← Blog

2026-03-10 · 1 min read

Listing OpenAI-compatible GPUs as a provider

Expose vLLM or a compatible proxy, attach credentials, go live, and sell capacity through Air Inference.

What buyers expect

Developers want a stable OpenAI-compatible /v1/chat/completions (and usually /v1/models). If you already run vLLM, llama.cpp server, LocalAI, or a RunPod OpenAI wrapper, you can list that endpoint on Air Inference.

Provider checklist

  1. Create an account with the provider (or both) role.
  2. Draft a listing — models, package price, token allotment, region.
  3. Save encrypted credentials (base_url + upstream API key).
  4. Only then set status to Live — the product blocks live without credentials.
  5. Respond to RFQs when buyers need custom quotes.

Keeping listings healthy

Stale upstream keys cause buyer 502s. Rotate keys in the dashboard when you rotate them upstream. Ops may force-pause listings that fail health probes.

Economics

Buyers pay Stripe to the platform. In v1, providers are paid off-platform; the documented fee is about 10% of GMV. Track what you are owed from paid orders and settle outside the app until Connect ships.

Ready to list? Start from the provider docs or open Listings after signup.