OpenRouter pivoting to brokerage economics
OpenRouter growing 29% MoM at $140M/year
This shows OpenRouter is becoming more like a high volume exchange than a premium software product. The business is processing far more tokens because developers are shifting easy and medium difficulty work onto much cheaper open source models, especially from Chinese labs, but OpenRouter only captures a small percent of token spend, so cheaper tokens mean lower revenue per token even while traffic explodes. OpenRouter wins when it becomes the default place where developers compare price, switch models, and keep usage flowing.
-
OpenRouter’s model is close to brokerage economics. Its pricing passes through the underlying model cost and charges about a 5% to 5.5% platform fee, so when usage shifts from expensive frontier models to lower priced open source models, OpenRouter can grow volume much faster than monetization per token.
-
The throughput jump fits a broader routing shift across the stack. Vercel turned AI Gateway into a general availability product in August 2025, Databricks now positions Unity AI Gateway as a control plane for routing and spend, and Ramp says its internal router cuts AI cost by more than 25% while handling trillions of tokens per day.
-
That makes OpenRouter less like a model lab and more like Costco for tokens. It aggregates 400 plus models and 70 plus providers behind one API, while enterprise gateways like Kong focus on internal guardrails, caching, and routing employees to the cheapest acceptable model instead of selling discounted external supply.
The next step is a market where routing logic matters more than any single model. As more apps automatically send hard tasks to the best model and easy tasks to the cheapest one, OpenRouter’s upside is owning the transaction layer for that flow, with scale, pricing visibility, and developer distribution becoming more important than high per token take rates.