Show summary Hide summary
The Roundhill Memory ETF (DRAM) hit a 52-week high of $62.68 on May 27, 2026, extending its record run since launching on April 2, 2026. The memory chip shortage, fueled by explosive AI data center demand, has created a structural supply-demand imbalance that is rewarding memory manufacturers and the ETF’s core holdings—SK Hynix, Micron, and Samsung—with unprecedented valuations and revenue growth.
🔥 Quick Facts
- DRAM ETF reached $62.68 on May 27, 2026—highest level since inception
- Fastest-growing ETF ever—hit $6.5 billion in assets in its first six weeks
- 70% of global memory chips destined for AI data centers in 2026
- Memory supply growth just 16% YoY—well below historical norms of 30%+
- SK Hynix crossed $1 trillion market cap on May 27 amid AI boom
Why Memory Became AI’s Bottleneck
The AI infrastructure buildout has created an unusual dynamic: memory chips are now more critical than processors for training and running large language models. Artificial intelligence systems require massive amounts of high-bandwidth memory (HBM) and standard DRAM to handle parallel processing and cache data. This fundamentally changed how semiconductor fabs allocate production capacity.
For decades, DRAM was a commodity where supply tracked demand smoothly. Now, AI data centers—operated by hyperscalers like OpenAI, Google, and Meta—are consuming memory at rates that exceed manufacturing capacity. Counterpoint Research confirmed that 70% of memory chip production is now reserved for AI infrastructure, leaving consumer electronics and traditional computing starved for supply.
Oil price drops to $96.28 per barrel as Iran deal progress eases supply tensions
DRAM stock hits record high at $62.68 as memory chip shortage fuels AI demand
Record Valuations Driven by Structural Supply Deficit
The Roundhill Memory ETF (DRAM) has become the fastest-adopted ETF since cryptocurrency products launched years earlier. Investors poured $10 billion into the fund since April 2, 2026, recognizing that memory chip makers face years of elevated pricing and demand. The ETF’s composition—with SK Hynix at 25-27%, Micron at 25-26%, and Samsung at 20-21%—gives investors diversified exposure to the global memory duopoly.
SK Hynix, South Korea’s largest memory manufacturer, achieved a historic milestone on May 27, 2026, joining the $1 trillion market cap club alongside Micron Technology. Both companies dramatically outpaced broader semiconductor indices this year due to their position supplying high-margin memory products to AI infrastructure customers. As a reference point, Marvell Technology surged 131% on AI demand, but memory stocks have outpaced even that stellar performance.
Pricing Dynamics: The “Memflation” Effect
Gartner’s April 2026 forecast predicted that semiconductor revenue would grow 64% in 2026—but memory revenue was expected to triple. This tri-fold increase stems from both volume growth and what analysts now call “memflation”—structural inflation in memory prices caused by AI demand.
| Metric | 2025 Baseline | 2026 Projection |
| DRAM Price Growth | +60% (Q1 2025) | +90–130% (2026) |
| Memory Revenue Growth | Baseline | 3x increase |
| Supply Growth (DRAM) | Historical avg 30% | 16% YoY |
| AI Data Center Share | ~40% | 70% of output |
| PC/Smartphone Impact | Stable | –10.4% (PC), –8.4% (phones) |
Prices have risen sharply at every tier. Q1 2026 saw DRAM contract prices surge 60% quarter-over-quarter, with TrendForce forecasting an additional 90–95% increase in the same period. Storage (NAND) prices are even more extreme, climbing 234% in 2026 according to Gartner. This dual squeeze—on both memory and storage—means that PC manufacturers face 17% price increases to consumers, driving down unit shipments.
“Memory chips have become a core pillar of AI infrastructure. The upcycle could last three to five years due to structural demand for data centers and AI systems.”
— SK Hynix Executive Leadership, May 2026
What This Means for AI Infrastructure Stocks
The memory shortage is working in a two-way amplification loop. AI data centers require exponentially more memory per unit than traditional data centers. This drives memory demand higher. Higher prices reduce profitability for AI service providers unless they pass costs to customers—which raises prices for AI cloud services and delays adoption in price-sensitive segments. Yet enterprise and hyperscaler demand remains inelastic. Companies building AI infrastructure will not reduce memory allocations just because prices doubled.
This structural advantage benefits memory manufacturers faster than it reaches AI application providers. Super Micro Computer reported 123% revenue growth by absorbing these costs into system pricing, but memory suppliers capture margin expansion directly.
Will the Cycle Peak or Sustain?
Semiconductor cycle history suggests peaks followed by corrections. However, 2026 appears structurally different. Unlike past memory booms, which were driven by cyclical PC or smartphone upgrades, the current cycle is anchored by AI compute infrastructure—a multi-year capital expenditure wave. Analysts predict the upcycle will persist through 2028–2029 as countries and companies invest in domestic AI clusters and data center buildouts.
Supply cannot ramp fast enough to reverse the dynamic. Building new memory fabs costs $20+ billion and takes 3–4 years. SK Hynix and Samsung have announced expansion plans, but new capacity won’t meaningfully arrive until 2028. Until then, DRAM and NAND shortages will persist, keeping prices elevated and memory stock valuations supported.
Is the DRAM ETF Bubble or Blueprint?
The Roundhill Memory ETF’s rapid ascent—from zero assets to $10 billion in 49 days—raises natural skepticism about whether retail investors are chasing a bubble. However, the underlying thesis holds merit: AI demand is structural, supply is constrained, and prices have room to rise further. Gartner estimates memory prices could climb another 50–60% before equilibrium returns.
The risk is not whether memory stocks will decline—it’s that prices peak unexpectedly if new capacity comes online faster than expected, or if AI adoption slows due to regulatory or economic headwinds. For long-term investors, memory ETFs offer diversified exposure to three global leaders without single-stock risk. For traders, near-term momentum may persist until Q3 2026 earnings reset expectations.
