Guotai Haitong: Memory prices are expected to continue to rise, giving the semiconductor industry a "buy" rating.

date
14:10 21/11/2025
avatar
GMT Eight
The bank believes that, since LPDDR is typically used in consumer electronic mobile terminals such as smartphones, this enterprise-level solution replacement may further impact the supply of consumer-grade memory.
Guotai Haitong released a research report stating that NVIDIA Corporation (NVDA.US) recently decided to change the type of memory chips used to reduce the power consumption costs of AI servers. They are replacing the commonly used DDR5 with LPDDR in servers and using low-power LPDDR instead of the traditional DDR5 RDIMM. The firm believes that this enterprise-level replacement may further affect consumer-level memory supply as LPDDR is commonly used in devices such as smartphones and other consumer electronics. With NVIDIA Corporation significantly increasing demand for LPDDR in servers, memory prices are expected to continue to rise and spill over into the consumer electronics market. The industry has been given a "buy" rating. Guotai Haitong's main points are as follows: - LPDDR may replace traditional DDR5 based RDIMM schemes to reduce the power consumption costs of AI servers. - According to Counterpoint Research, NVIDIA Corporation recently decided to change the type of memory chips used to reduce the power consumption costs of AI servers, replacing the commonly used DDR5 with LPDDR and using low-power LPDDR instead of the traditional DDR5 RDIMM. This change is expected to significantly increase industry demand as the value of memory in AI servers far exceeds that in smartphones. The firm believes that as LPDDR is commonly used in devices such as smartphones and other consumer electronics, this enterprise-level replacement may further affect consumer-level memory supply. - SOCAMM performance is better than traditional RDIMM, with memory bandwidth increasing by 2.5x. - DRAM manufacturers such as Micron, Hynix, and Samsung have released SOCAMM memory solutions aimed at promoting the use of low-power DDR in data centers. SOCAMM is a memory module based on LPDDR, optimized for AI servers and standardized by NVIDIA. Unlike traditional DDR used in servers, SOCAMM uses LPDDR for higher power efficiency. Servers typically rely on DDRECC memory for reliability, but NVIDIA Corporation has chosen to use LPDDR to achieve lower power consumption and move error correction logic to the CPU side, rather than relying on DDR5ECC. According to TrendForce, SOCAMM outperforms traditional RDIMM with over 2.5 times higher bandwidth, one-third less power consumption, and significantly smaller dimensions of 14x90mm, making server designs more compact and efficient. It also provides the highest LPDDR5X capacity, with each module being 128GB, improving AI model training and inference performance. - Memory prices are expected to continue to rise and spill over into the consumer electronics market. - According to Counterpoint Research, memory prices have already risen by 50% this year. On top of this, prices are expected to continue to rise by 30% in Q4 2025 and potentially increase by another 20% early next year. With NVIDIA Corporation significantly increasing demand for LPDDR in servers, there are broader and longer-term risk factors surrounding advanced chip availability, which may spill over into the entire consumer electronics market. Risk factors: Storage prices do not rise as expected; downstream demand does not meet expectations.