The most underestimated middleman business in the AI economy: no model training, no app — just move tokens from upstream APIs to users, and quietly handle hundreds of millions in volume.
No training. No app store. Just smart plumbing between API providers and end users.
Buy wholesale API credits, sell retail. Volume is the moat — the more traffic you aggregate, the better your upstream rates.
Charge monthly for dashboards, API keys, usage analytics and team management on top of raw token access.
On-premise gateways for companies that can't route through public APIs due to compliance or latency.
Pre-built packages tuned for education, law, or customer service — the same token, packaged for a niche.
The business looks simple — until one of these breaks.
Lose your only API provider and you lose every customer overnight. Diversify or die.
A race to the bottom you can't win against OpenAI's balance sheet. Compete on reliability, not price.
Stolen keys and resold accounts end in chargebacks, bans, and frozen payouts.
This is a field guide, not a product. It documents a business model that's already happening — mostly in Chinese, mostly under the radar.
It walks through how OpenRouter runs ~$100M in volume on a thin 5% cut, why open-source gateways like One-API hit 30k+ stars with almost no marketing, and the four revenue streams plus three traps that define the space.
If you're building in AI infra or just curious where the money actually flows, this should save you a few months of digging.