I spent the last few months examining six domains where blockchain infrastructure can add genuine value to AI systems. and where it doesn’t. This is not a “crypto fixes everything” pitch, but a look at new primitives that might actually work:
• Data Networks — Tokenized, domain-specific data networks.
• Decentralized Compute — Permissionless inference/training markets and their economics.
• Decentralized Training — Parallel, verifiable training without a single trusted coordinator.
• Robotics & Physical AI — Where autonomy reaches ROI first in the physical world.
• AI Verification — TEEs, zk proofs, and hybrids for auditing model outputs.
All six reports (20–30 pages each) are free to download. I’d be especially interested in the HN community’s take on:
1. Which of these primitives are technically viable at scale?
2. Where is blockchain overhead likely to outweigh the benefits?
• Data Networks — Tokenized, domain-specific data networks.
• Decentralized Compute — Permissionless inference/training markets and their economics.
• Decentralized Training — Parallel, verifiable training without a single trusted coordinator.
• Robotics & Physical AI — Where autonomy reaches ROI first in the physical world.
• AI Verification — TEEs, zk proofs, and hybrids for auditing model outputs.
All six reports (20–30 pages each) are free to download. I’d be especially interested in the HN community’s take on: 1. Which of these primitives are technically viable at scale? 2. Where is blockchain overhead likely to outweigh the benefits?