Austin Lyons says Nvidia (NVDA) and other GPU makers are "changing the dynamics" when it comes to how their chips improve LLM inference. Arm Holdings (ARM) is capitalizing on CPUs, something Austin considers a bottleneck for GPUs. He explains how Arm aims to fill in the efficiency gap. Turning to the AI memory space, Austin says Alphabet's (GOOGL) TurboQuant isn't a replacement for Micron (MU) and SanDisk (SNDK) but rather a tool to utilize their hardware more.