloud-computing costs, deciding in real time whether a task needs an expensive tool or a cheap one, instead of buying one fixed amount of capacity in advance.

Stripe already processes payments for many AI companies and has spent the past year building tools to help businesses track what they spend on AI, according to Stripe’s own announcement. Buying OpenRouter gives Stripe a direct view into which models developers actually pick and what they pay for them.

Ramp’s version sits inside the same platform companies already use to track software and employee expenses, and extends that same visibility to AI. Both companies are betting that as businesses spread their AI work across many models instead of picking just one, the more valuable business won’t be building the smartest model. It will be deciding which model gets picked next.

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.

The post Stripe and Ramp Are Betting on AI Traffic Controllers appeared first on PYMNTS.com.