We're aiming to make Nockchain the most secure money in the world.
Two compute networks already secure base:0x9b5e262cf9bb04869ab40b19af91d2dc85761722, with compute equivalent to $435M+ in hardware behind every block.
I’m out of $TAO.
The subnet model was the whole bet, and in practice it’s produced more extraction than output. A long tail of subnets exist to farm emissions, not to build anything anyone uses. When incentives reward the appearance of work over the work itself, you don’t get an AI economy you get a very expensive leaderboard.
I’ve spent the last while on $NOCK instead. Different thesis entirely: verifiable compute rather than incentivized compute. No committee deciding what deserves emissions, the work is either proven or it isn’t. That’s a much harder thing to fake.
Rotated my full TAO position into NOCK. Not a trade, a change of view on what this category is actually for.
Three weeks after the first AI-PoW block, the AI Compute Network's work rate peaked today at 19.6 ExaMAC/s, about 19,600 H100 equivalents at dense INT8, 170x the first reading.
Worth shining some light on the unit, and how the number is derived.
Can a blockchain reward miners for more than one kind of verifiable computation?
Bitcoin gives every miner the same basic task: search for a valid hash. This provides a consistent measure of mining power because more hashpower means a greater chance of producing the next block. Hashing works well for consensus even though its output has little value outside the blockchain.
Nockchain replaces hash grinding as the costly mining work with generating proofs over NockVM computation. More proofpower gives a miner a greater chance of producing the next valid block, analogous to hashpower in Bitcoin.
Nockchain extends this model beyond a single kind of work. Its base ZKPoW network rewards NockVM proof generation. A separate AI PoW network lets inference providers reuse existing model computation while competing for PoW rewards. Each has its own difficulty and incentive structure.
For proof-based mining to work, proofpower has to remain tied to fresh computation. A miner's chance of winning should scale with effective proofpower without allowing previous work to be reused or amortized across new mining puzzles.
The protocol addresses this by cryptographically binding each puzzle to fresh inputs. Extending the model into a broader compute market introduces another requirement: there also has to be external demand for the computation being incentivized.