What Is Bittensor (TAO)? How the Network Works
What is Bittensor (TAO)? A plain-English look at subnets, miners, validators and how the network pays people for AI work. How it works only, no price talk.
TL;DR
- Bittensor is an open network where many small AI "markets" (called subnets) compete to produce useful work.
- People who do the work are called miners. People who score the work are called validators. Both are paid in a token called TAO.
- A 2025 upgrade called Dynamic TAO gave each subnet its own token, so the market decides which subnets get rewarded.
- TAO has a supply cap of 21 million, and its first halving was in December 2025.
- This post covers how it works, not what it's worth. Most tokens lose value, and scams are common.
Bittensor is a blockchain network that pays people for useful AI work. Many separate sub-networks, called subnets, each focus on one job. That might be running AI models or storing data. The network rewards the best work in a token called TAO.
Risk Notice: Most tokens lose value, and scams that copy real projects are common. This post explains how Bittensor works. It doesn't say you should buy, hold or sell anything, and it doesn't talk about price.
Where did the Bittensor idea come from?
It started with a paper called Bittensor: A Peer-to-Peer Intelligence Market by Yuma Rao. The page I read doesn't show a date. It describes a marketplace where AI models rate each other's work. Fixed tests aren't the only judge.
Those ratings sit on a shared ledger. Models that others rate highly earn a bigger share of the rewards. The paper also adds a rule meant to stop small groups from voting only for themselves.
You don't need to follow the maths. Here's the simple version, mate. One company doesn't decide which AI is good. Many players judge each other, and the network pays the ones judged useful.
What are subnets, miners and validators?
Bittensor's own docs describe the network as a set of independent subnets. Each subnet makes one kind of "digital commodity". The docs name compute, inference, storage and prediction. Inference just means running a trained AI model to get an answer.
The docs list these roles:
| Role | What they do | |---|---| | Subnet creator (owner) | Sets the rules for what counts as good work in that subnet | | Miners | Do the work, such as running a model | | Validators | Score the miners' work | | Stakers | Back validators with TAO |
Here's how I think about a subnet. It's a small job market. The owner posts the rules. Miners compete. Validators act as the judges. The chain then pays out based on those scores.
Need the basics of what a token is first? Our post on NFT vs token is a good start.
How does TAO get paid out?
TAO is the network's token. It has a supply cap of 21 million, like Bitcoin. Fortune reported that the first halving took effect on 15 December 2025. It cut new TAO issued each day from 7,200 to 3,600. A halving just means new coins are made at half the old rate.
A 2025 upgrade called Dynamic TAO, or dTAO, changed how rewards flow. Bankless covered the launch on 14 February 2025. Before it, subnets shared one pool of rewards ranked by a single group of validators. After it, each subnet has its own token, paired with TAO in a liquidity pool. A liquidity pool is a shared pot of two tokens that lets people swap one for the other.
So demand from users, stakers and validators helps decide which subnets grow. A useful subnet can attract more. A weak one can fade.
How are rewards split inside a subnet?
Oak Research explained the split after dTAO. It's a third-party analysis, first published on 27 March 2025 and updated in December 2025. Its numbers were:
- 41% goes to miners.
- 41% goes to validators.
- 18% goes to the subnet owner.
The same piece says each block adds 1 TAO shared across all subnets. It calls subnet tokens alpha tokens. They're swapped against TAO inside each pool, not traded on regular exchanges.
Treat these as a snapshot. Network upgrades can change the rules. Check Bittensor's own docs for the latest version.
How does this connect to real AI?
This is what makes Bittensor different from a normal crypto project. The tokens pay for work that's meant to be AI work. A subnet might reward the best answers to questions, the best predictions, or the best storage.
That's the big idea behind "decentralised AI". One company doesn't run everything. Many people run pieces. For another project built around AI and data, see our explainer on Ocean Protocol.
There's a hard problem hiding here. Someone has to judge whether the work is good. In Bittensor, validators do that. If the judging is weak or can be gamed, rewards go to the wrong people. The design tries to guard against this. It's still an open challenge for the whole field.
What are the risks?
Fortune's report calls Bittensor not the easiest tech to understand. It notes that its promise depends on continued adoption. Bankless's dTAO piece lists more.
Here's what it flagged:
- Prices of subnet tokens can swing hard, especially early on.
- Validators and stakers can back the wrong subnets.
- Subnet owners who don't build real value risk their own token losing worth.
Add the usual crypto risks on top. Fake sites, fake wallets and impersonators love AI crypto names. Our guide to how crypto scams work covers the warning signs.
Do you build token systems yourself? Auditing a design is worth a look. See this builder piece on a builder's checklist for judging token design.
Frequently Asked Questions
What is Bittensor in simple terms?
It's a network that pays people for useful AI work. Separate subnets handle different jobs. Miners do the work, validators score it, and the network pays out in TAO.
What is a subnet in Bittensor?
A subnet is one small network inside Bittensor with its own job and its own rules. One might run AI models. Another might store data or make predictions.
What is dTAO?
Dynamic TAO is a 2025 upgrade. It gave each subnet its own token, paired with TAO in a pool. That lets demand decide which subnets earn more.
How many TAO can ever exist?
The cap is 21 million, according to Fortune. The first halving was in December 2025. New supply keeps falling over time.
Not financial advice
This is general education. It isn't financial advice, and it doesn't recommend TAO or any other token. Most tokens lose value. Only risk money you can afford to lose.
What to do next
Read our short guide to what AI tokens are and how they differ. Then check any Bittensor claim you see against the project's own docs.
About the Author
James Killick
AI Product, Web3 & Sales Automation Specialist
James Killick runs this site →Digital Product Development
Devwiz is a Sydney-based digital product development agency specialising in launching and scaling SaaS products, web apps and mobile apps. James is a shareholder and advisor, shaping product strategy and PLG approach.
// Related
Keep Reading
Crypto AI Agents: Can They Own and Trade NFTs?
Crypto AI agents explained: how an AI agent can hold a wallet, own NFTs and make trades, what the guardrails are, and the risks you should know first.
What Is Decentralized AI? Who Runs the Computers
What is decentralized AI? A plain-English guide to who runs the computers, how the pieces fit, what's proven so far, and what's still hard. Not advice.