With the advancement of artificial intelligence (AI), storage demand is rapidly increasing across data centers and on-device (embedded) systems. There is a growing view that the status of NAND flash, long discounted as a "low-margin memory," could change. As the AI industry shifts from training to inference, the importance of storage devices that store and reuse large volumes of data for long periods has grown.
NAND flash has long been called the "albatross" of the memory industry. Because the technological barriers to entry are relatively low, oversupply has recurred, and high price volatility has made it difficult to secure stable profitability. Its structure, which makes it hard to command a premium through performance competition like DRAM, has also been cited as a limitation. Even as the AI market surged, NAND flash drew relatively little attention behind high-bandwidth memory (HBM).
What cracked this perception was Nvidia's recently presented vision for next-generation AI infrastructure. Nvidia said that as the AI industry moves from training to inference, the role of storage devices that store and reuse large-scale data will grow rapidly, not just compute performance. In response, it unveiled an architecture that actively uses not only high-speed memory next to the graphics processing unit (GPU) but also large-capacity solid-state drives (SSD) outside the server. In effect, it elevated storage from a mere auxiliary tool to part of the AI inference pipeline.
In this process, NAND flash, which had been relatively undervalued, is being redefined from a "cheap storage semiconductor" to a core component of AI infrastructure. Given that SSD and NAND flash capacity per server could increase significantly, Nvidia's blueprint is being taken as a signal that the demand structure for NAND flash may change.
Demand indicators are already responding. According to the Korea International Trade Association, in January this year NAND flash exports totaled $1.3 billion, up 108% from a year earlier, while export volume rose 28%. Market researcher TrendForce said NAND flash prices rose 33%–38% in the fourth quarter of last year from the previous quarter and are likely to see similar hikes in the first quarter of this year. That is why major players such as Samsung Electronics, SK hynix and SanDisk have raised NAND flash prices for enterprise SSDs (eSSD).
What is interesting is that supply is not expanding aggressively even as demand rises quickly. Samsung Electronics and SK hynix plan to slightly reduce or keep at a limited level their NAND flash wafer output this year compared with last year. They are prioritizing capital expenditures for high-revenue DRAM products such as HBM, and production efficiency is inevitably declining as they transition processes to QLC (quad-level cell) to address large-capacity SSDs for AI data centers. Industry sources say, "For NAND flash now, defending prices and restoring profitability is more rational than pushing excessive output expansion."
This supply-demand structure is likely to support the upward trend in NAND flash prices for the time being. In particular, the fact that storage capacity per AI accelerator and per server is increasing rapidly shows that the nature of NAND flash demand is changing. Beyond simple data archiving, SSDs are being incorporated into the infrastructure that stores and manages the large-scale intermediate data and context information generated during AI inference.
The China factor is also at play. While Chinese-made general-purpose NAND flash supply is increasing, centered on YMTC, entry into the SSD market for AI data centers—which requires high performance and high reliability—remains limited due to U.S. export controls and technological barriers. In response, Samsung Electronics and SK hynix are reducing the share of general-purpose NAND flash for mobile and PCs, adjusting their product mix around server and enterprise SSDs to defend profitability. While mindful of China's low-price offensive, their strategy is to focus on areas with higher technological and reliability barriers rather than getting dragged into short-term price competition.
The current rise in NAND flash prices is interpreted less as a simple market rebound and more as the result of intertwined shifts in the AI industry's structure, supply strategies and the China factor. As AI evolves from a "computing technology" to a "technology that remembers and judges," the role of storage is moving from the periphery to core infrastructure. Attention is on whether NAND flash, long treated as an albatross, can change its standing thanks to the new use cases presented by Nvidia, and whether this trend will become a new revenue pillar for Samsung Electronics and SK hynix after DRAM.
A semiconductor industry official said, "As the AI trend shifts from training to inference, the strategic importance of storage such as NAND flash and enterprise SSDs is growing," adding, "NAND flash, overshadowed by HBM, could become a major pillar in expanding profitability for Korea's memory industry."