Michael Dell, CEO of Dell Technologies, recently issued a severe warning, pointing out that the memory demand in the global AI accelerator market is experiencing explosive growth. It is predicted that by 2028, this demand will increase 625 times compared to 2023. This near "quantum leap" in demand expansion is placing unprecedented pressure on the global semiconductor supply chain, and the imbalance between supply and demand is unlikely to be resolved in the short term.

Double Explosion in Capacity and Scale

This remarkable increase is mainly driven by both the capacity per unit and the scale of deployment. With the rapid advancement of AI technology, the memory configuration of a single AI accelerator will jump from 80GB in 2023 to 2TB in 2028, a 25-fold increase. At the same time, in order to meet the growing demand for computing power worldwide, the total number of AI accelerators deployed in data centers will also expand 25-fold. The combination of these two increases has created an exponential market demand.

Short-term Production Bottlenecks Are Difficult to Overcome

Currently, the global storage industry is in a critical phase of recovery from the cycle bottom, but the production capacity is clearly unable to keep up with the demand rhythm. It usually takes about four years to build and start production of a new storage chip wafer factory. Moreover, leading companies are being cautious about expanding production, which results in slow progress in front-end manufacturing capacity. Although the market faces multiple risks such as rising prices and extended procurement cycles, the shortage of memory is expected to continue until 2028 due to the strong drive for "sovereign AI" construction in various countries.