Phison CEO Warns NAND Supply May Stay Tight for 4 Years as AI Storage Demand Accelerates

Phison CEO Pua Khein Seng has issued another warning about the global NAND Flash market, arguing that new production capacity may take approximately 4 years to reach mass production while artificial intelligence continues accelerating demand for enterprise storage. According to comments reported by Commercial Times, Pua initially believed manufacturers could bring meaningful new NAND capacity online within around 2 years, but executives from NAND producers told him the process could require twice that time.

"I originally estimated it would take about 2 years, but the answer they gave was 4 years."
— Quote by: Pua Khein Seng

The warning does not necessarily mean NAND supply will remain at exactly the same shortage level until 2030. Instead, Pua is highlighting how long completely new production capacity can take to move from investment through construction, equipment installation, qualification, and eventual volume manufacturing. Phison is already building strategic NAND inventory to protect enterprise SSD supply for cloud service providers and AI customers, with the company increasingly prioritizing storage infrastructure connected to generative AI and AI agents.

AI inference is becoming an especially important part of that equation. Training requires enormous compute resources, but inference at scale also generates and accesses rapidly growing quantities of model data, context, vector databases, KV cache, and other information that must ultimately be stored. Enterprise SSDs are therefore becoming an increasingly important part of AI infrastructure rather than simply secondary storage behind GPUs and HBM. DIGITIMES reports that Pua sees this transition as a major opportunity for Phison as NAND remains constrained for years.

The market has already moved dramatically since Phison first began warning about tightening supply. NAND TLC pricing had already increased between 50% and 75% within several months as enterprise demand accelerated. Phison has since reported a record 65.3% gross margin during Q2 2026, although higher NAND procurement costs mean the company does not expect every increase in flash pricing to translate directly into higher profitability.

There is also a more optimistic interpretation of the supply outlook. TrendForce previously estimated that the NAND shortage could begin easing during 2H 2027, driven by process migrations, higher density NAND, additional Chinese production, and weaker consumer electronics demand. That does not directly contradict Pua's 4 year figure because suppliers can increase bit output from existing factories much sooner than they can construct completely new production capacity.

China is also becoming increasingly important to the supply equation. YMTC reached approximately 14% of global NAND bit shipments during Q2 2026, becoming the third largest supplier by shipment volume, while additional manufacturing capacity is still being developed. Even so, enterprise SSDs already represented approximately 48% of NAND bits shipped during Q2, illustrating how rapidly AI and data center workloads are absorbing available production.

Phison is responding through what it calls its Phison 3.0 strategy, shifting more attention toward AI Storage, Edge AI, enterprise SSDs, and complete AI platforms rather than depending primarily on conventional NAND controllers and storage modules. With agentic AI expected to increase storage requirements substantially, the company sees the current supply imbalance not only as a challenge but as an opportunity to reposition itself deeper inside the AI infrastructure market.

The important distinction is that a new NAND fab taking 4 years does not automatically mean SSD shortages will remain equally severe until 2030. Manufacturers can still increase effective supply through higher layer counts, denser QLC NAND, process migrations, and improvements in yield.

The problem is that AI storage demand is moving just as quickly. If enterprise SSDs, inference storage, and technologies such as context memory continue absorbing more NAND, additional bit output may simply be consumed before it can meaningfully reduce prices for consumer SSDs. For PC builders, the era of consistently falling SSD prices may therefore be much less predictable than it was only a few years ago.

Do you think SSD prices will remain elevated through 2030, or will higher density NAND and additional Chinese production restore supply much sooner?