Run a Node
Two roles for contributing useful work to the network. Worker payouts are live; validator rewards are not.
| AI Worker Node | Validator Node | |
|---|---|---|
| Status | Live | Public evidence-only preview |
| Stake | Not required; future bond amount/policy not finalized | Not required; validator staking is disabled |
| What you do | Serve generation jobs (chat, image, video) | Verify that workers’ results are honest |
| Hardware | GPU required (4 GB VRAM minimum, 12+ GB recommended) | Modest CPU + bandwidth; GPU optional for re-execution |
| Earnings | Current hourly AIPG budget, pro-rata to den | None in the preview |
| Risk | None today; future bond risk for objective fraud | No routing, reward, strike, stake, or slash authority |
Worker bonding and validator staking are part of the Base-backed roadmap. The current worker path does not require stake, and the first validator release is evidence-only.
AI Worker Nodes (Live)
If you have a GPU sitting idle, you can put it to work. AI worker nodes pick up generation jobs from the network, run the model or workflow, and return the result. The current hourly rail pays AIPG pro-rata to recorded den work units.
Two flavors:
You can run both on the same machine if you have the VRAM, or pick one based on what your hardware does best.
The worker bond (planned, not yet required)
Running a worker today does not require a stake — connect with an API key and
start serving. A bonding mechanism is in development: workers will stake to the
WorkerRegistry Diamond facet. The amount and rollout policy are not finalized.
This would be collateral, not a fee.
As long as you serve honest results, your stake stays untouched and you can
withdraw it whenever you stop running the worker.
Once the bond ships, validator evidence can first lower routing trust and caps. Slashing should come later, only for objective fraud with assignment-bound evidence, quorum, and a dispute path.
What you earn
The live rail distributes a fixed AIPG period budget pro-rata to den, so there is no guaranteed fixed token amount per request. Your share depends on:
- Model size (bigger models pay more per token)
- Throughput (faster workers serve more jobs)
- Quality (workers with clean attestation history get preference)
- Confidential workloads pay a premium (see Confidential Computing)
Payout history is visible on console.aipowergrid.io/payouts. A USD-denominated, multi-asset payout rail is designed in core but is not active; do not select USDC/ETH/USDS expecting the current payout CLI to honor it.
Validator Nodes (Public Preview)
If you want to help secure the network rather than provide compute, validator nodes are the audit role. The first release is intentionally low-authority: validators run unpredictable probes, sign evidence, and feed dashboards while the Grid hardens targeted assignments, reference checks, and quorum.
Authority: evidence only. Validator rewards, routing influence, staking, strikes, and slashing are disabled. The preview is for proving independent operation and hardening the evidence protocol before economics are attached.
Hardware: Modest. Validators don’t need to serve every job, so they don’t need worker-grade GPUs. A reliable server with stable bandwidth is enough for sampling. Validators who want to re-execute jobs themselves (rather than rely on cross-validation) can add GPU capacity.
The operator path mirrors the worker quickstarts: install the verified preview,
create a local signing wallet, link its public address in Console, create a
validator-purpose API key, run aipg-validator init, check the Grid connection,
and start the validator loop. The private key remains on the validator host.
→ Download and run the validator
→ Protocol, threat model, and preview limits
Get AIPG to Stake
Token: 0xa1c0deCaFE3E9Bf06A5F29B7015CD373a9854608 (BaseScan)