Bittensor is a blockchain built to pay for AI work. A blockchain is a shared record of who owns what, kept by many computers at once instead of one company's server. Bittensor's own chain, called Subtensor, does not do the AI work itself. It runs markets, called subnets, where people compete to produce something useful, such as answers from an AI model, training runs or computing power, and it pays them in its coin, TAO.

Each subnet has two sides. Miners do the work. Validators check the work and rank the miners, and a rule called Yuma Consensus, which runs on the chain, turns those rankings into rewards. Validators whose rankings stay close to what the others see earn more, and one who talks up a favourite miner is penalised. In Bittensor, mining has nothing to do with adding blocks to the chain; it means producing the work.

Anyone with the money and the skills can start a subnet, up to a limit of 128, and the Opentensor Foundation, the nonprofit behind Bittensor, does not run them. Each subnet's rules live off the chain, in code kept by the team that created it. Bittensor also runs a copy of Ethereum's programming system, so most Ethereum apps can run on it, paying their fees in TAO.

For you as a holder, TAO is the coin of a marketplace for AI work, and its worth depends on how much useful work that marketplace produces.

TAO has several jobs, and each one is a reason someone needs to hold it. The first is reward: the chain creates new TAO and pays it out for the work done in subnets. Bittensor's own documents describe TAO as the coin that "serves to incentivize activity across the platform".

The second is entry. Miners and validators pay a fee in TAO to register in a subnet, and that fee moves with demand. A validator needs a stake of at least 1,000 behind it to qualify, counting coins that others have lent it. Apps built on Bittensor's copy of Ethereum pay their fees in TAO too.

The third is taking part without running anything. Holders can lock their TAO behind a validator they trust, which is called staking, and share in its rewards. Since 13 February 2025 they can also stake TAO into a particular subnet and receive that subnet's own coin, called alpha, in return. Bittensor's documents describe this as stakers voting "with their TAO" for which subnets matter, because a subnet's share of new TAO depends on that support.

For you as a holder, demand for TAO rests on people wanting to work in, check, build on and back these subnets. If the subnets keep producing work that people pay for, that demand has a reason to last.

Most of the money flowing through Bittensor is new coins, not fees. Every subnet receives a share of the new TAO each day, and its own alpha coins are split three ways: 18% to the subnet's owner, 41% to the miners and 41% to the validators and the holders staking with them. Validators keep a cut of what they earn for their stakers, 18% by default, which they can change.

How new TAO is shared between subnets has changed twice. From November 2025 to June 2026 it followed how much TAO people were staking into or pulling out of each subnet. Since June 2026 it follows each subnet's coin price, averaged over time, which the documents describe as a return to the earlier method.

Nobody keeps the chain's transaction fees. They are recycled, meaning taken back out of the count of coins issued, as if they had never been created. The fees to register miners and validators are recycled the same way. The cost of registering a new subnet, for subnets created after 1 October 2025, is burned, which means destroyed for good. Each subnet's pool also charges a small fee of about 0.05% on every trade, which goes to whoever produced that block.

For you as a holder, the design is clear: new coins pay the workers, and fees shrink the supply instead of filling anyone's pocket. How much has been recycled or burned in total is not measured anywhere.

On 1 June 2026, the market valued all TAO at $2,450,595,510, the 40th largest crypto asset at the time. One coin traded at $255.07, 66.3% below its all-time high of $757.60 in March 2024, and 38.3% lower than a year earlier.

The usual way to check a price is against what the network earns, and for Bittensor that cannot be done. No public tracker records the fees the chain earns, so there is no earnings figure to set against the price. That is missing data, not proof that Bittensor earns nothing.

The one measured business belongs to a single subnet. Chutes, subnet 64, sells AI computing time, and it took in $3,693,238 in fees between 26 December 2025 and 6 September 2026. That is one company's revenue on Bittensor, not the chain's, like a shop's sales rather than the mall's rent.

What can be measured is how much TAO holders have staked into subnets: $560,569,221 on 14 October 2026. Set the market value against it and the market was worth about 4.4 times the staked total. That comparison says little, though. Both figures move with TAO's price, they are four months apart, and staked coins are not income.

For you as a holder, TAO's price today rests on what the subnets might become, not on earnings anyone can count.

On paper, Bittensor is moving towards a shared system for deciding its rules. In practice, as of June 2026, it is not there yet. Its own governance page says the new system "is not yet deployed on mainnet" and that the live network "uses pallet_sudo for privileged operations."

Sudo is a master key. Whoever holds it can make top-level changes to the network, including replacing the software the chain runs. Before, a group of three called the Triumvirate held that access directly. Under the new design the Triumvirate would only be the first step of a public vote, with no master key at all, but that design is still in development. Who holds the key today is not stated in any source available.

Inside each subnet, power sits closer to home. A subnet's governor can raise or lower how many validators it allows, and its creator keeps its rules in the team's own code.

That concentration has had consequences. In April 2026, Covenant AI, a large subnet team, left Bittensor citing governance concerns, and TAO's price fell by between 15% and 27%, depending on the period measured.

For you as a holder, the rules that protect TAO, including the cap on its supply, can today be changed by whoever holds one key, not by a vote of holders.

