AI Horde turns spare GPUs into free inference

The clever part here is the incentive design: AI Horde turns spare hardware into shared infrastructure instead of a one-way donation pool. That makes it a genuinely useful community layer, but the volunteer model also guarantees uneven latency and availability.

• –The OpenAI-compatible proxy at [oai.aihorde.net](https://oai.aihorde.net) is the fastest path for teams that already speak OpenAI-style APIs.
• –Contributors earn kudos, so supply and demand are tied together instead of separating donors from users.
• –Because workers choose what to serve, model availability is decentralized by design, which is great for openness and bad for strict reliability.
• –The open-source, self-hostable stack at [aihorde.net](https://aihorde.net) makes it appealing for communities, demos, and forkable deployments.