One of the projects that keeps coming up in the PoUW conversation is @Tensorcash The core idea is to use the computation behind AI responses as mining work, while making it possible to verify that the model you paid for was actually used. When using an AI API, we usually pay based on the model name the provider displays. But from the response alone, it’s difficult to tell whether they actually ran that model or quietly substituted a cheaper one. TensorCash aims to address this by attaching verifiable records to the inference process. The verification covers which model and execution settings were used. It doesn’t guarantee that the answer itself is factually correct. Miners generate responses using AI models and use the resulting proofs to compete for block rewards. When serving customer requests, they can sell the responses and compete for mining rewards using the same computation. Without customer traffic, they can keep mining by generating their own prompts. The ecosystem includes @inferenceFI AI chat and API services, alongside Bob Labs’ mining pool and AI services. tensorcash:native is used for mining rewards, inference payments, and network fees. tensorcash:native has a maximum supply of roughly 21M, with around 4.4M mined so far. The founder is pseudonymous and goes by Imosuke Takakuni. Among the people around the project with public professional backgrounds are inferenceFI’s two founders, both listed as early contributors to the protocol: @enrico_otta : Former Goldman Sachs MD and head of rates product structuring. He later founded Deep Grey Research and is now inferenceFI’s CEO. @samjavadi : Previously at Barclays, Goldman Sachs, and Morgan Stanley. Now inferenceFI’s COO. A recurring criticism of $Pearl has been: “If so little of the compute is actually doing AI inference, how is this different from ordinary GPU mining?” TensorCash feels like an attempt to address that gap by tying mining directly to the process of a model generating a response. Running models alone doesn’t create demand. But if paying customers start using that inference, this could move closer to what PoUW has been promising: mining through useful computation.