One organisation still runs the chain itself. Bittensor's own documents say that "the work of validating the blockchain is performed by the Opentensor Foundation on a Proof-of-Authority model." In plain terms, the computers that confirm transactions belong to an approved operator rather than being open to anyone who stakes coins, the way a members-only club differs from a public market. The validators inside subnets, who check AI work, do a different job.

The same documents say most of the chain's code is written by engineers working for the Opentensor Foundation. A startup called Latent Holdings builds the main tools and the documents, keeps a public explorer and runs one subnet.

In mid-2026 the code moved house. Between 29 June and 13 July 2026, the chain's main repository, where its code is kept, moved from the Opentensor Foundation's account to one called RaoFoundation. Its only tie to Bittensor in the project's own words is a page signed "Const, core dev @ Rao Foundation". What Rao Foundation is, who owns it and how it relates to the Opentensor Foundation is not explained anywhere, and no announcement of the move was found.

The work itself has not slowed. Between 17 July and 15 October 2026, 17 different authors made 829 changes to the chain's code, and older tool repositories were closed because they were merged into it.

For you as a holder, the network is built actively and in the open, but it is run by a very small circle, and part of that circle is not explained.

TAO has a hard limit of 21,000,000 coins, written into the chain's code. Once that many exist, the code creates no more. On 16 October 2026, 11,622,564 TAO had been issued, 55.3% of the cap.

New TAO is created with every block, and a block comes every 12 seconds. It started at 1 TAO per block. On 15 December 2025 the total issued passed 10,500,000 and the reward halved to 0.5 TAO, which the documents describe as 3,600 new TAO a day. It will halve again when 15,750,000 have been issued, then at 18,375,000. These halvings are set by how many coins exist, not by a date, so nobody can say exactly when the next one comes.

Recycling slows the growth. Because fees and registration costs are taken back out of the count, the supply grew by 0.3973 TAO per block in the most recent stretch measured, about 2,861 coins a day, roughly a fifth less than the 0.5 created.

For you as a holder, that pace adds about 1.04 million TAO a year, around 9% of today's supply, so holding the same number of coins means owning a smaller share each year unless you stake. The cap is real, but as the sections above explain, it sits in code that one master key can still change.

This is hard to answer for Bittensor, because the usual signs of use are missing. No tracker counts its transactions, its active users or the trading in its subnet pools, and none records the chain's fees.

What can be seen is how much TAO holders have put behind subnets. On 14 October 2026, the TAO staked into subnet pools was worth $560,569,221, after a high of $633,401,903 on 27 September. This is not money parked in lending or trading apps; it is holders backing subnets, and part of it comes from new coins flowing into the pools. The figure also rises and falls with TAO's price, so a drop does not always mean people left.

Only 26 days of that figure can be compared, because the tracker began counting all the pools on 19 September 2026, although they have existed since February 2025. So no trend can be read from it in either direction.

Some activity is plain. Subnets are run by independent teams, up to 128 of them. One, Chutes, has a measured paying business selling AI computing time. The chain's code changes hundreds of times a month, by more than a dozen people.

For you as a holder, Bittensor is clearly being built and used, but how much its subnets are worth to the people who pay for their work is out of view right now.

Bittensor's worst day came on 2 July 2024. At 7:06 pm UTC, someone began moving coins out of people's wallets. The Opentensor Foundation spotted the unusual transfers at 7:25 pm, and at 7:41 pm it put the chain's validators behind a firewall and switched the network to a safe mode, "thus stopping all transactions."

The cause was not the chain. A poisoned copy of Bittensor's own software package, version 6.12.2, had been uploaded to a public download site. Anyone who installed it between 22 and 29 May 2024 and then unlocked their wallet keys sent those keys to the attacker. The Foundation said the attack "DID NOT affect the blockchain or Subtensor code," and gave no loss figure. News reports put the losses at about 32,000 TAO, roughly $8 million. A larger figure of $28 million, from an investigator, adds up several thefts that year and is not the size of this one.

In August 2025, a security team found fake look-alike packages, with names such as "bitensor", built to steal from Bittensor wallets. They were not the official software and not a fault in the chain.

For you as a holder, two lessons stand out. Your coins are only as safe as the software you use to hold them. And the network's ability to stop every transaction within 35 minutes of the attack starting protected people that day, but it also shows how much power one organisation holds.

Some of what you will read about Bittensor describes how it is meant to work, and some describes how it works today. Here the difference matters more than usual.

Its documents describe the system of subnet coins as removing a central judge, so that the value of subnets is set "in a wholly distributed fashion". Outside descriptions call Bittensor a "decentralized" machine-learning network. Both are fair about how subnets are valued. Neither describes who runs the chain: its own documents say one foundation validates it, and its governance page says a master key is still in use.

The money claims check out exactly. The 21,000,000 cap is in the code. The documents' 3,600 new TAO a day matches the code's 0.5 TAO every 12 seconds, and the chain's own records show the halving on 15 December 2025.

Some numbers you will see elsewhere are wrong. A popular data provider's model dates the halving in October 2025, two months early, and gets the maximum supply slightly wrong. One reading stored on the chain still shows 1 TAO per block, because the software no longer updates it. And the documents' own formula for sharing new coins between subnets may predate a change made in August 2026.

The founding paper, written under the name Yuma Rao, describes the original idea before subnets and subnet coins existed. It is a guide to intent, not to today's design.

For you as a holder, the safest numbers for TAO are in its code and on its chain, and the claims about who controls it are best read with their date beside them.