OpenRouter hits 25T tokens a week as $113M backs model routing

AI infrastructure money is moving beyond the labs that train frontier models and into the layer that routes traffic among them. OpenRouter said on May 26, 2026 that it raised a $113 million Series B led by CapitalG, Alphabet's independent growth fund, with participation from NVentures, ServiceNow Ventures, MongoDB Ventures, Snowflake Ventures, Databricks Ventures, Andreessen Horowitz, and Menlo Ventures.

The sharper signal is usage. OpenRouter says weekly volume has climbed from 5 trillion tokens six months ago to 25 trillion tokens, or about 100 trillion tokens per month. Its platform gives developers and enterprises one API for more than 400 models from providers including Anthropic, Google, OpenAI, xAI, and DeepSeek.

That matters because production AI is becoming a routing problem. Teams no longer pick one model and leave the system alone. They shift traffic by cost, latency, capability, context length, reliability, and data-handling rules. Once agents begin making many calls in the background, those choices become part of the operating system for AI products rather than a one-time vendor decision.

OpenRouter also says it is used by more than 8 million global users, including AI-native startups and large enterprises. Its public rankings and usage data have become a proxy for real-world model adoption, because they reflect what developers actually call rather than what a benchmark table suggests. That gives the company a useful position between model makers, application builders, investors, and researchers.

The new capital will go toward routing, governance, and optimization. Those are not cosmetic enterprise features. A company running agents across multiple providers needs per-request policy controls, team-level permissions, spend visibility, failover, and audit-friendly reporting. Without that layer, the practical cost of using many models can erase the technical benefit.

The open question is whether OpenRouter can stay neutral while becoming more important to the market it measures. If model quality keeps changing quickly, the routing layer may become one of the most valuable control points in AI infrastructure. This round suggests investors are now pricing that layer as its own category, not just as developer convenience.