OpenRouter Raises $113M Series B at $1.3B as Token Volume 5x's

OpenRouter, the AI model exchange that sits between applications and the major model providers, announced a $113 million Series B led by CapitalG — Alphabet's independent growth fund. The round values the company at roughly $1.3 billion, more than double the $547 million it commanded a year ago.

The news arrives alongside a striking growth stat: OpenRouter is now processing 25 trillion tokens per week, a 5x jump from roughly 5 trillion tokens just six months ago. That works out to more than a quadrillion tokens a year flowing through a single inference router.

Why a model router is suddenly a $1.3 billion business

When OpenRouter launched, the company was effectively a hobbyist marketplace for trying different chatbots side by side. That has changed. As enterprises moved from pilots into production over the last year, "which model do we use for what?" turned into a real infrastructure problem.

OpenRouter sells a single API and dashboard that fronts more than 400 models — including Anthropic, Google, OpenAI, xAI, and DeepSeek — and handles failover, cost routing, rate limits, and model selection automatically. Teams can swap a cheaper or faster model in for a workload without rewriting their app.

That positioning matches where the market is going. A growing number of enterprises now run multi-model stacks instead of standardizing on one provider, partly to negotiate prices and partly to avoid being caught flat-footed when a competitor ships a better model.

What's in the round

Beyond CapitalG, the cap table reads like a who's-who of the AI build-out. Participants include NVentures (NVIDIA's venture arm), ServiceNow Ventures, MongoDB Ventures, Snowflake Ventures, and Databricks Ventures, alongside earlier backers Andreessen Horowitz and Menlo Ventures.

The strategic logic is straightforward: each of those data and infrastructure companies needs to plug AI inference into its products, and OpenRouter is the layer that lets them swap models without picking a single horse. NVIDIA's involvement is the most pointed signal — the company benefits whenever inference volume grows, regardless of which model wins.

OpenRouter says it now serves more than 8 million users worldwide, from AI-native startups to large enterprises.

What's driving the 5x volume surge

The token surge mirrors what the rest of the inference market is seeing. Anthropic disclosed last week that its annualized revenue had climbed past $30 billion, and OpenAI is tracking around $24 billion in ARR. Enterprises that started 2025 testing chatbots are now running coding agents, document pipelines, and customer-facing workflows that burn through tokens at industrial volume.

The shift to agentic systems matters here. A single agent task — like writing code, browsing the web, or filling out a spreadsheet — can consume hundreds of thousands of tokens across multiple model calls. That's pushing total inference demand up faster than the price-per-token is coming down.

For OpenRouter specifically, agent traffic is a tailwind because routing logic gets more valuable the more model calls a workload makes. Teams want the cheapest model that can do the job, and they want to fail over to a backup when their primary provider rate-limits them or goes down.

Industry impact: the inference middle layer is now a category

OpenRouter's raise validates a category that didn't really exist 18 months ago. AI model routing, inference governance, and multi-model observability are now their own line items in enterprise AI budgets. Competitors and adjacent players — Portkey, LiteLLM, Vercel's AI Gateway, AWS Bedrock — are all racing for the same middle layer.

The bigger implication is for the model providers themselves. As routing layers mature, enterprises can switch between Anthropic, OpenAI, Google, and open-weights models with a config change. That puts real pressure on model providers to compete on price and reliability, not just frontier benchmarks.

What's next

OpenRouter says it will use the new capital to expand routing, governance, and optimization capabilities — and to keep adding models. The company is betting that the number of useful models per enterprise will keep going up, not down, as specialized models for code, vision, voice, and reasoning continue to ship.

For developers building on top of LLMs, the practical takeaway: routing through OpenRouter (or a competitor) is increasingly cheaper and more resilient than hardcoding a single provider. The bet on AI being a "single winner" market is looking weaker with every release cycle.

Bottom line

OpenRouter's Series B is less a story about one company and more about how enterprise AI is actually being deployed. A $1.3 billion valuation for a model router would have looked strange two years ago. With 25 trillion tokens flowing through the platform every week, it now looks like the picks-and-shovels play of the inference era.