Understanding stake-weight

Bittensor's dynamic-TAO upgrade (February 2025) split every subnet's economy into its own Alpha token with a constant-product AMM against TAO. A natural question follows: if each subnet's Alpha is its own siloed market, does staking TAO to one subnet do anything for your position on any other subnet? The answer is yes, but through a specific, discounted mechanism — not by staking the same TAO twice.

The two components

For a given hotkey on a given subnet, the chain computes a single stake_weight value that drives everything else — validator permit eligibility, incentive/dividend distribution, all of it. It has two parts:

stake_weight = alpha_stake + tao_weight × root_tao_stake

• alpha_stake — Alpha tokens this hotkey holds staked directly on this subnet, at full 1.0× weight. Fully siloed: staking Alpha on subnet A contributes nothing to subnet B's calculation.
• root_tao_stake — this hotkey's stake on the root network (netuid 0), read fresh regardless of which subnet's stake_weight is being computed. Root stake isn't subnet-specific, so the same root position feeds into every subnet's calculation independently.
• tao_weight — a network-wide discount multiplier applied to the root-stake contribution. Currently 0.18 (governance-adjustable — see below).

Source: pallets/subtensor/src/staking/stake_utils.rs, get_stake_weights_for_hotkey_on_subnet.

Why the discount exists

Public documentation describes tao_weight as intended to decay from parity toward a lower steady-state over roughly 100 days following the Feb 2025 dTAO launch, gradually shifting influence toward directly-staked Alpha as each subnet's own market matures. In practice, live on-chain state shows the value still sitting at 0.18 well over a year past that original schedule — some sources point to a further reversion planned for a different emissions model. Treat 0.18 as current reality, not a fixed constant: it's a live storage value (SubtensorModule::TaoWeight), set by governance, and worth re-querying before relying on it for anything beyond rough intuition.

What this means practically

The two components trade off differently:

Directly staking Alpha on a subnet is the stronger per-subnet lever — full weight, no discount — but the position only counts there. Root TAO stake is weaker per-subnet (0.18× that of an equivalent direct stake) but counts on every subnet at once, since each subnet's stake_weight formula reads the same root position independently, with no cross-subnet accounting that "spends down" a fixed pool as it's counted on more subnets.

This is also the mechanism that makes childkey delegation work: a child hotkey on a given subnet inherits a proportional share of its parent's root stake specifically — not the parent's stake on that subnet — which is exactly the root_tao_stake term in the formula above, evaluated for the parent instead of the child directly.

Where stake-weight actually gets used

stake_weight isn't just an internal number — it's the input to two separate, consequential mechanisms covered on other pages here:

• Validator permits are assigned each epoch to whichever registered UIDs on a subnet rank highest by stake_weight, up to that subnet's max_validators cap.
• Weight-setting and rewards feed stake_weight (as active_stake) directly into the Yuma Consensus dividends calculation — it's not just permit-eligibility, it determines how much of a validator's emission is actually earned.