Building the Bitcoin Equivalent for AI Inference Compute with David and Daniil Liberman (Gonka)

In this episode, Austin chats with David and Daniil Liberman of Gonka about the case for using crypto to build a permissionless, meritocratic AI compute network focused on inference rather than decentralized training. They argue that centralized AI infrastructure is rapidly concentrating both revenue and compute, and contrast Bitcoin’s immutable, open rules with corporate systems whose promises can change. With Bitcoin as their north star, Gonka’s protocol aims to verify inference outputs and validate the underlying hardware to guard against fraud, censorship, and model substitution, using proof-of-work-like measurements of real, always-on hardware rather than proof-of-stake incentives.

00:00 - Why Crypto for AI

03:10 - Inference Over Training

05:02 - Bitcoin as Compute Incentives

07:21 - OpenRouter Critiques

11:36 - Open Source and Safety

18:51 - How Gonka Works

20:14 - Verifiable Inference Proof

24:22 - Privacy and Encryption

29:07 - Decentralized Training Roadmap

34:16 - Keeping Frontier Labs Honest

42:44 - More on Why Bitcoin Provides an Ideal Compute Model