Routing is coming for the frontier AI labs
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Aïda Amer/Axios
Businesses seeking more affordable and secure ways to use AI are turning to routing — the process of matching each query with the best, most efficient model for the task.
Why it matters: The popularization of routing could stymie the usage of the frontier AI models that are the most expensive to use and make.
State of play: Routing is booming.
- Stripe agreed to buy unicorn OpenRouter for more than $8 billion, the company announced last week.
- OpenRouter had 8 million users and 400 models available on the platform as of May, per Bloomberg.
- Meta is reportedly working on an OpenRouter competitor called Switchboard, per The Information, which the company may use to lower its own AI costs.
- Even the frontier AI labs have launched model pickers that allow users to select which model or level of intelligence they want to use for a certain task.
How it works: Routing companies say that it's extremely quick and easy to get started. (In a demo, I was able to use a DeepSeek model in under 30 seconds using TrustedRouter, an open-source routing company.)
- A simple query may call on a smaller flash model, while a complex task could go to a more expensive one.
- Companies can set their own priorities for the router, including cost, speed, performance and which models or providers they trust with their data.
Flashback: The routing boom was fueled by concerns about AI costs.
- Usage of open models, which can be cheaper, has skyrocketed: The five most popular models used on OpenRouter are all open weight or open source. OpenAI and Google each have only one model in the top 10, and Anthropic has none.
- Average token costs have plummeted since a peak in mid-May due to price cuts from frontier AI labs looking to gain back share from their competitors.
Zoom out: There's also a growing push for AI sovereignty, or protecting your enterprise IP and not sharing it with frontier labs, that routers aim to address.
- "I see people excited about us and moving to us because they don't like what Anthropic is becoming," Joseph Perla, founder of TrustedRouter, told Axios.
- TrustedRouter is encrypted end to end to make compute "confidential," Perla said.
- Palantir recently hosted a sovereignty boot camp in NYC packed with speakers, including one who said "Are we going to throw our years of knowledge to a frontier AI model? Absolutely not."
Reality check: People are not broadly arguing the frontier AI labs are going to die because of model routing.
- "It's not to say you won't ever use commercial models, but it's like you want to have the options to use models that are specific to you," Akshay Krishnaswamy, chief architect at Palantir, told Axios.
- Demand for intelligence is such that there can be multiple winners, because these platforms also "put their models in front of more users and more workloads," Matan Grinberg, CEO of Factory AI, a routing company, told Axios.
The bottom line: Routing turns models from destinations in their own right into interchangeable commodities, which could threaten margins for frontier AI models.
