Compute providers · Beta
Community GPU network
Alongside the major model providers, Gatewayz is building a second side of supply: independent GPU operators serving inference. It's in beta, and it's built on the assumption that no node can be trusted by default.
Untrusted by design
A community node is someone else's machine. So the network doesn't take a node's output on faith: work is checked before it counts.
Spot-check replay
A sample of requests is selected before they're forwarded, then replayed against the node to confirm it served what it claimed. Because prompts aren't stored, the sample is set aside at the moment of routing.
Track record decides the rest
Most work isn't replayed individually. It's accepted only if the node's recent checks have a low failure rate, and a node that fails checks repeatedly is disabled.
What operators should know
- The network is in beta. Requirements, terms and how work is accounted for are on the provider page.
- A community node sees the content of the requests it serves. That's inherent to serving inference, and it's why verification and the terms matter.
- Network activity — nodes, utilization and model coverage — is shown on the public GPU dashboard.
Run a node
Read the requirements and terms before you connect hardware.