Following the trend of DRAM, NAND flash prices have continued their ascent in the first quarter of this year. However, Samsung is proceeding with caution. The prevailing view holds that it is too soon to declare a return to normal supply levels. The recent uptick in NAND flash prices is attributed more to supply constraints than to an actual surge in demand.
The industry also believes that the demand for NAND flash in server applications has not met expectations. Investments in new infrastructure by IT companies are increasingly directed towards generative AI-related processors and leading memory solutions, such as high-bandwidth memory (HBM) and high-performance DRAM, rather than the traditional solid-state drives (SSDs) used in servers. This shift has led some analysts to suggest that the NAND market may require additional recovery time.
According to industry insiders, on Feb. 22, Samsung Electronics reduced its NAND flash production capacity by 50%. SK Hynix is reported to be making similar reductions. The aggressive cutbacks by key manufacturers have led to a steady increase in NAND flash prices. In December of the previous year, the average fixed transaction price for general-purpose NAND flash products (128Gb 16Gx8 MLC) was $4.72, an 8.87% increase from the previous month. Prices have been climbing since October last year.
Yet, experts concur that the price hike in NAND flash is primarily driven by supply-side actions rather than an increase in demand. “While there’s a noticeable uptick in NAND flash orders from major IT clients due to the base effect of last year’s low, the volume remains subdued,” a semiconductor industry expert noted. “There’s some recovery in the mobile sector, but the server and PC markets are only marginally improved from last year.”
The slow recovery of the NAND flash market for servers is attributed to large IT firms prioritizing investments in generative AI infrastructure. This new focus leans toward processors, HBMs, and DRAM rather than NAND. While traditional server setups demand substantial NAND storage, generative AI applications prioritize data generation over storage, relegating NAND to a lesser role.
“Unlike traditional data center investments that rely heavily on large-scale storage with high-capacity NAND-based SSDs, investments in generative AI servers prioritize main memory and processors, with HBMs, DRAMs, etc., making storage less of a focus. To cut costs, some entities opt for hard disk drives (HDDs) over NAND-based SSDs for storage,” explained an IT industry insider.
With half of its production lines currently idle, Samsung Electronics is carefully managing its supply to avoid any potential negative impact on NAND profitability. “As we mentioned in our previous conference call, our stance on reducing production for NAND remains unchanged,” a Samsung Electronics representative said.
