NAND Flash Storage Shortage Could Extend a Decade, Phison CEO Warns | SuperSSD

The NAND industry – and semiconductor industry as a whole – may be facing a prolonged supply constraint that could reshape technology costs for consumers and businesses alike. Pua Khein-Seng, CEO of Phison (a key producer of SSD controllers), has issued a troubling forecast suggesting that NAND flash memory shortages beginning in 2026 could persist for up to ten years. This is driven primarily by the rapid expansion of AI infrastructure.

The Core Prediction

In a recent interview with CommonWealth Magazine, Pua stated that NAND flash will face severe shortages starting next year, with supply remaining tight throughout the coming decade. This timeline extends significantly beyond the one to two-year shortage periods that other industry observers have projected.

Phison manufactures NAND memory controllers found in solid-state drives, memory cards, and USB storage devices from major brands. The company’s position in the supply chain gives it visibility into both production capacity and demand trends across the industry.

According to Pua, the foundations for this shortage were laid years ago. Between 2019 and 2020, NAND manufacturers reduced capital expenditures after repeated cycles where increased production led to price collapses and poor ROI. The caution meant that capacity expansion slowed precisely when it would have been needed to meet future demand.

To make matters worse, the situation was compounded when major memory manufacturers like Micron Technology and SK Hynix redirected substantial capital toward high-bandwidth memory production starting in 2023. HBM, which is essential for AI training applications, offered more attractive profit margins, leaving traditional NAND flash with reduced investment.

The AI Data Center Factor

The immediate driver of increased NAND demand is the AI industry’s shift from model training to inference operations. During the training phase, AI systems primarily require HBM and powerful GPUs. However, the inference phase, where LLMs are deployed to process user queries and generate outputs, creates substantial storage requirements.

Pua emphasized this dynamic in his interview: “Since 2022, cloud companies have been scrambling to secure GPUs, mainly to train models. Training uses HBM and has little to do with flash. Now those models appear to have matured. After burning through so much capital, how will cloud providers make money back? Through inference.”

He continued, explaining that to monetize their AI investments, cloud providers need users, users generate data, and data requires storage. This creates what he describes as a fundamental need for data centers to expand storage capacity over the next decade, since storage remains a core function of these facilities.

The Shift from HDDs to SSDs

A major factor in Pua’s long-term forecast is the anticipated transition from hard disk drives to solid-state drives in data center environments. In 2020, SSDs were less than ten percent of data center storage, with HDDs comprising more than ninety percent. That ratio has already shifted to about 20 percent SSDs and 80 percent HDDs.

Pua projects that SSDs will eventually account for 80 to 100 percent of data center storage – a transition being accelerated by converging economics. In 2020, HDDs cost around $0.016 per gigabyte while SSDs cost $0.020 per gigabyte. Today, SSD prices have fallen to $0.05-$0.07 per gigabyte, while HDD prices have risen to $0.02 per gigabyte. Industry projections suggest SSD pricing could match HDD pricing within five to eight years.

Beyond cost, SSDs offer advantages in speed, energy efficiency, and reliability that make them increasingly attractive for data centers handling AI workloads. Additionally, HDD supply constraints have emerged, with some providers reporting lead times extending beyond one year, making SSDs a more practical option for capacity expansion.

Price Impact on Consumer Electronics

A sustained NAND shortage would have cascading effects throughout the consumer electronics market. Flash memory is a fundamental component not only in computers and smartphones but also in graphics cards, tablets, cameras, and even devices with less visible storage requirements such as monitors and headphones.

Higher NAND prices would translate to increased costs across essentially all electronic devices. Several major storage manufacturers, including SanDisk, Western Digital, and Micron, have already announced price freezes or increases across their product lines as demand has reached what industry observers describe as unprecedented levels. High-capacity consumer SSDs could become prohibitively expensive.

Industry Response

Pua’s forecast aligns with emerging perspectives from financial analysts. Investment bank Morgan Stanley has advised clients about an impending memory supercycle, driven by continued HBM demand and broader memory market dynamics. However, Pua’s ten-year timeline for NAND specifically extends beyond most current market predictions.

The question facing the industry is whether manufacturers can scale production quickly enough to close the supply gap. Building semiconductor fabrication facilities is a multi-year process requiring billions of dollars in capital investment. The challenge is compounded by the fact that many memory makers may prefer to focus resources on higher-margin HBM production rather than traditional NAND flash.

Not a 100% Certainty

Several factors could influence whether Pua’s forecast proves accurate. Technology advancement remains a variable, as manufacturers continue developing higher-density NAND through additional 3D layers and improved cell architecture. However, scaling has become more challenging as the industry has already transitioned from single-level cells to quad-level cells and implemented 3D NAND technology.

The trajectory of AI demand itself is another uncertainty. If AI adoption continues at current rates or accelerates, storage requirements could indeed remain elevated for years. Conversely, if AI investment slows or proves less profitable than anticipated, demand could moderate.

It should also be noted that Pua’s position as CEO of a major memory controller supplier means he has a vested interest in the industry’s health and investment levels. His forecast could be intended to encourage capacity expansion or prepare stakeholders for challenging market conditions.