Bittensor uses token prices to decide how much each AI subnet earns

How Bittensor allocates $TAO across its AI subnets

Bittensor is a decentralized network built around competing AI subnets, each one focused on a specific digital commodity: compute, inference, storage, or prediction. To fund that competition, the protocol mints new $TAO continuously. Each day, 3,600 $TAO are emitted into the network, equal to 0.5 $TAO every 12 seconds. Those tokens are then divided among the active subnets, but not equally.

Bittensor uses a price-based model for determining how $TAO emissions are distributed across subnets. Each subnet's share of block emissions is proportional to its EMA (Exponential Moving Average) token price, normalized over all subnets with emissions enabled. In other words, the market's appetite for a subnet's work directly shapes how much that subnet earns. Subnets that attract genuine demand see their token price rise, which in turn draws a larger slice of the emission pool.

That said, price alone does not determine outcomes. The protocol runs results through an emission gate that sharply reduces the shares assigned to weaker or underperforming subnets, concentrating rewards where real output is being produced. The V440 Emission Gate introduced a more selective framework for subnet rewards, directing emissions according to performance and demand rather than allowing weaker subnets to capture the same economic weight. New subnets also face a deliberate ramp-up period of roughly four weeks, which blunts the effect of speculative launch pumps on emission allocations.

Where the rewards go once they reach a subnet

Once a subnet's share of the block reward is established, the split inside that subnet follows a fixed structure. Each block mints $TAO which is split 41% to AI miners (who run inference), 41% to validators (who score that work), and 18% to the subnet creator. Validators score miner outputs and submit those assessments on-chain. The validator scores of miners' performance determine the proportion of the subnet's emissions allocated to each miner, according to the Yuma Consensus algorithm.

The design is intended to be self-correcting. Bad-faith validators, those whose scores deviate too much from the consensus, earn less and get diluted by honest ones over time. Meanwhile, as a subnet ages, the TAO that can no longer be injected as liquidity is instead swapped for alpha on the subnet's own pool, buying pressure that transitions mature subnets from liquidity injection to chain buybacks.

The broader result is a network where funding flows toward AI services that the market actually values, rather than being distributed by committee or predetermined formula, according to @opentensor's documentation.

Sources:
Bittensor Emission Documentation, LearnBittensor
Bittensor Official Emissions Docs, Bittensor.com
What Is Bittensor (TAO)? Decentralized AI Explained, Bitcoin.com