Start here

Inference-Based Decentralised Value

AI inference will soon trade as a commodity, like electricity or metals. Its quality, however, cannot be checked. Nothing in an answer shows which model produced it, so whoever buys an answer has to trust the provider. As more capital and decisions come to rely on AI, that trust becomes both unavoidable and a risk. Therefore, a solution is needed that makes AI inference verifiable and trustless at scale.

TensorCash introduces such a solution. It is designed so that every answer can carry a proof of which model produced it, and users, aggregators and verifiers can check that proof without trusting the provider. The model's word choices are tied to a public number from the blockchain, one that no one can choose or predict in advance. Anyone can run the promised model with that same number and see whether it chooses the same words. This is called proof of inference.

We found that the same proof solves a second problem. Since Bitcoin (Nakamoto, 2008), blockchains have been secured by proof of work: miners compete to add the next block with work that is hard to produce and easy to check. It keeps the network safe, but the work itself, guessing numbers, serves no purpose beyond that. On TensorCash, we use the proof of inference as the proof of work. The work that secures the chain is answering real questions, and the same GPU run that serves a customer competes for the next block.

As a result, mining produces something people actually use. An inference provider is paid for the answer, and can earn a block reward for the same work. In this way, TensorCash brings together the best of two systems that have lived apart. AI brings real demand and compute at industrial scale. Bitcoin brings a proven design, with a scarce coin and no central authority. The result is a Bitcoin-derived blockchain with its own coin, $TSC, secured by useful inference.

Welcome to the verification and settlement layer for global AI inference.

Imosuke Takakuni