Published on: 2026-05-28
The Roundhill Memory ETF crossed $10B in assets roughly seven weeks after its April 2 launch. The record matters because investors are no longer buying only AI compute scarcity; they are buying the memory bottleneck behind it.
The Roundhill Memory ETF has turned a narrow corner of the semiconductor supply chain into the fastest $10B ETF launch on record. Investors are treating memory, one of technology’s most cyclical businesses, as if it has become a durable AI infrastructure tollbooth.
The tension is no longer whether AI needs more memory, but whether today’s pricing power can survive the next supply response. Is DRAM pricing a new bottleneck premium, or the cleanest expression of a cycle near its most profitable point?
DRAM became the fastest ETF to reach $10B after launching on April 2, with reports placing the milestone at 43 to 50 days.
The fund had $11.64B in assets, only 18 holdings and a 99.33% top-10 weight as of May 27. That makes DRAM a concentrated memory-cycle vehicle, not a broad semiconductor ETF.
SK Hynix, Samsung and Micron-linked positions account for roughly three-quarters of the portfolio. The ETF’s performance depends heavily on a small group of memory leaders.
Conventional DRAM prices are expected to rise 58% to 63% quarter over quarter in Q2 2026, while NAND prices are projected to rise 70% to 75%. Pricing power is the core bull signal.
The deciding test is margin durability. SK Hynix’s 72% Q1 operating margin and Micron’s $23.86B fiscal Q2 revenue make the trade powerful, but vulnerable if forward pricing weakens.
Picture Source: Marketwatch
DRAM reached $10B because it launched at the intersection of two shortages: investor access and AI memory supply. Most semiconductor exposure still leans toward GPU designers, while Roundhill’s fund gives investors direct exposure to companies producing HBM, NAND and DRAM, the memory categories now absorbing AI server demand.
Roundhill describes DRAM as the first-ever memory stock ETF and a pure memory exposure vehicle, unlike broader semiconductor funds.
The record was not ordinary thematic ETF adoption. DRAM reportedly crossed $10B in 43 to 50 days, depending on the measurement, after surpassing $1B within its first 10 trading days. That speed shows investors were not only buying a theme; they were building a liquid vehicle for one of the market’s clearest AI scarcity trades.
Liquidity changed the story. Since launch, DRAM averaged $213 million in daily trading volume and more than 11,000 option contracts per day during its first 10 trading days. Once a fund becomes large, options-enabled and active enough for institutional tickets, it stops being only a fund that tracks memory stocks. It becomes a price-discovery instrument for the AI memory bottleneck.
The AI trade began with accelerators, but compute cannot scale without memory bandwidth. High-bandwidth memory sits beside advanced GPUs, server DRAM supports large model workloads, and enterprise SSDs handle the data movement behind AI inference. A faster accelerator loses economic value if the memory stack around it becomes the constraint.
Contract pricing is now validating that constraint. TrendForce expects conventional DRAM prices to rise 58% to 63% quarter over quarter in Q2 2026, while NAND Flash prices are projected to rise 70% to 75%. Those are not routine cycle numbers; they point to allocation pressure as suppliers redirect capacity toward HBM, server DRAM and enterprise SSDs.
That is why the memory trade has moved beyond AI sentiment. The equity rally is being supported by contract prices, customer allocation, and margin expansion, making the next earnings cycle more important than the next ETF inflow number.
DRAM is concentrated by design. The fund has 18 holdings, a 99.33% top-10 weight and large direct or swap-linked positions in SK Hynix, Samsung and Micron. SK Hynix accounts for 25.94%, Samsung for 18.38%, while Micron exposure appears through direct shares and multiple total return swap positions.
