Wisdom of the $TAO: the future is decentralized AI

What It Covers

Jason Calacanis and Mark Jeffery explore BitTensor's decentralized AI network, where 128 competing subnets use TAO token emissions to fund AI product development. Guests Ala Shaabana (BitTensor cofounder) and Crucible Labs founder explain how subnet projects like Ridges (vibe coding), Targon (private inference), and Hippias (decentralized storage) are undercutting centralized competitors by 5–400x on price.

Key Questions Answered

• •BitTensor subnet economics: BitTensor distributes roughly $100M annually in TAO tokens to miners across 128 subnets, functioning like Bitcoin block rewards but directed at building real AI products. Subnets cost between $100K and $1M to register, with price fluctuating by demand. Investors can gain exposure by staking TAO to receive subnet-specific tokens, accessible via Coinbase, without going through traditional venture channels.
• •Ridges vibe-coding benchmark: Subnet 62, called Ridges, scores 73–88% on the SWE benchmark and 96.3% on the polyglot test, matching or exceeding Claude Code and Codex performance. It launches at $29/month, five to seven times cheaper than competitors. Ridges was built for approximately $10M in token emissions, while Cursor raised against a $29B valuation to reach comparable capability.
• •Crucible Labs allocation strategy: Ala Shaabana built Crucible Labs to automate TAO reward allocation across top-performing subnets, removing the need for active daily management. Users stake TAO inside the Crucible allocator, which distributes staking rewards proportionally to subnets based on performance metrics. This stake-and-forget model mirrors index investing but applied to decentralized AI infrastructure projects.
• •Targon private inference model: Subnet 4, Targon, provides industrial-grade private AI inference using trusted execution environments, meaning user prompts are never consumed for model retraining. Miners supply GPU compute under Targon's technical specifications and earn subnet tokens proportional to inference volume delivered. Investors include Shopify founder Toby Lutke and Ram Sriman, Google's first investor, signaling institutional validation of the subnet model.
• •OpenClaw arbitrage mining: Mark Jeffery built an OpenClaw agent that autonomously mines subnet 85 (VidIO), which uses AI to compress and upscale video at 50–90% lower cost than alternatives. The agent uses Hippias for storage at up to 4,000x discount and Targon for compute at roughly one-tenth market rate. Current yield runs approximately $30/day against $10/day in costs, though competitive miners frequently displace new entrants within 48 hours.
• •Hippias tokenomics structure: Hippias, a decentralized storage subnet competing with Filecoin at 400–4,000x lower pricing, ties token value deterministically to product revenue. Miners must stake Hippias tokens to participate, creating continuous demand. This design eliminates the common crypto failure mode where a protocol's revenue grows while its token depreciates, making token performance structurally correlated with actual storage utilization and business growth.