Rent compute
Launch an isolated NVIDIA workspace.
Choose a compatible public OCI image, fund the maximum lease and connect through temporary SSH or Jupyter credentials.
Independent GPU infrastructure
Prism connects renters to independent NVIDIA capacity through isolated workspaces, metered USDG escrow and a public settlement trail.
01 / Network
Raw GPU workspaces come first. Batch containers and managed inference build on the same node identity, access and settlement lifecycle.
Rent compute
Choose a compatible public OCI image, fund the maximum lease and connect through temporary SSH or Jupyter credentials.
Supply capacity
Enroll an NVIDIA host, publish a signed offer and receive 90% of confirmed usage after settlement.
02 / Settlement
Maximum lease cost is locked in USDG escrow before provisioning.
Compatible independent nodes are ranked by price, performance and reliability.
The GPU enters an isolated Kata VM-backed workspace with temporary access.
Billing begins only after CUDA and interactive access are confirmed ready.
Confirmed runtime becomes an onchain proposal, refund and public receipt.
03 / Isolation
Each lease receives exclusive GPU passthrough into a VM-backed container. Supplier addresses stay private, workspace keys are temporary and host egress policy blocks private networks and metadata endpoints.
Permissionless suppliers are not trusted computing environments. Prism rejects confidential and sensitive workloads until independently attestable confidential-GPU nodes exist.
04 / Proof
Finalized receipts connect platform-attested usage records to Robinhood Chain settlement events without exposing terminal contents, notebooks or files.
Open proof feed ↗GPU infrastructure should account for every second.