Market still feels stuck in a 2023 mental model on AI memory. Everyone is debating HBM capacity, which is fine, but the more interesting bottleneck is memory optionality per token. Call it HOT if you want: heterogeneous offload throughput. In the CP...