Ocean Network

Ocean Network runs a peer-to-peer GPU cloud, pooling graphics cards held by individuals and small teams into a compute market and renting them by the hour, with the whole card dedicated to one tenant for the booked window rather than split across users.

Billing follows the reserved window, with H200 capacity from $2.16 an hour at a price fixed before the job starts and no charge once the window ends. Inference is served through OpenAI-compatible endpoints with models loaded by name, and image and video workflows ship as templates with weights already in the image so a first run does not begin with a multi-gigabyte download. An MCP interface lets coding agents pick hardware, submit containerised jobs and collect results, while owners of idle cards can register as nodes and earn per minute of use. A beta version shipped in March 2026.