The era of picking one AI model and sticking with it is over. But running requests across a dozen different providers without losing your mind? That's a new kind of infrastructure problem, and it just made OpenRouter a $1.3 billion company.
What OpenRouter Actually Does
OpenRouter sits between your application and 400+ AI models from providers including OpenAI, Anthropic, Google, xAI, DeepSeek, and 60+ other labs. Instead of integrating separate SDKs for each one, you point your OpenAI-compatible client at https://openrouter.ai/api/v1 and pick any model by its full identifier.
The platform handles three things at once: a unified API gateway, a real-time routing engine, and a market intelligence layer that has become the de facto scoreboard for model adoption. If an AI model launches on Wednesday, you can see its usage share on OpenRouter by Friday.
By the Numbers
The growth curve is absurd even by AI infrastructure standards.
| Metric | Value | Growth |
|---|---|---|
| Series B | $113M led by CapitalG | — |
| Valuation | $1.3 billion | 2x+ since Series A |
| Weekly tokens | 25 trillion | 5x in 6 months |
| Monthly tokens | ~100 trillion | — |
| Annual run rate | ~1.5 quadrillion | 15x YoY |
| Users | 8 million | 3x from 2.5M |
| Team size | ~50 people | — |
| Revenue per employee | ~$2M net | — |
The Menlo Ventures blog puts these numbers in context: OpenRouter now processes more tokens than Microsoft Azure Foundry (over 50% of their run rate), and sits at roughly 15-30% of Google's and 20-40% of OpenAI's token volumes.
Revenue has doubled since the round was finalized in February, according to Menlo. That's $80-100M ARR by some estimates, a 13-16x multiple on the $1.3B valuation, which one HN commenter called "more of a business than 99% of AI companies."
The Investor Signal
The cap table tells you everything about how the industry sees this company. CapitalG (Alphabet's growth fund) led the round. But the participation list is even more telling:
NVentures (NVIDIA), ServiceNow, MongoDB, Snowflake, Databricks, Andreessen Horowitz, Menlo Ventures.
These aren't just financial investors. ServiceNow, MongoDB, Snowflake, and Databricks are enterprise platforms whose customers are deploying AI at scale. They want OpenRouter to exist because their own growth depends on the multi-model AI future being real.
Mo Jomaa, partner at CapitalG, compared OpenRouter's role to foundational infrastructure companies: "Every platform shift creates infrastructure gaps. OpenRouter is solving the infrastructure gap for inference in the AI era."
Andrej Karpathy called OpenRouter "the de facto transfer switch of AI today."
Competitive Landscape
OpenRouter isn't the only game in town, and the space is getting crowded fast.
| Layer | Players | Strength |
|---|---|---|
| Model router | OpenRouter, Inworld | Best model selection, developer UX |
| Self-hosted gateway | LiteLLM, Bifrost, RelayPlane | Full control, in-VPC deployment |
| Full-stack cloud | Modal ($4.65B, $355M raised) | Serverless compute + inference |
| Cloud gateways | Cloudflare AI Gateway, AWS Bedrock | Infrastructure bundling |
| Observability | Portkey, Maxim | Monitoring and governance |
Modal just raised $355M at a $4.65B valuation, a different bet (serverless compute for AI workloads) but in the same broad thesis that the AI infrastructure layer is where the value is consolidating.
The risk for OpenRouter is that hyperscalers bundle routing into their platforms. AWS Bedrock, Google Vertex AI, and Microsoft Azure AI Foundry all offer multi-model access. OpenRouter's edge is speed of iteration and breadth: 400+ models that no single cloud can match. But enterprise procurement still favors the bundle.
What Surprised Me
Two things stand out.
First, the team efficiency. Fifty people processing 1.5 quadrillion tokens a year, that's the same order of magnitude as Google and OpenAI's token volumes with a team smaller than most startup engineering departments. Alex Atallah, the CEO, previously co-founded OpenSea. This team knows how to build infrastructure that scales with zero friction.
Second, the revenue story. Doubling revenue between a closed February round and a May announcement suggests they're not just growing on hype. The numbers were already in hand when they went to market.
The broader point is one people keep missing: the AI industry is consolidating around infrastructure, not models. Every lab is releasing models that are "almost as good as GPT-5" or "beats Claude on math." The models are becoming commodities. The switching layer that decides which model runs which request at which price, that's the durable business.
OpenRouter's $1.3B valuation feels reasonable for a company on track to process more inference traffic than any single lab within 18 months. What feels unreasonable is that only 50 people are doing it.
Sources
- Business Wire / AFP: OpenRouter $113M Series B announcement (May 26, 2026)
- Menlo Ventures: "OpenRouter Now Processes More Than a Quadrillion Tokens a Year"
- TechStartups: "OpenRouter raises $113M as AI token usage surges to 100 trillion monthly"
- The Information: "Startup That Helps Developers Pick AI Models Nears $1.3 Billion Valuation"
- New York Times: "A One-Stop Shop for A.I. Models Raises $113 Million"
- LinkedIn / Stephan Palazzolo: OpenRouter funding details
- Hacker News discussion (various)
- Modal: Series C $355M announcement
- Deloitte: 2026 enterprise AI adoption study (67% of enterprises >1B tokens/month)
- RelvPlane: LLM Gateway Comparison 2026
- Sacra: OpenRouter revenue, valuation & funding