The global memory chip market remains locked in a severe supply-demand imbalance that shows no signs of abating anytime soon. According to JPMorgan strategist Jay Kwon, the AI-driven memory crisis will extend for at least two more years as demand broadens from graphics processing units (GPUs) into central processing units (CPUs) and other enterprise infrastructure.
The Structural Supply Crunch
In a research note published Monday, Kwon emphasized that the memory total addressable market (TAM) is expanding due to both pricing power and volume growth. “Supply-demand shortage continues for the next two years,” he wrote. The strategist argues that investors have underestimated how memory demand is migrating beyond AI accelerators into mainstream compute, creating a sustained upward revision cycle for memory manufacturers.
This isn’t a typical cyclical downturn. The proliferation of large language models, inference workloads, and enterprise AI deployments has fundamentally altered the demand curve for DRAM and NAND flash. Cloud service providers (CSPs) are securing multi-year supply agreements, giving manufacturers unprecedented revenue visibility.
Why This Time Is Different
Historically, memory markets have been notorious for violent boom-bust cycles driven by capital expenditure timing and inventory corrections. However, several structural factors suggest this upcycle has more staying power:
- AI demand broadening: Inference workloads on CPUs are becoming a major new driver alongside GPU training clusters.
- Disciplined capacity: Major suppliers (Samsung, SK Hynix, Micron) have maintained capital expenditure restraint, avoiding the overcapacity traps of past cycles.
- Advanced node complexity: Transition to 1-alpha/1-beta DRAM and 200+ layer NAND increases production difficulty and limits rapid supply response.
- Long-term contracts: Hyperscalers are locking in supply through 2027-2028, reducing spot market availability.
JPMorgan’s Bullish Stance on Memory Equities
Despite a 25% correction in memory shares during Q3 2026 amid peak-out sentiment, Kwon remains constructive. He identifies three temporary headwinds: near-term EPS misses, slower-than-expected CSP AI capex, and mainstream memory content optimization. However, he sees “strong upside to memory stock prices trading at earnings and P/E valuation metrics on a smoothening earnings trajectory from long-term agreements and a higher shareholder return profile.”
Kwon specifically favors:
- Micron Technology (MU) – U.S. pure-play with strong HBM (High Bandwidth Memory) positioning
- Kioxia (KI5.F) – Japanese NAND leader with enterprise SSD exposure
- Winbond (2344.TW) – Taiwanese specialty memory player benefiting from automotive/industrial demand
AlphaSpace Data Point: Micron’s 202% Surge
According to Yahoo Finance AlphaSpace data, Micron shares have exploded 202% year-to-date versus a 13% advance for the S&P 500 (^GSPC). The stock has pulled back roughly 10% over the past month on valuation concerns, but the fundamental backdrop remains robust. Micron’s HBM3E qualification with NVIDIA and expanding data center SSD portfolio position it as a primary beneficiary of the AI memory supercycle.
The Bottom Line: Four-Year Visibility
SanDisk CEO David Goeckeler reinforced the bullish narrative on his recent earnings call: “We spent a lot of time over the last two or three quarters really working very deeply with our largest customers on committing demand. We have over four years of visibility now. We feel very good about where the franchise is.”
That level of forward visibility is extraordinary for a historically cyclical industry. It suggests the memory shortage may persist even beyond JPMorgan’s two-year horizon as AI infrastructure build-out continues globally.
FAQ: Memory Chip Supply Crisis
1. Why can’t memory manufacturers just build more factories to meet demand?
Building a leading-edge memory fab costs $15-20 billion and takes 3-4 years to ramp. The extreme complexity of sub-10nm DRAM and 200+ layer 3D NAND means yield learning curves are steep. Additionally, the top three suppliers (Samsung, SK Hynix, Micron) control ~95% of DRAM supply and have demonstrated unprecedented capital discipline, prioritizing returns over market share gains.
2. How does High Bandwidth Memory (HBM) change the supply equation?
HBM requires 2-3x more DRAM die per unit compared to standard server DRAM, along with advanced packaging (TSV, interposers) that creates additional bottlenecks. As AI accelerators shift from HBM2E to HBM3/3E/4, each generation consumes more wafer capacity. This structurally reduces bit supply available for mainstream PC/server markets.
3. What could break the bullish thesis for memory stocks?
Key risks include: (a) A sharper-than-expected slowdown in AI capex from major hyperscalers (Microsoft, Google, Amazon, Meta), (b) Chinese memory producers (CXMT, YMTC) achieving faster-than-expected yield maturity on advanced nodes, or (c) A global recession triggering enterprise IT spending cuts. However, current order books and multi-year contracts provide substantial downside protection through 2026.
