Preparing for a reasonable pullback and building positions in batches $MRVL not only because Jensen Huang is calling for GPUs, HBM, and other markets have been hot for two or three years now. Next is network interconnection. In the future, a single cluster can have hundreds of thousands of cards, and how to achieve high-speed communication between cards is more important than the performance of a single card. $MRVL can benefit from the full-stack interconnection switching dividend. If the AI computing power arms race continues to escalate, MRVL