Most AI + blockchain conversations skip over a basic constraint. AI systems need large amounts of data and computation. Blockchains are not naturally built for either.
So the practical path isn’t about forcing AI fully on-chain. It’s more about making blockchains better at handling structured data and verification. If data can be compressed, referenced, and retrieved more efficiently, many integration problems become less severe.
Heavy computation can stay off-chain where it’s efficient. What matters is that outputs remain transparent and auditable.
This changes the framing slightly. Instead of “AI running on blockchain,” the emphasis becomes infrastructure that reduces friction between data, logic, and trust.