Michael Dell, the founder and CEO of Dell Technologies, has made a bold prediction: by 2028, the overall global demand for AI accelerator memory is projected to soar to 625 times its 2023 levels. This forecast hinges on two pivotal factors. Firstly, the memory capacity of each AI accelerator is anticipated to surge from the 80GB found in NVIDIA's H100 GPU in 2023 to a staggering 2TB by 2028, representing an approximate 25-fold escalation. Secondly, the worldwide deployment scale of AI accelerators is expected to expand 25-fold over the same timeframe. The convergence of these two trends will fuel a monumental 625-fold spike in total memory demand. At present, the global semiconductor industry is grappling with structural supply-demand disparities, characterized by constrained expansion in memory chip production capabilities and a buoyant demand for the core components essential for AI computing power, culminating in a protracted supply shortage.