AI Gateway Market Splits Into Three

This is no longer one market, it is three businesses that happen to sit in the same request path. OpenRouter wins when a developer wants the cheapest useful model from a giant menu and is fine buying tokens through a broker. Vercel wins when an app team already lives inside its developer platform and wants one API key, one bill, and built in observability. Kong wins when a large company needs the gateway to sit inside its own walls and control which employee or system can call which model, at what cost, with logs, limits, and redaction.

• OpenRouter looks most like a wholesale exchange. Its core job is aggregating hundreds of models, routing traffic to the best priced provider, and taking roughly a 5% fee on token spend. That is why cheaper Chinese open source models can expand volume fast while lowering revenue per token.
  1 sacra 3 sacra 5 sacra
• Vercel is packaging gateway functionality into an existing app platform. A startup building on Next.js can use AI Gateway for model access, routing, and observability without adopting a separate infra vendor. The gateway is part of a broader developer workflow, not the whole business on its own.
  2 sacra 4 sacra 5 sacra
• Kong sells to enterprises that care less about public model discovery and more about internal control. Its gateway handles semantic routing, caching, prompt compression, token limits, and sensitive data filtering across many internal LLMs. That makes the buyer the CIO or platform team, not an individual developer swiping a card.
  1 sacra 4 sacra

The next step is deeper separation, not reconvergence. Wholesale marketplaces should keep pulling in developer traffic as token buying gets more price sensitive. Public gateways should become default features of developer clouds and coding platforms. Enterprise gateways should grow into policy and metering layers for agent traffic, where the real value is controlling cost, access, and auditability across thousands of internal users and systems.