OpenRouter just raised $113 million at a $1.3 billion valuation, more than doubling what it was worth a year ago. That’s not the interesting part.

The interesting part is what CapitalG, Alphabet’s growth fund, is betting on: a world where developers stop choosing between AI models and start routing between all of them.

OpenRouter is infrastructure for hedging. It gives developers a single API that connects to dozens of models from OpenAI, Anthropic, Google, Meta, and others. You send a request, OpenRouter finds the cheapest or fastest model that can handle it, and you get a response. No vendor lock-in, no rewriting code when a better model ships next month.

Usage grew five times over the past six months. That’s the kind of curve that makes investors write checks, but it also says something about where the industry is. Developers aren’t waiting to see who wins the model race. They’re building on the assumption that there won’t be a winner.

This isn’t a new idea. Model routers have been around for a while. But OpenRouter’s traction suggests the market has shifted from “maybe we should abstract this” to “we can’t afford not to.” When models change every few weeks and pricing swings with new releases, writing directly against one provider’s API starts to look like technical debt.

It’s also a hedge for Google. CapitalG led the round, which means Alphabet is funding a startup that makes it easier to use competitors’ models. That’s either very confident or very pragmatic. Probably both. If routing becomes the default way developers interact with AI, Google wants a stake in the pipes.

The broader pattern

OpenRouter isn’t the only company riding this wave. Martian, another AI router, raised $9 million in March. LangChain and LlamaIndex both have routing features. Every major AI developer tool now includes some version of “use any model.”

The question isn’t whether routing will be common. It already is. The question is who captures the value. Does it go to standalone infrastructure companies like OpenRouter, or does it get absorbed into larger platforms?

Right now, OpenRouter has an advantage: it’s neutral. Developers trust it because it doesn’t have a horse in the race. That’s worth something, maybe $1.3 billion worth, but it’s also fragile. If AWS or Azure launches a comparable service and bundles it into their existing contracts, neutrality stops being a moat.

What this means for model providers

If routing becomes standard, the model providers face a different game. Instead of locking developers into ecosystems, they’re competing on performance and price for every single request. That’s good for developers and probably good for innovation, but it changes the economics.

OpenAI’s advantage isn’t just that GPT-4 is good. It’s that millions of apps are hardcoded to use it. If those apps start routing requests to whatever model is best today, OpenAI has to stay best, or cheaper, or both. Same for Anthropic, Google, and everyone else.

It also means the big model labs can’t rely on API revenue growing in lockstep with usage. If developers route to the cheapest model that works, margin pressure gets worse, not better.

CapitalG’s bet is that this is inevitable, and that the routing layer will be valuable enough to justify a billion-dollar valuation. OpenRouter’s bet is that developers would rather pay for flexibility than commit to a single vendor.

Both bets assume the same thing: the future of AI development is multi-model by default. OpenRouter’s growth over the last six months suggests that future is already here.