← Blog

The model access layer just got a $7 billion price tag

OpenRouter proved model routing is real infrastructure. Choosing intelligence is not the same as giving an agent a place to work.

August 17, 20262 min readAgent Anything
openrouterinfrastructureagent fleetattexaai inference

Bloomberg reported on August 16 that Stripe reached an agreement to acquire OpenRouter for more than $7 billion, just three months after OpenRouter raised its Series B at a $1.3 billion valuation.

OpenRouter gives developers one interface for reaching models from OpenAI, Anthropic, Google, Meta, DeepSeek, and many others. It handles access, routing, and payment across hundreds of models from dozens of providers.

That is a real number, and it says something true about the market: choosing a model, paying for it, routing around provider constraints, and switching without getting locked in are not convenience features. They are infrastructure.

We have used OpenRouter since a couple of months after it launched in early 2023, back when model routing was not yet an established category. Today, it is one of several inference providers available through Agent Anything, not the only one, by design.

Picking a model answers one question: which intelligence handles the request?

It does not answer where an agent runs, what organizational resources it can reach, what state persists after the request, what boundaries apply while it works, or what can be proven afterward about consequential actions. Those are different questions. OpenRouter was never trying to answer them, any more than a great payment processor is expected to also run your warehouse.

That is the layer we have been building.

In our stack, Agent Anything is where people and AI coordinate around context, knowledge, and work. Agent Fleet is where delegated agents go to work. Attexa Verified Actions governs consequential actions at the boundary, and Attexa Witness makes the resulting evidence independently verifiable.

Agent Fleet gives an agent an isolated machine with a durable workspace, a declared model route, a network posture that can be narrowed or removed by policy, explicit access to the data and services it needs, and an inspectable record of its execution, outputs, and reported completion.

The model can change. The provider can change. The workplace can remain.

OpenRouter makes intelligence swappable. Agent Fleet makes the workplace durable.

Or, put more simply: your AI should have an address.

The market just made one previously invisible layer of AI infrastructure visible. Model access matters. But access is not execution.

Once an organization chooses the intelligence, it still needs a place for that intelligence to work: somewhere isolated, persistent, bounded, and accountable.

That is Agent Fleet.

That is the part we have been building.

Get in touch

Want to talk through how this applies to your organization?

Request access