For the past two years, markets focused largely on compute, the GPUs, hyperscalers, and cloud infrastructure powering large language models. Now, investors are shifting toward what many analysts increasingly call AI’s real bottleneck and it is memory.
The newly launched Roundhill Investments Roundhill Memory ETF, trading under the ticker $DRAM, has surged roughly 98% since its April 2026 debut and amassed more than $6 billion in assets under management (AUM) in just over five weeks, making it one of the fastest-growing ETF launches on record.
The rally reflects a deeper structural change in AI infrastructure economics. Training and running advanced AI systems increasingly depends not just on processing power, but on high-bandwidth memory (HBM) , the specialized chips that allow AI accelerators to move enormous amounts of data with minimal latency. In practical terms, memory has become the plumbing of generative AI.
Why Memory Suddenly Matters So Much
In AI systems, memory chips determine how quickly models can access and process information. As models become larger and inference workloads expand across enterprises, demand for faster DRAM and NAND storage has exploded.
That has transformed what was historically one of the semiconductor industry’s most cyclical and commoditized segments into one of its hottest growth areas.
Companies such as SK hynix, Samsung Electronics, and Micron Technology are now central beneficiaries of the AI buildout because they dominate the global HBM market. Analysts increasingly expect supply shortages in advanced memory chips to persist into 2027 or even 2028.
That dynamic helps explain the extraordinary momentum behind DRAM.
Unlike broader semiconductor ETFs, which remain heavily tilted toward logic chips and GPU names, DRAM offers concentrated exposure to the memory layer of the AI stack. The ETF also gives U.S.-listed investors access to Asian champions that are difficult to capture through traditional semiconductor funds.
Inside the DRAM ETF
As of today, the ETF’s largest holdings were heavily concentrated in a handful of dominant memory producers. According to Roundhill data, the top positions included:
Combined, Micron, Samsung, and SK hynix account for roughly three-quarters of the ETF’s exposure.
That concentration has amplified returns, but also raises volatility risks. The ETF has already experienced sharp swings amid broader semiconductor selloffs, underscoring that thematic AI trades remain highly momentum-driven.
AI Exposure Is Expanding Closer to Home
The enthusiasm around AI-themed ETFs is not confined to U.S. markets.
On the Abu Dhabi Securities Exchange (ADX), the KraneShares Artificial Intelligence & Technology ETF (AGIX) has also attracted attention from regional investors. Since listing on ADX in July 2026, AGIX has gained roughly 20%, reflecting strong regional appetite for AI-linked equity exposure.
AGIX differs meaningfully from DRAM. While DRAM is effectively a targeted bet on AI memory infrastructure, AGIX takes a broader approach across the AI ecosystem, including cloud computing, semiconductors, data infrastructure, and private AI-linked companies with wider diversification.
For GCC investors, the rise of products like AGIX is important because it signals the increasing localization of thematic ETF access. Investors in the UAE can now gain exposure to global AI trends through locally listed vehicles on ADX, avoiding some of the friction associated with offshore brokerage access and U.S.-domiciled products.
At the same time, DRAM’s explosive rise highlights how quickly ETF narratives can evolve inside the AI trade. Just as investors once focused almost exclusively on GPUs, markets are now repricing the less glamorous but increasingly indispensable layer beneath them.
In AI, memory is becoming the constraint that defines the next stage of growth.
