DRAM sets a new all-time high again: how has the world’s first pure memory ETF delivered nearly a 150% increase within the year?

On April 2, 2026, the world's first exchange-traded fund (ETF) focused solely on pure memory chips—Roundhill Memory ETF (ticker: DRAM)—was officially listed on the Cboe BZX Exchange. Within just two months of launch, the ETF's price rose from approximately $28 at issuance to over $50, with a cumulative increase of nearly 150%. Its assets under management (AUM) surpassed $10 billion, making it one of the fastest-growing ETFs in history. As of June 2, 2026, DRAM's latest trading price was $68, with a 24-hour increase of 7.6%.

This explosive growth of the ETF is not an isolated event but a concentrated reflection of the global AI infrastructure expansion wave in the capital markets. The memory chips it tracks are currently at the most critical bottleneck in AI computing power expansion.

Why can an ETF launched only two months ago become the fastest-growing product globally?

Two months to complete a decade—DRAM ETF growth milestone

DRAM has completed the growth path that traditionally takes several years for ETFs in just about two months. Its explosive power is not only due to the rise in secondary market prices—going from $28 at issuance to over $50, which is already highly attractive—but also because the inflow of capital far exceeds expectations.

According to publicly available data, DRAM's assets under management surpassed $1 billion within 10 days of listing and exceeded $5B within 25 days. This speed broke previous records, making it the fastest ETF in history to reach $10B in AUM. By the end of May 2026, the fund's AUM stabilized at around $10.3 billion, with weekly capital inflows remaining positive.

Such strong capital attention reflects market recognition of the "pure memory theme" as a differentiated investment direction. Compared to traditional semiconductor ETFs like SOXX or SMH, which broadly cover logic chips, equipment manufacturing, and other segments, DRAM restricts its investment scope strictly to memory and storage chips, offering a more focused and pure exposure to AI infrastructure.

Why are memory chips the most critical bottleneck in current AI computing power expansion?

To understand DRAM's price performance, we first need to answer a core question: what role do memory chips play in the AI computing power system?

Enhancing AI computing power depends not only on the continuous evolution of compute chips like GPUs but also on the efficiency of data transfer between processors and storage. High-bandwidth memory (HBM) is a key supporting component of AI acceleration cards, while DRAM and NAND flash memory support server system operation and large-scale data access.

Currently, supply tightness for HBM, DRAM, and NAND flash is expected to persist beyond 2026, driven by explosive demand for high-performance memory from AI applications. However, supply-side expansion faces multiple technical bottlenecks. Specifically, the continuous increase in the wafer size of HBM's new process reduces the number of chips that can be cut from a single wafer, weakening supply elasticity; the introduction of extreme ultraviolet (EUV) lithography in advanced DRAM processes further constrains capacity ramp-up speed.

The Achilles' heel of AI computing power—tight supply and demand of memory chips

While supply is constrained, demand is still accelerating. JPMorgan's latest research report significantly raised its forecast for the global storage market size from 2026 to 2028, expecting a total market size of $1.7 trillion by 2028. Micron Technology has confirmed that all its HBM capacity for 2026 has been fully booked, significantly strengthening its pricing power. SK Hynix holds about 60% of the HBM market share, becoming a key supporter of NVIDIA's AI ecosystem.

How do the concentration and risk exposure features of pure memory-themed ETFs look?

The high concentration of DRAM is not a design flaw but an inevitable result of its thematic positioning. The ETF currently holds 20 component securities, with the top three holdings—SK Hynix, Samsung Electronics, and Micron Technology—accounting for nearly 70% of the total, with SK Hynix alone comprising about 27% to 28% of the fund's weight.

The three giants and regional distribution—an overview of DRAM ETF holdings concentration

The combined weight of South Korean companies in the ETF is approximately 52%–55% (mainly SK Hynix and Samsung Electronics), U.S. companies account for about 32%–35% (mainly Micron Technology), and the remaining exposure is distributed across Taiwan (about 7%–8%), Japan (about 3%–4%), and other regions. These three regions together account for nearly 100%. This geographic concentration also reflects the regional distribution of global memory chip capacity: Korean companies dominate HBM and DRAM, Micron in DRAM and NAND, and Taiwanese firms like Nanya Technology and Winbond are included as supplementary components.

How is the demand structure for memory evolving from HBM to DDR5?

Over the past two years, market attention to memory storage has mainly focused on HBM, as it is the direct supporting component for AI training chips. However, as AI applications evolve from training to inference and the era of intelligent agents, the category structure of memory demand is undergoing profound changes.

From HBM to DDR5—evolution of AI memory demand structure

UBS's latest report indicates that the underlying demand structure of the AI industry has begun to shift. Before 2023, large model demand mainly came from training; from 2024 to 2025, the focus is gradually shifting to inference; and starting in 2026, the industry is accelerating into the era of intelligent agents—AI that can not only answer questions but also autonomously plan, execute tasks, and call tools, leading to exponential growth in storage resource consumption.

Within this new framework, DDR5's role is becoming prominent. Intelligent agents require significant CPU involvement in task orchestration, state management, and tool invocation, with DDR5 being the core supporting memory for CPUs. UBS believes that in the coming years, the largest demand increase may come from DDR5 rather than HBM. JPMorgan's forecast also supports this view, raising its projection for server memory demand from 2026 to 2028 by 5% to 22%, with over 60% of the increase attributable to AI servers.

This means the demand structure covered by DRAM ETFs is shifting from a "single category" to a "multi-point flowering" pattern—HBM continues to be strong, DDR5 accelerates, and enterprise SSDs are rapidly expanding under AI inference demand. JPMorgan estimates that the enterprise SSD market will exceed 500 exabytes (EB) by 2026, accounting for 43% of NAND's total demand.

