Commentary from Marco Mezger, Global President MEMPHIS Electronic.
Device manufacturers were able to enjoy two years of low memory prices. However, this year, the tide has completely turned. Memory manufacturers have pulled the ripcord on both NAND flash memory and DRAM, announcing that they will focus on new memory generations in the future.
This development is not new, but the cycles in which memory generations are replaced are becoming shorter and shorter. The major memory manufacturers have stopped producing DDR3 as recently as at the end of 2025. Twelve months later, they are announcing the end of their DDR4 components.
It's time to ask ourselves: Will memory manufacturers determine when designs are transitioned to new processors? How did it come to this?
To answer this question, you need to understand the current market dynamics. This will help you assess the impact on your own production better.
The DRAM industry has consolidated from a market with over 17 manufacturers in the 1990s to one dominated by three leading players: Samsung, SK Hynix, and Micron. This is due to the high investments required for the continuous development of DRAM technologies and the construction of corresponding production facilities. Capital expenditure is comparable to that of large infrastructure projects and is often amortized over 10 or more years. Only a few companies have the necessary resources.
DRAM production is reaching its physical limits. Until recently, the storage density of DRAMs doubled every two years. Today, it takes four years to achieve this. At the same time, AI and other data-intensive applications are demanding ever more computing power, resulting in an increasing discrepancy between computing capacity and available memory bandwidth per processor.
Memory manufacturers are trying to fill this gap with high-bandwidth memory (HBM), where their margins are highest. Although HBM currently accounts for less than 10% of the industry's bit output, it generates a third of total sales. This explains how the DRAM market can grow by 25% this year, even though actual production capacity is declining. But it also explains the focus on HBM as a premium memory. By 2030, nearly all new factories will be geared towards producing HBM.
This will have an impact on conventional DRAM technologies. Wafer production for DDR4 and LPDDR4 is expected to decline by 90 percent by 2027, and DDR5 and LPDDR5 will be the dominant memory technologies. Beyond 20230, DDR4 and LPDDR4 will only be offered by second-tier memory manufacturers, such as Nanya and Winbond, as well as specialized fabless manufacturers like Intelligent Memory.
If there’s no memory supporting a processor, chip manufacturers will have to adjust their product roadmaps. By 2030, DDR3 production will decline sharply, and only a few products will still use it.
For DDR4-based designs, industrial manufacturers should not rely on products from the major memory players. However, it will take until the end of 2026 for Winbond and Nanya to ramp up their production capacity to fully meet demand. Intelligent Memory is another supplier that is focusing its entire portfolio on the industry and will have availability.
It does seem likely that the processor will no longer dictate product design in the future, but rather the availability of memory products. This requires an adjustment of the procurement strategy, and specialist distributors are ready to assist with this.
Author: Marco Mezger, Global President MEMPHIS Electronic.