Simon Chen, Chairman and CEO of ADATA, has forecast that the memory shortage will continue for the next 10 years. The remark was reported by the Commercial Times on July 20, 2026. It's a strong prediction that extends Micron's view that shortages will "persist beyond 2027" all the way to the mid-2030s. Tracing actual capacity expansion plans confirms that supply won't increase immediately, but the industry as a whole has yet to substantiate a 10-year timeframe.
The Demand Beyond Cloud That the "10-Year Shortage" Anticipates
Chen cited power—particularly green power—and memory as the resources that will be in shortest supply over the next decade. Responding to concerns that AI investment has become overheated, he countered: "After 2030, we can debate whether an AI bubble will occur in 2040 or 2050." His position is that it's premature to judge the ceiling of demand based solely on current capital expenditure by cloud providers.
His demand scenario assumes that AI will expand from data centers into the physical world. Chen cited robots, autonomous vehicles, and unmanned factories as examples. He also expects it to spread further into smart homes and low-earth-orbit satellites, driving up memory content on both the cloud and device sides. His conclusion is that even Samsung, SK hynix, Micron, plus major Chinese manufacturers combined won't be able to fully satisfy demand within 10 years.
Chen also has his own reasoning on the supply side. Major manufacturers, having experienced past market downturns, are reluctant to build capacity that exceeds demand all at once. Chinese manufacturers, too, face constraints in procuring critical manufacturing equipment, and building fabs and cleanrooms takes time. However, the Commercial Times article doesn't include a quantitative model showing supply and demand through 2036. "10 years" is not a settled industry forecast—it's the timeframe of the demand scenario Chen envisions.
HBM Is Increasing Wafer Consumption
According to SK hynix, HBM (High Bandwidth Memory) requires more wafers than conventional DRAM to produce the same memory capacity. This is because stacking multiple DRAM dies and connecting them via TSV (Through-Silicon Vias) increases die area, reducing the number of dies obtainable from a single wafer. Stacking and advanced packaging processes add further overhead.
What matters here is that memory production volume can't be measured by bit count alone. As the share of HBM for AI accelerators rises, the wafers and manufacturing floor space needed to ship the same total capacity increase. Micron similarly explains that as HBM generations advance, wafer consumption per bit rises, further squeezing supply of non-HBM products.
Samsung expects its 2026 HBM revenue to more than triple compared to 2025, and is expanding HBM4 production capacity. While the decision to increase high-value AI memory makes sense, it continues to compete with manufacturing resources that could otherwise be allocated to conventional DRAM used in PCs, smartphones, and general-purpose servers. This is where the paradox arises: the more AI demand grows, the harder it becomes to shake the sense of shortage even as production increases.
Capacity Expansion Through 2033 Won't Deliver Immediate Relief
In its June 2026 earnings briefing, Micron stated that DRAM and NAND demand significantly exceeds supply, and that tightness will persist beyond 2027. While it expects supply to gradually improve starting in 2028, it says it cannot foresee when supply will catch up with demand. New fabs require securing skilled labor, permits, and power infrastructure. Time is also needed to bring in manufacturing equipment and ramp up production.
| Company/Site | Supply Expansion Milestone | Disclosed Scope |
|---|---|---|
| Micron Idaho Fab 1 | Mid-2027 | First wafer output |
| Micron Idaho Fab 2 | End of 2028 | First wafer output |
| Micron Taiwan Tongluo | Mid-2027 | Full-scale shipments from existing fab |
| SK hynix Yongin | 2033 | Target completion of Fab 4 construction |
This timeline shows the gap between announcements of fab construction and the point at which sufficient volume actually reaches the market. SK hynix moved up the completion of its Yongin Fab 4 by 12 years, from 2045 to 2033. Even so, the company has determined that Yongin alone won't be enough to meet future demand, and has laid out a phased investment plan totaling 1,100 trillion won, including Cheongju and southwestern Korea. Since developing Yongin took roughly nine years, the next major site is unlikely to ramp up quickly either.
Customers, too, have moved toward long-term contracts. Micron has signed 16 strategic customer agreements centered on 2026 through 2030, covering approximately 20% of the DRAM volume and one-third of the NAND volume it plans to sell over that period. As more contracts lock in future supply years in advance, the shortage felt by spot-market buyers and customers without such contracts could persist longer than the overall recovery in production volume.
ADATA's Position: Profits Rise Along with Prices
ADATA doesn't manufacture DRAM chips itself; it's a company that procures chips from upstream manufacturers and sells them as modules and SSDs. In Q1 2026, the company posted revenue of NT$26.1 billion and operating profit of NT$12.280 billion, with a gross margin reaching 55.69%. DRAM accounted for 68% of revenue composition, while SSDs made up 25%.
Inventory secured before price increases has a major impact on profitability. ADATA's inventory grew from NT$23.557 billion in Q4 2025 to NT$36.438 billion in Q1 2026. The company has further indicated plans to build inventory to over NT$50 billion by Q3. Revenue for May 2026 also came in at NT$12.943 billion, up 210.44% year-over-year.
Accordingly, statements anticipating sustained long-term price increases and supply shortages are directly tied to ADATA's procurement strategy and earnings outlook. This isn't a reason to dismiss the prediction, but it should be read separately from chip manufacturers' production plans. What primary sources support is that: tightness will persist beyond 2027, new fab ramp-ups will span from 2027 to 2033, and HBM consumes significantly more wafers.
Whether the "10 years" prediction becomes reality can be first assessed by whether the supply improvements Micron projects from 2028 onward outpace demand growth. In making that assessment, what matters isn't the number of new fabs completed, but actual wafer output volume and the share of increased capacity that gets allocated to conventional DRAM.