The Roundhill Memory ETF (DRAM) has crossed $20 billion in assets under management, a milestone that underscores a growing appetite among investors for targeted bets on the hardware powering artificial intelligence. The fund, which tracks companies focused on memory chips and related technology, has ballooned in size as firms from cloud giants to enterprise buyers race to build out AI systems — and the memory to run them.
Why memory is suddenly a hot ticket
Memory chips — the components that store and rapidly retrieve data — have long been a commodity business, subject to brutal boom-and-bust cycles. But the AI boom changed that. Large language models and other deep-learning workloads demand huge amounts of high-bandwidth memory, which sits close to the processor to feed it data as fast as possible. Without enough memory, even the most advanced GPU is worthless.
DRAM's rapid growth highlights how investors are looking beyond the flashiest names in AI, like Nvidia or OpenAI, and drilling into the enabling infrastructure. The fund holds stakes in memory makers, chip-equipment suppliers, and firms that design memory controllers. Its $20 billion haul signals that a broad base of buyers — institutional and retail — now see memory as a core AI play, not just a cyclical semiconductor subsector.
How DRAM got so big, so fast
Launched just a few years ago, the ETF quietly accumulated assets as AI spending surged. The fund's structure gives it exposure to both established players and smaller specialists. The biggest weighting is in companies that produce dynamic random-access memory chips, the kind used in servers and data centers. But DRAM also includes firms that make the machines that fabricate those chips, as well as companies that design the intellectual property for memory circuits.
That breadth has helped the fund capture gains across the supply chain. As hyperscalers like Amazon, Microsoft and Google poured billions into new data centers, they ordered more servers, and each server needed more memory. The ETF tracked that wave. The $20 billion mark is more than a number — it's a vote of confidence that the memory buildout has room to run.
What the milestone says about AI investing
The crossover also reflects a shift in how money flows into AI. Early on, investors piled into the companies making the most noise — the chatbot builders, the cloud platforms, the chip designers. Now, as AI moves from research labs to production, the focus is widening to the physical layer: the factories, the raw materials, the components that make inference and training possible.
Memory is a particularly tight bottleneck. Industry estimates suggest that AI servers require several times more memory than traditional servers. That demand has pushed prices up and margins higher for memory makers, which were struggling just a few years ago. The ETF's performance tracks that turnaround. For investors, the question is no longer whether memory matters for AI, but how much more of it the world will need.
The next test for DRAM will come in the quarterly earnings reports from its largest holdings. If memory demand continues to outstrip supply, the fund's growth could accelerate further. If a glut emerges, the same cyclical forces that once punished the sector could resurface. The fund's managers have not publicly commented on the $20 billion milestone, and the firm declined to provide additional details beyond the asset figure.
For now, the ETF sits at a symbolic threshold. Whether it crosses $30 billion or stalls will depend on how fast AI infrastructure expands — and whether memory remains the bottleneck that investors believe it is.
