# Stripe Bought the Routing Layer > Source: > Published: 2026-08-17 08:18:36+00:00 Stripe just acquired OpenRouter, and everyone's talking about the number. Big deal. But the real signal is what layer they're buying into. OpenRouter is the router. It sells developers a single door into more than 400 artificial intelligence models. A customer writes to OpenRouter once, then swaps between OpenAI, Anthropic or a cheaper open-weight alternative without touching the code, with the router picking on cost, speed or which provider happens to be up. It's abstraction. It's portability. It's the thing that keeps any one model vendor from owning your application. Stripe just bought it. And in doing so, made a choice about which piece of the AI stack is worth owning. For a long time, the winner-take-all bets were on the models themselves. Bigger models, better models, frontier models. That's where the capital went, where the companies fought, where the strategy lived. But the smarter move might be one step back. Not the model. The chooser. If you control the router, you control which model gets used. You see every inference decision. You know what's happening. You own the switching mechanism. And more importantly: you get to decide what the default is. OpenRouter takes about 5% of the inference spend that passes through it. That's real money, and it compounds. What's interesting is that Stripe is a payments company. They built the infrastructure for moving money online. Now they're buying the infrastructure for moving inference. Those aren't unrelated instincts. Payments sits between buyer and seller. Stripe sits in the middle. Now Stripe will sit in the middle of model selection too. They're not becoming an AI company. They're becoming the intermediary. This matters because model consolidation was starting to feel inevitable. A few shops would win (OpenAI, Anthropic, maybe Google and Meta). Developers would get locked in. But if OpenRouter stays genuinely open, truly vendor-agnostic, under Stripe ownership, it changes the game. You can use whatever's best without being hostage to any vendor's roadmap or pricing. You can write once to the router and swap implementations on the backend. The risk is obvious: Stripe could decide that some models are preferred. They could favor certain providers. They could use the router to push their own future AI ambitions. The routing layer becomes a chokepoint instead of a lever. For now, what matters is that Stripe made the bet. Not on building the model. Not on being the lab. On being the thing in the middle. The layer that matters most isn't always the loudest one.