JPMorgan believes that the storage demand driven by AI is spreading from GPUs to AI CPUs, potentially representing the next wave of demand growth that the market has not yet fully priced in. The report projects a compound annual growth rate of approximately 155% for the AI CPU-related market, meaning that storage demand is no longer solely reliant on GPUs and HBM.
Meanwhile, to meet HBM demand, DRAM wafer capacity is continuously shifting towards HBM, with JPMorgan predicting its share could rise to approximately 32% by the end of 2028. While overall DRAM monthly capacity is projected to expand from approximately 1.9 million wafers by the end of 2025 to approximately 2.85 million wafers by the end of 2028, new capacity typically takes about 2-2.5 years from construction to mass production.
Therefore, a key change in the AI storage cycle is that demand is spreading from a single GPU to more computing platforms, while supply expansion has a significant time lag. If AI CPU demand grows as expected, the storage shortage and high-price environment may last longer than traditional cycle experiences suggest.
