Skip to main content

What Is AETRON?

AETRON is a blockchain on which AI services run with a cryptographic proof of every computation. A node cannot pass off someone else's answer as its own, quietly swap in a simpler model, or change parameters on the fly. The network verifies each computation mathematically. This verification mechanism is called Proof of Intelligence (PoI).

The main network has been live since October 2025. It is built on Substrate, the same framework behind Polkadot and Kusama, and it provides the base infrastructure: a participant registry, emission, settlements, and governance.

The Core Idea​

Most decentralized AI projects answer the question "did this node really run the declared model?" with reputation scores or random spot checks. Both give partial protection and break down under coordinated attacks.

AETRON takes a different route. Every miner produces a cryptographic proof that a specific computation was carried out exactly as declared: same model, same parameters, same hardware. Emission is paid only for work that passes this check, so compute is spent on useful AI tasks instead of throwaway puzzles.

The Neuronet​

The basic unit of the network is the Neuronet, an AI service run by distributed infrastructure. A single Neuronet can serve many parallel requests and attract many miners, and the network can run hundreds or thousands of independent Neuronets at once across different task classes: language models, image generation, speech recognition, document analysis, AI agents, and more.

Each Neuronet is a self-contained economy. The more AET staked on it, the higher its throughput and the more miners it can support. In-demand Neuronets attract stake, gain capacity, and pay larger returns; weak ones stay at minimum capacity.

Who Takes Part​

AETRON has four roles, and one participant can combine several of them.

Owner​

Starts an AI service by paying a one-time registration fee in AET. The fee is not a deposit and is not returned: it is burned and recycled back into the emission pool. The Owner manages the Neuronet and its tasks through a web interface and chooses how to monetize it (free access, subscription, pay-per-request, or embedding it into an existing product).

There are two ways to run the request layer:

  • Managed (no-code). AETRON hosts the API gateway for the Neuronet. The Owner needs no servers and no DevOps: end users send requests, miners serve them, and the platform handles the routing. In the future, Managed will go beyond the API gateway and let an Owner launch a complete ready-made product, not just an endpoint, so an AI service can be brought to end users without building a front end of its own.
  • Self-hosted (advanced). The Owner runs their own gateway on their own infrastructure and registers its address on chain. This path means running servers and handling DevOps, in exchange for full control over the request layer.

Either way the request layer speaks the same interface. See Gateway API for the endpoints and API Quickstart for a first request.

Miner​

Provides compute and serves requests, earning a share of emission for each verified computation. Miners also re-check each other's work selectively. This mutual verification is the control loop that keeps the network honest: if a re-check finds an incorrect result, the emission for that work is not paid and the offender loses reputation and future task share.

Staker​

Delegates AET in support of a specific Neuronet and earns a share of its economy in proportion to the staked amount. A stake is both a vote for a service's usefulness and a direct increase of its capacity.

Block Validator​

Secures and stabilizes the base blockchain layer and votes on protocol proposals. Running a validator needs an enterprise-grade server and a minimum stake. This role earns a separate share of emission for maintaining the network.

What Makes It Different​

  • Mathematical verification instead of reputation or probabilistic spot checks.
  • Heterogeneous mining across five hardware classes: NVIDIA Ampere, Ada, and Blackwell, AMD CDNA, and Apple Silicon. The same check works across mixed hardware, so data centers, workstations, and consumer devices can all mine in one Neuronet. See Heterogeneous Mining.
  • Low verification overhead, on the order of a tenth of a percent of a miner's capacity, leaving the rest for useful work.
  • A built-in financial layer. AETRON ships with an EVM-compatible layer (Frontier), so Ethereum smart contracts run without rewriting and DeFi and AIfi instruments live on the same chain as the AI services.

Tokenomics in Brief​

AET follows Bitcoin's logic with one difference: emission goes to useful AI work, not to useless hashes.

  • Hard cap of 33 million AET, fixed in the runtime, with no premine and no discretionary emission.
  • Block time of 12 seconds and an initial reward of 1.5 AET per block, about 10,800 AET per day in the current epoch.
  • Halving by total issuance, not by date. The block reward is cut in half each time issuance passes a threshold, with the first halving at 16.5 million AET (half of the cap). Because recycled fees return tokens to the unissued pool, the exact halving date shifts over time rather than landing on a fixed schedule.

See Tokenomics and AET for the full emission schedule and Proof of Intelligence for how verification works.

Last updated on