Changes in AI Server Architecture
DDR4 and DDR5 Price Reversal Observed

There are predictions that server memory prices will surge in the future, as Nvidia has started equipping its AI servers with smartphone memory instead of conventional server memory.


Reuters Yonhap News

Reuters Yonhap News

View original image

On November 19 (local time), market research firm Counterpoint Research predicted that, due to the semiconductor shortage, server memory prices could double by the end of 2026.


Counterpoint Research highlighted Nvidia's change in memory policy as the main reason for the price increase. Nvidia has equipped its AI servers with low-power double data rate (LPDDR) memory, which is typically used in smartphones, instead of double data rate (DDR) memory designed for servers. Normally, servers use server memory equipped with error-correcting code (ECC), but Nvidia opted for smartphone memory products with higher power efficiency, as server memory products consume more power. Instead, error correction functions will be handled at the central processing unit (CPU) level.


As a result, demand for smartphone memory has surged, and suppliers are likely to shift production from server memory to smartphone memory, which could further exacerbate the shortage of server memory.



Low-cost server memory is already being traded at significant premiums due to supply shortages. While high-end DDR5 memory is trading at around $1.50 per gigabit, the price of older DDR4 memory, which is used in consumer electronics, is even higher at $2.10. This is higher than the price of high-bandwidth memory (HBM) 3E, which is installed in Nvidia servers, at $1.70.


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing