Top 5 Bittensor Subnets: A Look at the dTAO Ecosystem

A Bittensor subnet is a self-contained market for one kind of machine intelligence — serving models, training them, renting out graphics cards. There are well over a hundred of them. The five below have been among the largest, though sizes move constantly, and the dTAO system decides how much of the network's daily TAO is shared out to each.
What is dTAO?
Since an upgrade called dTAO in February 2025, every subnet has had its own token, known as its alpha. Staking TAO into a subnet swaps your TAO for that alpha at a pool price, much like a swap on Uniswap. That price is the network's vote: the higher it goes, the bigger the subnet's share of the daily emission. Money, not a committee, decides which markets get funded.
Which subnets are the biggest?
- Chutes (subnet 64) — serverless inference. Developers deploy open models such as Llama, DeepSeek or Mistral and pay per request, the way they would with a cloud function.
- Templar (subnet 3) — training large language models from scratch across dozens of machines that have never met.
- Targon (subnet 4) — confidential compute, running models on hardware built so the machine's owner cannot read what passes through it.
- Lium (subnet 51) — a peer-to-peer rental market for graphics cards, with validators checking the hardware is really there.
- Affine (subnet 120) — a reinforcement-learning arena where miners submit models and the best scorers take most of the pay.
Why does the ranking keep moving?
Because it is a price ranking, not a quality ranking. An alpha token's value can halve in a day if stakers leave, and in April 2026 the team behind Templar walked away from the network entirely, which moved both its token and TAO sharply. A subnet near the top this quarter can be mid-table the next, and nothing about a big alpha price guarantees the work underneath is good.
Treat any top-five list, this one included, as a snapshot — check a live subnet dashboard for current figures before you act on it, and read a subnet's own code repository before staking into it. Start with Bittensor explained in plain English, then how subnets get paid.