“General-purpose GPUs are not very efficient for AI inference and will not perform well in market segments that require highly optimized metrics for specific applications.”