Counterpoint Predicts: HBM Memory Demand for AI Server ASICs Will Hit 35x 2024 Levels by 2028

Counterpoint Research's report shows that by 2028, the demand for HBM memory from AI server computing ASICs will reach 35 times the 2024 level, and the average HBM memory capacity will also increase by nearly 5 times.

In the market, Google and its TPU chips are the main drivers of HBM demand in the AI ASIC camp, accounting for 58.5% of the camp's share, with Amazon taking 20.3%. HBM3E will remain the mainstay (56.3% share), while HBM4 and HBM4E combined will account for 37.7% of the market in 2028.

Regarding supplier trends, Samsung is expected to regain HBM market share from SK Hynix; in packaging technology, TSMC's CoWoS dominates, and Intel's EMIB-T technology is also gradually gaining market attention.