Betting on the matrix compute and data center energy consumption bottlenecks in the AI inference prefill phase.