How do the performance gaps of the three major memory giants support the ETF's valuation?

The reason the DRAM ETF's holdings can continue to attract market attention is fundamentally because the performance growth of its underlying component companies has moved beyond traditional "cyclical fluctuations" into a growth channel driven by structural demand.

Underlying engine—overview of the performance gaps among the three major memory giants

Latest financial reports show SK Hynix's revenue increased by 198% year-over-year, with net profit up 165%, and management has raised forward guidance. Micron's quarterly revenue jumped from about $8 billion last year to over $23 billion. Samsung Electronics also benefited from its capacity advantages in HBM and DDR5, with its stock price rising over 160% this year.

Notably, all three companies have surpassed a market capitalization of $1 trillion, becoming some of the most closely watched AI infrastructure assets globally. Bloomberg data indicates Micron's net profit is expected to jump from $17k in 2025 to $66.8 billion in 2026, potentially reaching about $120 billion in 2027. If these expectations are gradually realized, the profit growth of the DRAM holdings remains highly visible.

However, from a valuation perspective, market optimism is already quite high. Currently, the forward P/E ratios for Micron and SanDisk are about 10x each, but this valuation is based on continued profit growth. Historical data shows Micron's P/E at cyclical peaks once reached 46x, and SanDisk's at 58x, indicating that current valuation expansion more reflects profit growth expectations rather than a valuation bubble.

How to verify the sustainability of the super cycle? What risks should be watched?

Any asset that experiences short-term explosive growth must answer a fundamental question: can this growth continue? For DRAM, three factors will determine its medium-term trajectory.

Sustainability of capital expenditure. The four major cloud and platform companies—Amazon, Meta, Alphabet, and Microsoft—are expected to spend up to $725 billion on AI infrastructure in 2026. Some companies are supporting this pace by increasing debt; if subsequent capital expenditure growth slows, the profitability outlook and stock prices of chip companies will be directly impacted.

Risks of the memory pricing cycle turning point. Although current contract prices for DRAM and NAND are still on an upward trend, the industry has historically been highly cyclical. TrendForce predicts that traditional DRAM contract prices in Q1 2026 could increase by 55%–60% month-over-month, reflecting high sensitivity to supply and demand. If demand growth slows marginally or supply capacity is gradually released, a retreat from high prices could significantly compress the profits of memory companies that rely heavily on pricing leverage.

Concentration risk. With about 70% of DRAM's weight concentrated in three companies, any negative news about these underlying stocks will have a significant impact on the ETF's net value. Additionally, the ETF's high dependence on the Korean market exposes it to currency exchange risks and policy environment changes.

JPMorgan admits in its report that storage stocks are still trading at a discount to earnings, mainly due to market doubts about whether the storage value share can continue to increase. However, the firm believes that AI has brought a new demand structure, and the traditional cyclical valuation framework is no longer applicable. This is essentially a proposition that requires time to verify: between the "structural turning point" and the "cyclical peak," market judgment mechanisms have yet to form consensus.

Summary

The new all-time high in DRAM prices fundamentally reflects the concentrated impact of the AI infrastructure investment wave on the capital markets. The ETF's rapid scale expansion is driven by its differentiated "pure memory" positioning and the strong performance of its underlying components. Behind this is the structural logic that memory chips are the core bottleneck of AI computing power—HBM supply shortages, DDR5 demand acceleration, and enterprise SSD expansion—forming a multi-layered demand-driven system.

However, risks such as concentration, pricing cycle turning points, and valuation sustainability remain. The optimistic expectations embedded in current prices will determine the ETF's future market performance. For market participants, understanding the structural logic and cyclical constraints of DRAM is key to evaluating the value of this emerging investment tool.

Frequently Asked Questions

What type of fund is the DRAM ETF, and what is its investment theme?

DRAM is an actively managed ETF issued by Roundhill Investments, focusing on the global pure memory theme. It was listed on the U.S. Cboe BZX Exchange on April 2, 2026. The fund invests at least 80% of its net assets in memory and storage chip companies, with a focus on HBM, DRAM, NAND flash, and similar categories, distinguishing itself from broader traditional semiconductor ETFs.

What are the main holdings of the DRAM ETF, and how concentrated are they?

Currently, DRAM holds about 20 component securities, with the top three—SK Hynix (about 28%), Samsung Electronics (about 21%), and Micron Technology (including derivatives, about 26%)—accounting for roughly 70% to 75% of the fund's total weight.

The DRAM ETF's cumulative increase is nearly 150%. What are the main drivers?

The core driver is the structural demand for memory chips driven by AI computing expansion. HBM, as a key supporting component for AI acceleration cards, is currently in tight supply with limited capacity expansion; simultaneously, as AI applications extend from training to inference and intelligent agents, demand for DDR5 and enterprise SSDs is accelerating, collectively boosting the profitability outlook and stock valuations of memory companies.

What risks should investors be aware of when investing in the DRAM ETF?

Main risks include: (1) high concentration in three companies, meaning stock-specific news can significantly impact ETF NAV; (2) the cyclical nature of the memory industry, with contract prices potentially turning; (3) potential slowdown in AI capital expenditure affecting high-capital-demand memory needs; (4) high dependence on the Korean market, exposing the ETF to currency and policy risks.

What are the expense ratio and management approach of this ETF?

The DRAM ETF is actively managed, with an expense ratio of 0.65%. The management team rebalances quarterly, adjusting holdings based on market and revenue share dynamics within the memory and storage sectors, while also adhering to a maximum of 25% weight for any single company.