The headline for the week is dispersion. The gap between the best and worst 24-hour performer in our 36-subnet sample is over 13 points. When dispersion is this wide, subnet selection dominates TAO direction — you can be right on Bittensor and still lose money on the subnet you picked.

Breadth: the tape is narrower than it looks

Roughly half the set is advancing, and almost all of the strength sits in two cohorts: distributed training (SN3 Templar, SN9 IOTA, SN56 Gradients) and agents (SN36 Autoppia, SN20 BitAgent). Search and content-integrity subnets — SN5 OpenKaito, SN22 Desearch, SN32 Detection, SN34 BitMind — are funding the move. That is a rotation, not an expansion.

Rotations inside a flat TAO price are zero-sum. It is worth being blunt about that: none of this week's subnet strength requires new capital entering Bittensor. It only requires existing stake moving.

Training subnets: real progress or reflexive emissions?

SN3 Templar is up 17.9% on the week and SN9 IOTA 13.1%, both on volume expansion above 25%. The uncomfortable part is that emissions in this cohort have been rising at the same time, which mechanically flatters the APY prints. When a subnet's emission share expands, delegators see a higher yield and stake more, which pushes the alpha price up, which attracts more attention. That loop runs in reverse just as efficiently.

Where APY is rising for bad reasons

Six subnets in the set raised effective APY by more than a full point in 24 hours. Five of them are down on the week. SN16 BitAds (+1.6pp APY, -9.6% price), SN89 Vericore (+1.2pp, -10.3%) and SN32 Detection (+1.4pp, -8.7%) are the clearest cases. The yield went up because stake left. If you are screening on APY alone, this is precisely the trap the screen walks you into.

The clean version of the check: look at APY change and 7-day price change together. Yield up and price up is demand. Yield up and price down is exit.

Liquidity is the quiet risk

Turnover ratios across the set range from about 6.4% of market cap (SN64 Chutes) to under 2% (SN75 Hippius, SN7 Subvortex). The small end of the set is where drawdowns get amplified: thin books mean that the same amount of selling moves price three or four times as far. If you are running a book of more than a few thousand TAO, treat anything under a $10M cap as a position you may not be able to exit at screen prices.

Three things to watch

  1. Whether the training cohort holds its gains through a week of flat emissions — that separates fundamental adoption from an emissions loop.
  2. Volume in the search cohort. Falling price on rising volume (SN5, SN22) usually resolves lower before it resolves higher.
  3. Compression in SN64 and SN19 APY. If the two largest subnets keep absorbing stake, yields there fall further and the yield-chasing flow pushes further out the risk curve.

Nothing here is a recommendation. It is a description of what the numbers in our sample set are doing and the interpretation we would apply to them.

TAO Compound is an analytics tool, not financial advice. Subnet figures referenced above come from the sample data set in this build; the live TAO price shown elsewhere on the site is from CoinGecko.