KEY POINTS
- The Roundhill Memory ETF (DRAM) has surged approximately 107% since its April 2 launch and reached $10 billion in AUM faster than any thematic ETF in history, driven by explosive demand for HBM and DRAM chips powering AI data centers.
- The fund gained 31% in the past week alone as the semiconductor rebound disproportionately benefited memory names like Micron and SK Hynix, which are capacity-constrained on high-bandwidth memory production.
- Traders should watch whether DRAM can hold its gains through the next earnings cycle and whether the fund's narrow focus on memory semiconductors creates amplified drawdowns during the next sector correction.
The Roundhill Memory ETF has returned approximately 107% since it began trading on April 2, making it the best-performing ETF launch of 2026 by a wide margin. The fund, which trades under the ticker DRAM on the CBOE, reached the $10 billion AUM milestone faster than any previous thematic fund in ETF history, and its trajectory shows no sign of slowing: assets under management have climbed from roughly $6.5 billion in late April to north of $12 billion today.
The fund's thesis is simple and specific: it owns companies that derive at least 50% of their revenue from semiconductor memory products, including high-bandwidth memory (HBM), dynamic random-access memory (DRAM), and NAND flash. That means concentrated exposure to names like Micron Technology, SK Hynix, Samsung Electronics' memory division, and a handful of specialized suppliers — the companies building the chips that sit inside every AI accelerator and data-center server on the planet.
Why Memory Is the AI Bottleneck
The AI infrastructure buildout has created a supply-demand imbalance in memory that is more acute than in any other part of the semiconductor stack. Every Nvidia H100 and B200 GPU requires multiple stacks of HBM chips to function, and the transition to next-generation HBM3E and HBM4 technologies has further tightened available capacity. Micron, SK Hynix, and Samsung are the only three companies in the world capable of manufacturing HBM at scale, and all three have reported that their production is sold out through 2027.
That supply constraint is the engine driving DRAM's performance. When Micron surged 9% this week on strong AI memory demand and confirmation that its HBM3E ramp is running ahead of schedule, DRAM captured the full benefit of that move with no dilution from non-memory holdings. The fund's narrow focus is a feature, not a bug, when the trade is working.
The semiconductor rebound this week disproportionately favored memory names. DRAM gained 31% over the past five trading days and 72% over the past month, outperforming the broader VanEck Semiconductor ETF (SMH) by a factor of three. The reason is straightforward: memory companies have the most direct exposure to AI capex, and the hyperscaler commitment to spend $750 billion on data-center infrastructure in 2026 flows through the memory supply chain before it reaches almost any other component.
The Risk of Concentration
Thematic ETFs that double in two months deserve both admiration and caution. DRAM's concentrated portfolio means it behaves more like a leveraged bet on AI memory demand than a diversified sector fund. During last week's semiconductor selloff, DRAM fell harder than SMH because its holdings are more volatile and more correlated with each other than a broader chip basket.
The fund's rapid asset growth also raises secondary concerns. At $12 billion-plus in AUM, DRAM is becoming a meaningful holder of its underlying stocks, particularly the smaller names in the memory supply chain. If a market shock triggers rapid redemptions, the forced selling of relatively illiquid positions could amplify the drawdown beyond what the underlying stocks would experience on their own. This is a generic risk for all thematic ETFs, but it is amplified when a fund grows from zero to $12 billion in 71 days.
For context, the previous record for fastest thematic ETF to $10 billion was held by the ARK Innovation ETF (ARKK) during the pandemic-era growth mania. That fund subsequently lost 75% of its value from peak to trough, a cautionary tale about what happens when thematic momentum reverses. DRAM's underlying fundamentals — genuine supply constraints, contractual commitments from hyperscalers — are arguably stronger than ARKK's were, but the velocity of inflows creates its own fragility.
How to Trade It
The DRAM trade is a pure expression of the thesis that AI memory demand will remain structurally undersupplied. If you believe Micron, SK Hynix, and Samsung will continue to report sold-out capacity and accelerating HBM revenue through 2027, DRAM is the most direct vehicle available. If you believe the AI capex cycle is closer to peaking than the market thinks, DRAM will be one of the fastest ways to lose money in this market.
The next catalyst is Micron's earnings, expected later this month. A strong HBM revenue print and raised guidance would validate DRAM's run and likely push the fund to new highs. A miss or a cautious outlook on memory pricing would test whether the $12 billion in assets sticks around or heads for the exits. Traders holding DRAM should size positions with the understanding that a 30% drawdown is entirely possible in a single bad week, and a 30% rally is equally plausible in a good one. This is a high-conviction, high-volatility trade — exactly the kind that separates the tourists from the professionals.
