Bittensor & the SubnetROI model — a full guide

1. What is Bittensor, and why is it suddenly so popular?
2. Registrations, lock costs, slots & the floor
3. Hardware sizing, capex & power
4. Getting started as a beginner
5. The model: formulas, inputs & caveats
6. The Daily Digest

Bittensor is a decentralized machine-intelligence network: it pays people (miners) in its own token (TAO) for useful work — compute, data labeling, inference, agents, robotics data — instead of one central company owning the AI. Anyone can join, pick a task, and get paid for doing it well.

Why it's popular (2026):

• It works, and it pays. The top subnets earn real revenue: SN51 Lium ~$8–10M/yr, SN4 Targon ~$5.5–6M/yr, SN64 Chutes ~$4–5M/yr. Miners at the top of good subnets make thousands of $/day.
• Wall Street arrived. Grayscale filed an SEC registration for a Bittensor ETP (Sept 2026); Kraken became the first major CEX to list Bittensor alphas (July 2026), giving the tokens real markets and deeper liquidity.
• The network keeps growing. 129 live subnets, up from ~60 two years ago; new registrations are capped (~1/day network-wide) so supply of slots is scarce.
• Anyone with a GPU can start — you don't need to be a dev. Pick hardware from our sizing table, pick a subnet, register, point your axon at it, and let the ROI planner tell you whether it's worth it.

Every subnet charges a registration lock — TAO you burn to register (mint) a miner slot on that subnet. Think of it as a joining fee that the network itself sets by supply and demand:

• Lock cost — the current TAO you pay to register. It doubles on every registration (supply gets scarcer) and then decays back down over ~100,800 blocks (~2 weeks), floored at the network minimum (~462 TAO today).
• Floor estimate — the lowest the lock typically decays to before the next registration resets it. Registration near the floor is ~2.5–3× cheaper than right after a doubling.
• Next slot — the next registration window. Registrations are rate-limited (~1 per 7,200 blocks ≈ 1/day), so the cost step-up lands on whoever registers next.
• Time to floor — days until the current lock decays to its floor. The planner's Pro tab models this for you, including the Kraken far-side scenario (price-impact modeling on alpha).

Miner dereg risk: track the table's dereg column — when a subnet gets flagged, it means validators are pulling out and the subnet may be deregistered; ROI disappears fast.

Your P&L starts with what you run. The planner's hardware sizing takes GPU capex, GPU power (watts), electricity rate, infra, registration amortization and computes:

• Power cost — (your kW ÷ 1000) × 24 h × $/kWh. A 350W GPU at $0.12/kWh ≈ $1.01/day.
• Infra — your monthly hosting/infra bill, annualized.
• Registration amortization — the lock TAO × TAO price ÷ amortization days.
• Capex amortization — GPU capex ÷ amortization days.
• Payback — capex ÷ net $/day → how many days until your hardware + lock pay for themselves.

What a good machine looks like (drive the planner's tier selector to see the exact math on live data):

Day 1 — open the planner, set electricity + infra, and look at the default sort (best net $/day). That's the current "best subnet to mine" for your hardware.

Picking a subnet — look for: high emission %, decent miners_tao_per_day, low dereg risk, real liquidity, and (for Kraken subnets) an alpha roadmap. The planner's Methodology tab + methodology.html expands every field.

Buy a machine — start with a used 24GB GPU (~$2,000): it's the sweet spot most TMC setups target.

Pitfall for newbies — the metrics are emission-based: if a subnet's emission % looks huge but the alpha price is collapsing, your actual $/day may be near zero. Always cross-check the α price column.

All math is client-side, estimates from live emission share. No model is run on data older than 30 days; the planner recomputes on TAO price changes, and — if all live sources are down — falls back to a bundled snapshot from Sep 2026 so the page never 500s.

The core

• your TAO/day = miners_tao_per_day × incentive_share
• your USD/day = your TAO/day × TAO price
• net USD/day = your USD/day − power − infra − amortized registration − amortized capex
• payback (days) = GPU capex ÷ net USD/day

Every input & what it means

Caveats — read me. Emission is computed from the current emission % snapshot; it is NOT a prediction. α price is volatile: a subnet's miners income is denominated in its own alpha, so a price drop can erase your net $/day even when emission looks great. Hardware capex assumes you buy used GPUs at the tier capex; performance varies. Bittensor subnets carry real risk — subnets get deregistered, α prices crash, and your burn/registration is non-refundable. This tool is for research, not investment advice.

Every morning, SubnetROI emails a digest with the top 10 subnets — sorted by net $/day for YOUR hardware. It's the fastest way to stay on top of shifts. Subscribe below (free), or upgrade to unlock the Kraken roadmap + live registration alerts.