How Does Bittensor Work?

Bitcoin-style halving schedule with 21M hard cap. First halving occurred ~December 2025, reducing daily emissions from 7,200 to 3,600 TAO. Emissions split 41% miners, 41% validators, 18% subnet owners.

Directly modeled on Bitcoin's emission schedule. The 41/41/18 three-way split between miners, validators, and subnet owners adds complexity but the halving mechanism itself is well-understood.

Pro-rata staking rewards distributed per subnet via Yuma Consensus. Validators stake TAO and earn proportional rewards within each subnet they validate. ~72% of circulating TAO is staked.

Standard pro-rata staking at the base layer, but rewards are mediated by Yuma Consensus scores within each subnet, adding a quality-assessment overlay to standard staking rewards.

Direct delegation to validators across subnets. Delegators receive ~82% of validator rewards, validators retain ~18% commission. With Dynamic TAO, staking into specific subnets via alpha tokens acts as a market-driven delegation mechanism.

Standard direct delegation at the validator level, but Dynamic TAO introduced subnet-specific alpha tokens that allow stakers to express preference for individual subnets, creating a two-tier delegation system.

TAO must be burned to register a new subnet on the network. Registration cost creates a barrier to entry and deflationary pressure proportional to subnet creation activity.

Standard burn-on-action mechanism. The cost of subnet registration fluctuates based on demand, acting as a rate-limiter for subnet creation.

OpenTensor Foundation manages network development, upgrades, and treasury allocation. The Foundation operates the Proof of Authority consensus validators and the Triumvirate governance body.

Standard foundation-managed treasury pattern, but the Foundation's control extends unusually deep — they run all PoA validators and the governance Triumvirate is composed of Foundation insiders.

Yuma Consensus scores AI work quality by aggregating validator weight matrices. Validators evaluate miner outputs and submit ranking vectors; the algorithm resolves these into emission allocations, weighting trusted validators more heavily to resist manipulation.

Entirely novel consensus mechanism for evaluating subjective AI output quality. No precedent in crypto — combines reputation-weighted scoring with emission allocation. The algorithm must handle the fundamentally difficult problem of objectively scoring subjective AI work quality. Versions v2 and v3 exist, indicating ongoing iteration.

Taoflow determines subnet emission allocation based on net TAO staking flows rather than token prices or governance votes. Subnets receiving net staking inflows earn higher emissions; subnets with outflows receive reduced emissions. Replaced the centralized root-network valuation model in late 2025.

Novel flow-based emission allocation mechanism with no direct precedent. Uses staking activity as a market signal for subnet value. Scale-invariant design prevents large subnets from having structural advantages. Introduced alongside Dynamic TAO (dTAO) alpha tokens in February 2025, with full transition by December 2025.

The OpenTensor Foundation controls both block validation (PoA) and governance (Triumvirate). This creates a single entity with the ability to censor transactions, halt the network, and push through governance changes unilaterally. The 2024 network halt after the exploit demonstrated this centralized control. The Senate (top 12 validators) provides limited checks but validators are themselves dependent on Foundation infrastructure.

Post-halving, daily emissions dropped from 7,200 to 3,600 TAO. Unlike Bitcoin where fee revenue can partially replace block rewards, Bittensor generates trivial transaction fees since it is not a DeFi chain. Validator and miner incentives will decline with each halving without a compensating fee market, potentially leading to reduced network security and participation as the block reward subsidy shrinks.

Taoflow allocates emissions based on net staking flows, and staking flows are influenced by expected emission yields. This creates a reflexive feedback loop: subnets with high emissions attract more staking, which increases their emissions further (virtuous cycle), while subnets losing stake see emissions decline, causing more unstaking (death spiral). The system can amplify herd behavior and create winner-take-all dynamics among subnets.

Yuma Consensus scores subjective AI work quality, and these scores directly determine miner emission rewards. Validators have financial incentive to collude on scoring to direct emissions to affiliated miners. The trust-weighted algorithm mitigates this but evaluating AI output quality is fundamentally subjective — there is no on-chain proof of correct scoring unlike proof-of-work or proof-of-stake. Gaming the scoring mechanism is the rational economic play for sophisticated actors.

New subnet owners must burn TAO to register, then compete for emissions via Taoflow. If a subnet fails to attract sufficient staking inflows, the registration burn is a sunk cost with no recovery. This creates a high barrier to experimentation and may concentrate the network around established subnets, reducing the innovation that the subnet model is designed to encourage.