What is Bittensor (TAO)?

Bittensor's premise is that machine intelligence should have a market price. Rather than one model owned by one company, it runs subnets — independent markets each focused on a task, whether text generation, prediction, data scraping or inference. Miners in a subnet compete to produce the best output; validators score them; the protocol pays TAO in proportion to the scores.

The mechanism that makes it interesting is the scoring. Validators rank miners, and validators are themselves ranked on whether their scores agree with consensus, so producing useless work or scoring dishonestly costs you. It is an attempt to make quality itself the thing being mined, and the hardest part is that quality in AI is much harder to verify than a hash.

Dynamic TAO changed what holding TAO means

Before February 2025, TAO emissions were allocated to subnets by validator vote — a political process that concentrated rewards and invited lobbying. Dynamic TAO replaced it with markets: every subnet now has its own Alpha token and its own liquidity pool. Staking TAO into a subnet is effectively a swap — your TAO enters the pool and you receive that subnet's Alpha token.

That is a genuine structural change and it is not always explained. You are no longer simply staking a network asset; you are taking a position in a specific AI market, with the price of its Alpha token reflecting what the market thinks that subnet is worth. Emissions flow to subnets according to those prices rather than according to votes. It makes TAO closer to an index of AI subnets than to a single staking asset.

The Bitcoin-shaped monetary policy

21 million TAO, hard capped, with halvings — a deliberate echo of Bitcoin and structurally unlike most AI tokens, which inflate indefinitely. The first halving arrived in December 2025 and cut daily emissions from 7,200 to 3,600 TAO.

Whether Bitcoin's monetary design suits a network that needs to pay continuously for compute is the interesting question here. Bitcoin's subsidy buys security, which becomes cheaper to maintain as the network matures. Bittensor's subsidy buys intelligence, which does not get cheaper to produce.

• Payment to miners and validators producing and scoring AI work across subnets.
• Staking into subnets, which under dTAO means acquiring that subnet's Alpha token.
• Registering a subnet, which requires burning TAO — a real and growing supply sink as slots expand toward 256.
• Buying AI services from subnets, which is where the ~$43m of Q1 2026 revenue came from.

That revenue number is the one to track, and it matters more than any roadmap item. Most AI-adjacent tokens have a narrative and no customers. $43m in a quarter from services is small against Bittensor's market capitalisation and it is not zero, which puts it ahead of almost everything in the category.

21 million TAO maximum, no premine and no venture allocation — a fair launch in 2021 with tokens issued only through the emission schedule. Halvings occur on a supply-based schedule, and the first, in December 2025, halved daily emissions from 7,200 to 3,600 TAO.

Registration burns are the other side. Launching a subnet costs TAO, permanently destroyed, and that cost rises with demand for slots. As the network expands toward 256 subnets, burns become a meaningful counterweight to emissions.

What dTAO did to the supply picture

Alpha tokens introduce a second layer of economics beneath TAO. Capital staked into subnets is held in their pools rather than sitting liquid, which removes TAO from circulation for as long as those positions are held — and exposes the holder to a subnet's individual performance rather than to the network's average.

The practical consequence is that comparing TAO's yield to Ethereum's or Solana's is a category error. You are not earning a network rate; you are running a portfolio of positions in AI markets, most of which will fail.

Staking TAO means choosing a subnet and, under Dynamic TAO, swapping into that subnet's Alpha token. Returns depend on how that subnet performs — how much emission it attracts, and what its Alpha token is worth when you exit.

This is closer to venture investing than to staking. A subnet that attracts users and emissions rewards you well; one that does not can leave your Alpha token worth substantially less than the TAO you put in. There is no protocol guarantee of principal here, which is the opposite of what the word staking implies on most chains.

Delegating to a validator within a subnet is the common route for people who do not want to run infrastructure, and it adds the validator's competence to the list of things determining your return.

Verifying AI quality is the unsolved problem

Bitcoin verifies work with a hash — cheap, objective, unfakeable. Bittensor has to verify that an AI output is good, which is subjective and expensive. The consensus scoring mechanism is a serious attempt and gaming it is an ongoing arms race. Academic work has catalogued risk factors specific to decentralised AI subnets for a reason.

dTAO makes staking a stock-picking exercise

Alpha tokens mean subnet selection determines your return, and most subnets in any competitive market fail. Holders who treated staking as a passive yield before February 2025 are now running an active portfolio whether they realise it or not.

Subsidy versus real demand

Much of the activity on Bittensor exists because emissions pay for it. $43m of quarterly service revenue is genuine external demand and it is small relative to what the network emits. The halving tightens the subsidy, which is either the forcing function that makes subnets find customers or the thing that reveals how few had them.

Competing against the frontier labs

The counterfactual for most Bittensor subnets is an API from OpenAI, Anthropic or Google, priced aggressively and improving fast. Decentralised production has to be better or cheaper on some specific axis, not merely decentralised, and that case has to be made subnet by subnet.

Concentration

Validator stake and subnet ownership are concentrated, and dTAO redistributed influence toward capital rather than away from it. The people who decide what gets rewarded are a small group.