Inference Networks.
Decentralized model serving — most ambitious crypto-AI thesis, hardest to make a market in.
Thesis
Why this sector, why now.
Inference networks (Bittensor, Allora, others) are betting that model serving fragments away from centralized providers — that a market of specialized inference subnets out-performs OpenAI's general endpoint on cost, latency, or capability. The technical claim is plausible. The market claim is the question: do real customers buy decentralized inference when centralized inference works fine and gets cheaper every quarter? I have not seen the demand signal yet. Watching for a single non-crypto-native customer adopting at scale.
Signals I track
What would move my read.
- 01
Subnet specialization producing measurably better results than centralized baselines
- 02
Real revenue per subnet vs token emissions to validators
- 03
Whether enterprise pilots convert to production usage
Kill shot
What would kill the thesis
Centralized inference (OpenAI / Anthropic API) keeps cost-per-token falling faster than decentralized networks can match, and no killer use case for trustless inference emerges.
Coverage
Projects on the radar.
Going deeper
Bespoke Inference Networks dive for your fund.
Draft thesis · Grey voice in progress, edits land continuously