[AI Mining]

AI Mining with Useful Compute

A miner installs the AETRON client, and the client itself finds matching neuronets inside the network. From there on it's plain inference and model training, paid out in AET.

What mining looks like in AETRON

One client covers everything. It sees the compatible neuronets and plugs your GPU into them.

1

Download the client

A single file for Windows, macOS, Linux or Docker. The client identifies your GPU and its architecture on its own, whichever vendor you happen to run.

2

Connect to neuronets

The client walks through the available neuronets and keeps the ones that match your hardware. Out of those it picks the most profitable.

3

Earn AET

The client takes on requests and, on the side, re-checks the responses of other miners. Emissions arrive automatically based on the confirmed work.

How You Earn

1

AET emission (main income)

  • Run with the neuronet's requests. On top of that, the network assigns you re-checks of other miners' responses.
  • 80% of a neuronet's emission goes to its miners. How much of that share lands in your wallet depends on the confirmed work you delivered and on uptime.
  • Emissions are paid only for confirmed work. If you didn't close out the re-checks assigned to you, you get nothing for that epoch.
2

Income outside the protocol

  • The neuronet owner is free to sell an API on top of your compute to their own customers. That matters most where verifiability matters: healthcare or finance, for instance.
  • Part of that revenue the owner can pass to miners on top of the emission. The exact terms are set by the neuronet itself.
  • A separate case is the private mode with TEE attestation. There the customer is paying for verifiable confidentiality.
3

Restaking into a neuronet

  • AET you mined can go back into a neuronet as stake. That raises its throughput.
  • A staker's income tracks the size of their stake relative to the total stake in that neuronet.
4

Transparency and control

  • Every rule that governs emissions lives on-chain. The score is computed by a formula. There are no votes, no closed-door decisions.
  • Every work artefact, every re-check and every payout is written on-chain and can be verified.
  • The hardware, the set of neuronets and the restaking strategy stay entirely yours.

For data centers & GPU clusters

Scale earnings with enterprise-grade equipment

Multi-registration Hundreds of GPUs running in different neuronets at the same time, with no cap on slots or UIDs
Modern architectures supported All modern NVIDIA architectures, AMD CDNA3 and Apple Silicon work through the same client. No per-model integration is needed.
Revenue diversification Split the fleet between inference, training and private workloads. Inside the client you can see which neuronets are currently paying more
Professional infrastructure High uptime and stability translate directly into a larger share of the emission. Enterprise infrastructure earns more precisely because of that stability
Confidential Computing TEE attestation (NVIDIA CC and Intel TDX) opens up access to private models and enterprise workloads with premium terms
Why it beats classic mining:
Useful work instead of empty computationIncome from several sides: emission, restaking, and customers on top of the protocolDemand for verifiable AI keeps growing, and the network grows along with itVerification overhead is light, so most of the GPU time goes into earning

Mining Lifecycle

Register

Install the AETRON client and pay a small registration fee. As an alternative you can run a PoW puzzle instead. The reference checks and the benchmark are handled by the client itself.

Execute

The client serves the neuronet's requests according to its parameters: data type, quantization format, attention backend. After each response a small artefact is stored, which is what later lets the work be re-checked.

Artefact batching

Artefacts are folded into a Merkle tree. A single on-chain transaction covers up to a million requests. In parallel the client re-checks other miners' responses, around 7% of the network's traffic.

Earn

Emissions are paid out automatically. The more confirmed work and uptime you provide, the more you earn. A miner who loses arbitration on someone else's re-check gets no emission for that epoch.

New miners start with a lower reputation weight. It grows as the miner passes re-checks without errors. The exact thresholds are set by the community through governance.

Requirements & setup

Hardware

  • Any GPU on a supported architecture: NVIDIA Ampere/Ada/Hopper/Blackwell, AMD CDNA3, or Apple Silicon.
  • Your GPU's memory has to fit the target model. A 72B model in bf16, for instance, fits on a single B200 with 180 GB. Models like 7B, FLUX or Whisper run on consumer cards just fine.
  • Reliable power and cooling. Storage for models, datasets and checkpoints.
  • Solid internet with redundancy. Uptime feeds straight into your income.

Software

  • The AETRON client: one file for Windows, macOS, Linux or Docker
  • A bit of AET for the registration fee. Or the willingness to run a PoW puzzle instead
  • A compatible PyTorch and CUDA, ROCm or MPS. The environment is checked by the client itself on launch
  • Your own monitoring is optional, mostly relevant when you mine on a large GPU fleet

Getting started

  • Download the AETRON client and launch it. From there it deals with hardware and the available neuronets on its own
  • Pay the registration fee through the client. The reference checks and the benchmark are submitted without any input from you
  • Income, uptime and the re-checks you've passed are all visible inside the client
CTA Background Pattern

Start mining

Install the AETRON client and connect to neuronets in a few minutes. Your GPUs start producing AET, while the network gets useful inference and training in return.