The memory industry has been under pressure for months, but Phison’s latest warning significantly raises the tone of the debate. Its founder and CEO, K.S. Pua, has suggested that the memory shortage, especially in NAND Flash, might not be fully resolved for up to a decade if demand related to Artificial Intelligence continues at its current pace. This is no minor forecast or isolated comment: it comes at a time when the market is already dealing with sharp price hikes, limited capacity, and increasing pressure from AI data centers across the entire supply chain.
The key point is that Phison is not merely talking about a bullish phase within a typical cyclical market. What its top executive is describing is a structural imbalance between demand surging due to AI deployment and an industrial capacity that cannot grow at the same rate. According to various reports from the Asian and U.S. tech sectors, Pua indicates that the volume of data and memory required for the new AI infrastructure is advancing far beyond manufacturers’ expansion capabilities. His conclusion is stark: as long as AI applications do not slow down, memory demand will not weaken significantly.
This message aligns, at least in part, with much more cautious but also very firm forecasts from firms like Gartner. The consulting firm predicts that DRAM prices will increase by 125% in 2026, and NAND flash prices by 234%, not expecting meaningful relief until late 2027. Gartner does not endorse a ten-year outlook, but confirms that memory inflation will be one of the major factors shaping the tech market this year, potentially delaying or harming some non-AI-related demand until 2028. This provides context to Phison’s warning: it may sound extreme, but it’s not coming from nowhere.
A very aggressive warning, but not entirely isolated
It’s important to clarify the scope of these statements. Phison is not a major wafer manufacturer like Samsung, SK hynix, or Micron. Its role is more focused on controllers and storage solutions, giving it very direct insight into order volumes, availability, and price trends, but not making it a sole market arbiter. Still, its messages carry weight because they come from a company deeply connected to NAND supply chains and corporate and consumer clients. Additionally, this is not the first time Pua has warned of an exceptional memory situation. As early as late 2025 and early 2026, he predicted a “supercycle” of memory and significant supply pressure.
Recently, the discourse has escalated. According to the latest sector reports, Phison considers the second half of 2026 to be particularly challenging, with a Q4 scenario where “even with money,” stock of NAND could become very limited in some segments. The company is reacting in a way uncommon for its profile: increasing funding, stockpiling inventory, and attempting to secure supplies before the bottleneck worsens. Digitimes recently reported that Phison is seeking up to $1.4 billion in financing to address this tension, while other industry sources indicate its NAND inventory exceeds 50 billion Taiwanese dollars.
The underlying reason is clear. AI is consuming an ever-growing portion of high-value memory supply. It’s not just HBM for accelerators, but also DRAM for servers, enterprise SSDs, and NAND in various storage layers associated with data centers and inference systems. Micron, for example, has pointed out that memory and storage are becoming strategic assets within the new AI stack, and demand from data centers remains extremely strong. TSMC’s revenue forecasts also reflect this, with AI-driven demand sustaining high growth rates into 2026.
The real risk is not just price
The most striking aspect of Phison’s message isn’t just the expected duration of shortages but their potential impact on the consumer market. The company warns that memory allocated for AI and cloud infrastructure has high economic priority for major manufacturers, which could weaken PC, mobile, TV, and consumer electronics sectors. As early as February, reports featured statements from Pua noting that many consumer electronics manufacturers might be forced to pull certain product lines if supply cannot be assured. While this scenario is not universally confirmed across the industry, it underscores fears of a divided industry: one well supplied due to AI needs and another much more vulnerable to price shocks and delivery delays.
This is arguably the most critical point. Memory costs aren’t rising solely due to supply shortages; the market is reassigning value and capacity toward areas with greater demand, urgency, and growth potential. In this context, small and medium enterprises relying on NAND or DRAM for mass-market products have less bargaining power than hyperscalers, large server manufacturers, or AI ecosystem providers. Pua phrases this bluntly because he’s close to that imbalance, but broader market predictions also reflect this trend in more conservative estimates.
Can it really last ten years?
This is the big question, and honestly, no one knows for sure. The ten-year figure is not a consensus forecast but a very aggressive warning from Phison’s CEO. It’s entirely possible the market will find equilibrium sooner thanks to new plants, architectural shifts, efficiency improvements, demand moderation, or a different distribution of AI spending. Gartner, for instance, emphasizes that while “memflation” is deep, it’s not eternal.
What seems increasingly clear is that memory shortages won’t merely last a few quarters. AI has changed the nature of the market. What used to be predominantly driven by cycles of supply and price is now also shaped by strategic considerations. Memory is no longer just a component; it’s a critical resource of AI infrastructure. When that happens, normalization timelines depend not only on production capacity but also on who can pay more, who gets priority, and which sectors are deemed more profitable or strategic.
In that light, Phison’s warning may be exaggerated in timing, but not in fundamental diagnosis. AI’s pressure on memory is real, affecting prices, inventories, and procurement strategies, and could sustain this impact well beyond what much of the electronics industry might expect. The debate no longer revolves around whether tensions will occur, but for how long and who will be able to endure them best.
Frequently Asked Questions
Has Phison officially stated that the memory shortage will last ten years?
There have been statements attributed to its CEO, K.S. Pua, in industry outlets suggesting that the imbalance might not fully resolve in the next decade. However, this figure is not a sector-wide consensus but a particularly stark warning from Phison.
Which part of the memory market is most affected by AI?
The pressure impacts HBM, server DRAM, and enterprise NAND, but Gartner also forecasts significant increases in DRAM and NAND prices across the board in 2026, which could impact consumer products and segments not directly tied to AI.
When might memory prices start to ease?
According to Gartner, a meaningful relief is not expected before late 2027. That doesn’t mean shortages will disappear then, but current steep increases should not continue at the same pace indefinitely.
Why is NAND particularly worrisome in 2026?
Because AI, cloud computing, and data center demands are progressively absorbing more capacity. Experts like Phison warn that, especially in the second half of the year and Q4, supply could become very tight—even for financially capable clients.
Sources:
Gartner, “Gartner Forecasts Worldwide Semiconductor Revenue to Exceed $1.3 Trillion in 2026”
Digitimes, “Phison eyes US$1.4B fundraising to tackle severe NAND shortage in 4Q26”