One of the wildest things about AI memory shortage is that regular DRAM now costs more per GB than HBM which tells you just how much demand is reshaping the market.
And AI needs far more than just HBM so here's how the six layers of memory and storage break down:
• SRAM ($3,000/GB) ultra-fast on-chip memory for AI inference
• HBM ($20/GB) high-bandwidth memory feeding GPUs
• DRAM ($25/GB) Working memory for servers, PCs & phones
• NAND ($0.30/GB) flash storage for AI context & inference
• HDD ($0.016/GB) low-cost bulk storage for massive AI datasets
• Tape ($0.008/GB) long-term storage for AI data archives
The companies exposed across these layers include $MU, $SKHY, $SNDK, $STX, $WDC, $IBM, $NVDA & $CBRS.