The Open-Source AI Stack
RSS

The Stack · Infrastructure · Open source

Bittensor Compute (Subnet 27 etc.)

TAO-incentivized compute subnets inside the Bittensor protocol; emission-based rewards.

Apache 2.0 · beta · Project site → · GitHub →

Bittensor is a peer-to-peer protocol with a token-incentive design (TAO) and a subnet architecture: each subnet defines a task, miners compete to do that task well, validators score the results, and TAO emissions reward both. Compute-focused subnets let miners contribute GPU cycles for various AI workloads; Subnet 27 is the canonical compute subnet, and several others focus on training, inference, or specialized compute. Bittensor matters because it is the only decentralized AI protocol that has shipped a working token-incentivized network operating continuously since 2021. Compared to Akash (compute marketplace, reverse-auction settlement) Bittensor's compute subnets offer emission-based rewards: miners contribute capacity and earn TAO from the protocol's emission schedule rather than from direct payments. That makes the economics different (and more crypto-native). Production-readiness varies sharply by subnet. The protocol itself is mature; individual subnets range from research-grade to actively-deployed. Used by Templar (subnet that ran a 72B distributed pretrain from scratch) and various inference and training experiments. Limits: TAO economics are volatile, subnet quality is uneven, and the path from "miner contributing compute" to "stable production workload" still has friction.

Sources

Want a follow-up? Ask the chat about Bittensor Compute (Subnet 27 etc.) in context. It will compare to siblings at the same layer and ground every claim in the wiki.

Other projects at the Infrastructure layer

13 siblings · ordered open first