Skip to main content

Mining in AETRON

Mining in AETRON is not about solving hashes. A miner runs real AI work inside a Neuronet, serving requests and producing results that the network verifies cryptographically. Emission is paid for proven, useful work, not for raw computation spent on puzzles.

What a Miner Does​

A miner takes on two jobs at once:

  • Execution. It runs the model, serves inference or training requests, and produces a result along with a proof that the computation ran exactly as declared.
  • Verification. It re-checks a portion of other miners' work. The protocol assigns which results get re-executed, and a miner's own rewards depend on doing this verification when asked.

There is no separate validator role for AI work. Miners verify each other, and the protocol enforces it through the reward rules. See Proof of Intelligence for how verification works, and Inference Verification for what a re-check actually compares.

What Miners Work On​

Depending on the Neuronet's tasks, a miner might:

  • Generate or process text
  • Generate or classify images and video
  • Recognize or synthesize audio
  • Serve model inference on demand
  • Run training steps for a model
  • Power AI agents that use tools

Joining a Neuronet​

There are no fixed seats. AETRON has no UID slots and no cap of miners per Neuronet, so a new miner is not forced to evict an existing one. It joins whenever it meets the requirements:

  • Registration. Joining costs a small AET fee. The fee is burned and recycled rather than held as a deposit, and it rises briefly during bursts of registrations before decaying back down.
  • Proof of compute (Pulse). A miner must clear the Neuronet's minimum Pulse score, which proves it has real hardware capable of the work. Pulse is hardware-agnostic: it measures throughput, not a specific GPU brand.
  • Canonical settings. The miner must run the exact execution setup the task declares (model, precision, and so on) so its results can be verified against others.

For standard public Neuronets, miners join on their own once they meet these requirements. For private Neuronets that run in Confidential (TEE) mode, the Owner approves, invites, or removes miners directly.

Heterogeneous Hardware​

A miner does not need a specific GPU. AETRON verifies work across NVIDIA, AMD, Apple Silicon, and CPU within the same Neuronet, so data-center cards, workstations, and consumer machines can all mine, as long as they clear the Pulse threshold and run the canonical settings. See Heterogeneous Mining for the measured tolerances behind this, and Execution Spec for what "canonical settings" covers.

How Miners Earn​

A miner's share of emission grows with the work it can prove, not with a subjective score. The main factors are:

  • Verified work. Only results that pass re-checks count toward emission. Fraudulent or unverifiable work earns nothing.
  • Proof of compute. A higher Pulse score reflects more capacity to contribute.
  • Uptime and replay-pass rate. Being available and consistently passing re-checks keeps rewards full.
  • Reputation. Reputation falls when work fails verification or duties are missed, which reduces rewards.

A miner also needs a Neuronet with spare capacity for its work to earn, since work beyond a Neuronet's staked capacity pays nothing. Well-staked Neuronets with real demand are where mining pays off. See Staking for how capacity works.

Penalties​

Honest mining is the only profitable strategy, and Attack Resistance covers why the shortcuts do not pay. The protocol responds to bad behavior on a ladder:

  • Missing assigned verification or going offline reduces rewards and can lead to a temporary ban.
  • Producing work that fails verification (fraud) or falsely accusing an honest miner leads to a temporary ban first and a permanent ban on a repeat.
  • Penalties lower a miner's reputation; some fade over time, while fraud-related flags are permanent.

A temporary ban keeps the miner's identity and membership and lifts on its own, so an honest miner recovering from downtime is not pushed out for good.

Strategies for Successful Mining​

  • Match your hardware to the task. Pick Neuronets whose model and Pulse requirements fit what your machine can run.
  • Choose Neuronets with real demand and capacity. Your work only earns where there is traffic and staked capacity to cover it.
  • Stay available and pass re-checks. Uptime and a clean replay-pass record protect your reputation and your rewards.
  • Do your verification duties. Re-checking peers when assigned is part of earning, not an optional extra.

Mining in AETRON rewards real, verifiable contribution. The work you prove is the work you get paid for.

Where to Go Next​

Last updated on