— GPU NETWORK · AI COMPUTE ON ROBINHOOD CHAIN
GPU NETWORK is compute infrastructure paid for by usage. A fixed share of the fees the network produces flows into the GPU Reserve, which leases accelerator time and hands it back to participants as Compute Credits.
CONTRACT ADDRESS · $GPUNETWORK
Reserve design live · credits open to participants at launch
Core concept
Four modules / usage-funded / compute-first
THE SOURCE
Every swap, transfer and interaction on the network leaves value behind. GPU NETWORK stops that value from pooling somewhere idle and points it at accelerator time instead, so the system gets stronger the more people use it.
THE POOL
A fixed slice of protocol fees streams into the GPU Reserve without anyone pressing a button. The reserve is not a treasury waiting for a plan — it has one job, and that job is buying compute.
THE HARDWARE
The reserve leases GPU hours from screened datacenter partners — physical cards with measurable throughput. Pooling demand lets it book capacity at rates a single buyer rarely gets.
THE PAYOFF
Leased hours come back to eligible holders and participants as Compute Credits: metered GPU time you can actually run jobs on. The people who keep the network busy are the ones it serves.
How it works
CAPTURE
A preset share of those fees is skimmed by the protocol on every cycle. The rule is public, the split is fixed, and nothing waits on a manual decision.
SEE THE GPU GRID →STAGE 02 OF 05
The reserve
| LOCATION | REGION | GPU · UNITS | LATENCY | TARGET SHARE |
|---|---|---|---|---|
| Oregon | NA-WEST-01 | H100 · 2,400 | 14ms | 70% |
| Virginia | NA-EAST-01 | H100 · 3,000 | 19ms | 86% |
| Dublin | EU-WEST-01 | A100 · 2,600 | 26ms | 62% |
| Frankfurt | EU-CENT-01 | A100 · 1,800 | 29ms | 51% |
| Singapore | AP-SE-01 | L40S · 2,100 | 34ms | 77% |
| Mumbai | AP-SOUTH-01 | L40S · 950 | 47ms | 40% |
SHOWING 6 OF 17 TARGET REGIONS · PROCUREMENT PLAN, NOT LIVE TELEMETRY
Live metrics
The network
Trust & security
Each job gets its own container on hardware it does not share mid-run. Nothing leaks between tenants and nothing lingers after the job ends.
The GPU Reserve can only do one thing: lease accelerator time. Fees in, GPU hours out, under a rule anyone can read.
Fee captures, leases and credit grants are recorded so every hour of compute can be followed back to the activity that paid for it.
Access is tied to your wallet on Robinhood Chain. No account passwords, no custody — connecting only shares an address, never a key.
Compute credits SDK
A small client library that turns Compute Credits into running GPU jobs. The aim: from wallet to first job on real accelerators in under ten minutes.
Typed from the start
Every call and response ships with TypeScript definitions.
Credit-metered
Credits are drawn per second of GPU time — nothing billed while idle.
Bring any framework
PyTorch, JAX, TensorFlow, vLLM — point it at your own image.
Nearest-region routing
Jobs land in whichever region has free cards closest to you.
1# npm2npm install @gpunetwork/sdk34# pnpm5pnpm add @gpunetwork/sdk67# bun8bun add @gpunetwork/sdk
Use cases
02 / 04
“The rule is easy to explain: usage feeds the reserve, the reserve leases GPUs, and participants get the hours. No abstraction layer in between.”
Protocol design
How the loop works · SCENARIO_02
Key figure
100%
FEES MEANT FOR COMPUTE
Scenarios are illustrative examples of how credits could be used, not customer statements.
GPU RESERVE · $GPUNETWORK
Join the network that converts participation into real AI GPU time. Connect a wallet, collect Compute Credits, and run jobs on leased hardware as the reserve grows.
