Everyone is crowding into $MU and $SNDK for the memory shortage. But most people are overlooking $PENG. Same tailwind, way less attention. → AI moved from training to inference, and inference is memory-bound, not compute-bound. The GPU's onboard HBM becomes the wall. → DRAM prices doubled in Q1 and Reuters reported they could climb another ~63% this quarter. HSBC's Kazunori Ito sees little prospect of a quick drop. → The crowd buys the chipmakers. $MU, $SNDK, the obvious picks. But the shortage also reprices the system integrators who package that memory into AI factories. That's $PENG. Why $PENG is still a good play: → Memory segment grew +63% YoY in Q2 FY2026 on AI demand and pricing. → The MemoryAI KV Cache server (industry-first, March 2026) adds up to 11 TB of CXL 2.0 memory per node ≈ 10x faster than NVMe caching, 3.8x faster than RDMA. → Built on 30 years of SMART Modular memory IP + an SK Hynix DRAM collaboration competitors can't easily copy. The rerate is already starting: → $16 (52-wk low) → ~$71 recent high. Rosenblatt and Stifel both hiked targets into the $60s, Street high sits at $108. → The market is re-rating $PENG from a commodity hardware multiple to an AI-infrastructure multiple. While everyone fights over $MU, that gap is where the alpha lives. I'm in at $30, holding, and $100 is my minimum target. Not financial advice. DYOR.