Yuusuke Ohara: Agentic AI Drives Simultaneous DRAM and NAND Demand—Why China's Two Memory Giants Consider Cross-Entry

The rise of Agentic AI has reshaped a previously overlooked supply-demand structure. In his EE Times Japan column, Yuusuke Ohara notes that AI inference previously drove demand mainly for GPU-attached HBM, but Agentic AI also requires large numbers of CPUs, and KV Cache demands far more memory capacity than before, pushing DDR5 and LPDDR5/5X demand sharply higher. More critically, when DRAM capacity still falls short, systems begin using NAND flash as secondary memory to build memory pools, bringing NAND into tight supply as well.

The direct consequence of this mechanism is that DRAM makers and NAND makers, previously each holding one end of the market, simultaneously achieved record revenues. CXMT (ChangXin Memory Technologies) completed its IPO in July at an offering price of 8.66 RMB per share, closing the first day at 49 RMB, with market capitalisation reaching 3.3 trillion RMB and vaulting to the top spot among Chinese companies. YMTC (Yangtze Memory Technologies) filed its IPO application in August; reports suggest it expects to raise 33 billion RMB. Both companies now enjoy relatively ample funding.

Ohara argues that it is precisely this simultaneous tightness and simultaneous capital headroom that forms the backdrop for both specialist manufacturers considering cross-entry. Reports indicate CXMT is attempting to enter the NAND flash market, while YMTC is reportedly considering entering the DRAM market. The initial reports came from Taiwan-based DigiTimes, followed by Reuters and other outlets. The author attributes this move to operational stability considerations: when a company's fate rests entirely on a single product type, the risk of a cycle reversal remains enormous even when current profits are substantial. Cross-entry can smooth that risk.