Decentralized GPU Infrastructure
Open inference network. Send OpenAI-compatible requests, route through a global mesh of consumer GPUs, settle rewards on-chain. Pay per token. Get paid for spare cycles. OpenAI-compatible, so your existing client just works.
The infrastructure already exists. It is just not being used.
Centralized bottleneck
Three hyperscalers own most inference capacity. Outages, rate limits, and capricious policy changes ripple to every dependent product.
Stranded supply
Hundreds of millions of consumer GPUs sit idle outside business hours. Energy paid for, capacity unsold.
Geographic gaps
Data-center clusters concentrate in a handful of regions. Latency in everywhere-else is a tax on every prompt.
CAPEX dead weight
Building a new GPU cluster costs years and billions before the first token returns. The world cannot scale AI on that curve.
Three steps from idle GPU to passive income
Install the node
One curl command. Detects a local Ollama install or starts a sandboxed echo mode for testing without a GPU.
Join the mesh
The provider registers with the orchestrator, advertises its models, and long-polls for tasks. Survives reconnects automatically.
Earn $GPU
Every completed inference accrues claimable rewards. Claim to your wallet whenever you want, no minimums, no intermediaries.
Built for inference at the edge of the network.
Drop-in for any SDK. Same /v1/chat/completions surface, same streaming chunks, same Bearer auth.
Orchestrator scores live providers on latency, model match, and load. Failed tasks re-queue with no client retry.
Sign a message with Phantom or Solflare to mint an API key. No emails, no passwords, no recovery flows.
Rewards accrue per inference and claim straight to your wallet. Treasury-signed today, on-chain program after audit.
Anyone can run a provider. No KYC, no allowlist. Models you ship and prompts you serve are entirely your call.
Site, orchestrator, and provider script are open and small enough to read in an afternoon. No magic.
One line to start providing. No git, no clone, no setup.
$ curl -fsSL https://gpunetwork.xyz/install | sh$ curl https://gpunetwork.xyz/v1/chat/completions \
-H "Authorization: Bearer dgi_xxx" \
-H "Content-Type: application/json" \
-d '{"model":"gpu-mesh","stream":true,"messages":[{"role":"user","content":"hello mesh"}]}'Inference, served from everywhere at once.
Provider nodes run wherever there's electricity and a GPU. Each pulse below is a fictional location — the actual mesh is permissionless, so the map fills in as people install the node software on their own machines.
Real numbers, not a coming-soon page.
Start earning $GPU with the GPU you already own.
Five minutes to first inference. No paid plan, no team approval, no minimum withdrawal.