Bittensor is a decentralised machine-learning network where independent operators compete to train specialised AI models inside subnets. dTAO — "dynamic TAO" — is the protocol upgrade that gave every subnet its own alpha token and its own on-chain market against TAO. That single change turned subnet economics from administrator-set emissions into a live market that discovers value continuously. This guide explains the moving parts without the jargon.
Bittensor is a network of parallel machine-learning sub-economies called subnets. Each subnet defines a specific task (generate language, transcribe speech, predict prices, store data, etc.). Inside each subnet, miners submit work and validators score it. High-scoring miners earn emissions; the network runs forever as long as people want to train models on it. TAO is the native token — used for staking, rewards and transfers. There is no company running Bittensor; there is a foundation coordinating the open-source protocol.
Originally, every subnet had a fixed slice of TAO emissions assigned by governance. The community voted on which subnets were worth rewarding, and the vote determined who earned what. This worked, but it was slow, political and had no continuous price signal for subnet quality. A subnet could be producing excellent work and still have the same emission share as one that wasn't.
dTAO gave every subnet its own token — called the subnet's alpha token — and its own AMM (automated market maker) pool pairing that alpha against TAO. Emissions are now denominated in alpha tokens for each subnet. Users can buy alpha tokens with TAO by interacting with the pool; the pool's price becomes the market's live valuation of that subnet's work. High-demand subnets have rising alpha prices; low-demand subnets bleed.
The governance vote still exists and still matters, but it no longer has to predict which subnets are valuable — the pool price already encodes that information, updated block-by-block.
A subnet is a running competition with its own task definition, its own miner/validator roster, its own incentive mechanism and its own alpha token. Current subnets cover language models, text-to-image, speech, prediction markets, storage, finance, audio, code generation — the catalogue expands as the community registers new subnets. A live list is at taostats.io/subnets.
Each subnet mints its own alpha token. Miners and validators on that subnet earn emissions in alpha, not TAO directly. To convert alpha to TAO, they sell into the subnet's AMM pool. To buy exposure to a subnet, a user buys alpha from the pool with TAO. The supply of alpha is determined by the subnet's emission schedule; the demand is determined by who wants exposure.
Every active subnet has a constant-product AMM pool on Bittensor EVM (Chain ID 964). The pool holds TAO on one side and the subnet's alpha on the other. The pool's ratio defines the market price. Trading against the pool moves the price along the constant-product curve (x × y = k) the same way Uniswap v2 works.
BRICKZ aggregates across every active pool. When you sell TAO → USDT or TAO → an alpha token, BRICKZ routes your order through the pool giving the best execution at that moment.
Root stake is TAO staked to the protocol's root pool. It accrues steady yield (from emissions allocated to root) and carries voting weight in the mechanism that decides how emissions are distributed across subnets. Root stake is lower-variance than alpha stake — you get a fraction of everything, which dampens the ups and downs of any single subnet.
Alpha stake is TAO converted to a specific subnet's alpha token and staked into that subnet. Higher yield if the subnet outperforms (its alpha appreciates against TAO and you accrue more alpha from emissions), but higher risk — the alpha price can fall and your stake is exposed to subnet-specific failures.
Validators score miner work inside each subnet. They stake TAO, earn a share of that subnet's emissions in alpha, and typically take a small commission (0–20%) from delegators who stake through them. A validator must maintain uptime and accurate scoring to keep earning.
Miners submit work to a subnet — model outputs, predictions, storage proofs, whatever the subnet's task is. Validators score their output. High-scoring miners earn the bulk of alpha emissions; low-scoring miners earn little and eventually get deregistered.
The feedback loop is what makes dTAO dynamic — each subnet's price is the continuous outcome of real buying and selling, not a governance decision.
If you're mining a subnet, your rewards arrive as that subnet's alpha token on your coldkey. Before you can sell for USDT or fiat, you either:
On BRICKZ, both swaps happen through the same interface. Sell the alpha directly to USDT and the router finds the best path — often alpha → TAO → USDT in a single transaction.