At present, the entire industry is affected by the tight supply of memory, mechanical hard drives and solid-state drives. The main reason is that the development of the artificial intelligence industry has led to a rapid increase in hardware demand, and the hardware supply of conventional computers for both individuals and enterprises has been restricted.
On Twitter @BullsLab released an internal analysis report from South Korea's SK Hynix, which shows that SK Hynix believes that the current supply shortage of traditional memory will continue until at least 2028, and the inventory of such products has fallen to the lowest level at this stage.
It should be noted that the traditional memory here refers to conventional DRAM, not the HBM series high-bandwidth memory and SOCAMM series memory used by servers. The supply of such HBM and SOCAMM memory used by servers is even tighter.
Growth in demand:
SK Hynix predicts that B/G side (referring to commercial and government customers) server demand will grow by 24%. Due to the surge in demand brought about by artificial intelligence, DRAM memory is expected to usher in a super cycle, which is to maintain a trend of high demand and tight supply for a long time, rather than the cyclical fluctuations in memory that led to a decline in demand and price a few years ago.
Capacity growth is limited:
Since demand has surged and profits have increased significantly, flash memory suppliers should also be building more fabs to increase production capacity, right? This is indeed the case, but the fab construction cycle is long and the expansion of production capacity is slow during the technological transformation period, which means that it cannot meet the needs of the entire market in a short period of time.
SK Hynix expects that the new DRAM wafer fab will reach normal operation by 2028, and production capacity can be increased by then, but capacity growth may be limited due to technological transformation issues.
Increased demand brought about by artificial intelligence PCs:
In addition to servers requiring a large amount of HBM and SOCAMM series memory, SK Hynix said that as the AI PC market share gradually expands, the demand for traditional DRAM will also increase. AI PC requires larger memory, the memory capacity of a single system may increase, and AI PC shipments may account for 55% of total PC shipments by 2026.
What needs to be noted here is that the newly launched flagship devices in the past two years have basically supported NPU units for local AI operation. As the market share of AI PCs expands, users will need to pay for this part of the hardware resources regardless of whether they need to use local AI.
Storage (solid state drive):
SK Hynix believes that the demand for consumer-grade solid-state drives will not grow significantly, but here is a background: the AI industry has extremely high demand for mechanical hard drives (because of the massive amount of data). Due to the huge demand, mechanical hard drive production capacity cannot keep up. Many AI companies have begun to hoard solid-state drives to meet demand, which has also caused the price of solid-state drives to rise.
SK Hynix said that although the demand for consumer-grade SSDs will not increase significantly, the demand for QLC NAND-based products (mainly enterprise-grade SSDs) will increase, and if supply continues to be tight, the price of such SSDs will continue to rise.
Finally, SK Hynix also mentioned that IDM companies (referring to semiconductor companies that design, produce, and sell integrated circuits) are currently focusing their main investments on server DRAM and HBM memory, which has resulted in a relative shortage of NAND flash memory supply. Therefore, NAND inventory levels are expected to decline significantly, and price upward pressure will continue to exist.