// intro · overview

Overview

VouchGPU is a verifiable decentralized GPU compute network, anchored to Solana for instant settlement and on-chain attestation of every inference cycle.

updated 2026-05read 3 minsection · intro

What VouchGPU is

VouchGPU is a peer-to-peer GPU compute mesh designed for AI workloads. Renters get verifiable compute at consumer prices. Hosts plug in idle silicon and earn SOL for every proven cycle the network routes through their machine.

Unlike centralized GPU clouds (AWS, RunPod, Vast), every cycle on VouchGPU is attested on-chain. Validators re-execute and sample work. If a node lies about a forward pass, the chain catches it before the next block.

The three sides of the network

Renters — pay only for verified cycles. OpenAI-compatible inference endpoints. Spot or reserved pricing.
Hosts — bring your own silicon (RTX 4090, A100, H100, M3 Max). Post collateral, get routed jobs, earn SOL every epoch.
Validators — stake $GPU, re-execute samples, slash bad actors, collect protocol fees.

Why on-chain attestation matters

The trust problem in GPU compute

Centralized GPU clouds bill you for utilization they alone measure. You have no way to verify what actually ran on your GPU minutes. VouchGPU solves this by attesting every cycle on a Solana-anchored L1 — payouts settle the same epoch, only on proven work.

What you can deploy

SSH terminals on dedicated GPU silicon
Pre-built LLM stacks (Llama 3, Mistral, Qwen via Oobabooga)
JupyterLab, PyTorch, TensorFlow notebooks
Stable Diffusion / ComfyUI image generation
Whisper transcription pipelines
LangChain agents with vector store

Mainnet status

The network is in mainnet-beta. Hosts are onboarding. Renters can browse the mesh, deploy instances, and pay in SOL. Token generation event ($GPU) follows the V1 hardware ship.

Tokenomics