Jordan Nanos explains how hardware constraints, data center scale, and chip co-design shape modern AI performance and tokenomics from silicon to inference.