Goldman Sachs: high-bandwidth flash could break AI memory wall, up to $150B market
In a research note dated Oct. 12, Goldman Sachs said fast-growing AI inference demand is making insufficient memory capacity a bottleneck on compute efficiency, and that new NAND technologies such as high-bandwidth flash (HBF) could ease this "memory wall" and notably cut AI infrastructure costs. The bank estimates a potential HBF market of $40 billion to $150 billion, assuming GPU adoption rates of 15% to 50%. Goldman flagged SanDisk and SK Hynix as HBF technology leaders, keeping "buy" ratings on both, said Samsung Electronics should also benefit, and kept a "neutral" rating on Micron, whose business is weighted toward DRAM and HBM with less direct exposure.
If the HBF thesis holds, memory could become a new lever for cutting AI inference costs and reshape which memory makers benefit most.