Memory vs. GPU

Hey everyone, I've got a macro observation I'd love to get your take on: Everyone labels memory stocks like Micron and Sandisk as classic cyclical plays, but shouldn't NVDA and AMD fall into the exact same category? At the end of the day, there are no computers or AI servers without both memory chips and GPUs—they are two sides of the same coin. With AI and AGI demand scaling aggressively for years to come, if memory is treated as cyclical, GPUs should logically be viewed through the exact same lens. 1. It’s no longer an off-the-shelf commodity High Bandwidth Memory (HBM3e/HBM4) is custom-engineered, 3D-stacked silicon closely integrated with specific GPU architectures. Micron isn't just dumping DRAM onto a volatile spot market—capacity is locked into binding, non-cancellable Strategic Customer Agreements (SCAs) with major hyperscalers, backed by upfront cash deposits and price floors. 2. Wafer math prevents sudden supply gluts Producing 1 bit of HBM requires ~3x the silicon wafer capacity of standard DDR5 DRAM. Converting cleanroom space to HBM restricts the overall supply of standard server RAM and NAND, keeping pricing firm across the entire memory market. 3. Memory is the hardware bottleneck GPUs literally cannot function without HBM. With memory makers' capacity booked out through 2027, memory acts as a legally contracted infrastructure bottleneck rather than a simple side-show commodity. Curious to hear how others are playing memory vs. compute right now. Is Wall Street underpricing the structural shift in HBM?