The table shows why DRAM behaves less like a broad semiconductor ETF and more like a leveraged signal on memory pricing, Korean HBM leadership and Micron earnings.
| Signal | Latest Data | What It Shows | Why It Matters |
|---|---|---|---|
| ETF adoption | $10B in 43 to 50 days | Record-speed asset gathering | AI capital moved quickly into memory scarcity |
| Fund structure | 18 holdings, 99.33% top-10 weight | Narrow exposure | Returns depend on a small group of memory leaders |
| SK Hynix exposure | 25.94% direct holding | Largest single position | Korean HBM leadership drives the fund’s core thesis |
| Samsung exposure | 18.38% direct holding | Second major anchor | The ETF adds global memory exposure missing from many U.S. chip baskets |
| Micron-linked exposure | 13.35% swap, 9.62% swap, 5.31% common | U.S. memory leverage | Micron earnings are a major transmission channel |
| SK Hynix profitability | 72% Q1 operating margin | Extreme margin power | Sustainability is the real test of the bottleneck trade |
The most important row is not the $10B milestone. It is SK Hynix’s 72% operating margin, because that is where the AI memory thesis either becomes structural or starts to resemble a peak-cycle warning.
The margin debate starts with SK Hynix. The company reported KRW 52.5763T in Q1 revenue, KRW 37.6103T in operating profit and a 72% operating margin, all record levels. That is not just strong memory profitability; it is the kind of margin profile investors usually associate with scarce platforms, not commodity hardware.
Samsung confirms that the margin expansion is industry-wide rather than company-specific. The company reported KRW 133.9T in Q1 revenue and KRW 57.2T in operating profit, both all-time highs, while its Device Solutions division posted KRW 81.7T in revenue and KRW 53.7T in operating profit. The Memory Business set quarterly records as higher average selling prices and AI demand outweighed limited supply.
Micron gives U.S. investors the same signal in a more accessible listing. Fiscal Q2 2026 revenue reached $23.86B, up from $13.64B in the prior quarter and $8.05B a year earlier, while non-GAAP EPS reached $12.20 and operating cash flow rose to $11.90B. That makes Micron a major earnings transmission channel for DRAM, not just another holding in the basket.
Memory cycles usually turn before the income statement looks weak. Customers slow orders, suppliers add capacity, and average selling prices flatten before profits collapse. The next warning sign would not be a bad quarter. It would be a strong quarter with weaker forward pricing.
The sharpest risk is not that DRAM owns memory stocks. The risk is that it owns them after investors have already repriced memory from cyclical hardware into scarce infrastructure. For a fund built around memory scarcity, the first crack would likely appear in contract-price guidance before it appears in reported earnings.
Concentration is the first transmission channel. With 18 holdings and a 99.33% top-10 weight, DRAM would be highly exposed to any reversal in SK Hynix, Samsung or Micron. A narrow portfolio gives investors cleaner upside when memory prices rise, but it also reduces the buffer if one of the core suppliers signals weaker pricing, delayed HBM deliveries or slower server demand.
Cycle risk is the second. DRAM and NAND prices can stop rising before reported profits look weak, because average selling prices usually turn before margins compress. That makes forward contract pricing more important than headline earnings during the next reporting cycle.
Swap exposure adds another layer. DRAM’s exposure to Micron, SK Hynix, and Samsung includes total return swaps, which can give the fund economic exposure without holding every share directly. That structure is useful for access and portfolio construction, but it introduces counterparty and derivative-structure risk beyond ordinary equity ownership.
AI capex is the final pressure point. If hyperscalers slow server orders or delay data center expansion, the demand assumptions behind HBM, server DRAM and enterprise SSD pricing would weaken. The ETF does not need AI demand to collapse for the trade to break; it only needs memory pricing to stop confirming the scarcity thesis.
The Roundhill Memory ETF, ticker DRAM, is an actively managed fund focused on global memory semiconductor companies. It targets businesses tied to HBM, NAND and DRAM, which are critical components in AI servers, data centers and high-performance computing infrastructure.
DRAM reached $10B quickly because it launched when investors were looking beyond GPU designers for the next AI supply constraint. Its exposure to SK Hynix, Samsung and Micron gave the market a concentrated way to trade memory pricing power.
The ETF’s largest positions include SK Hynix, Samsung Electronics, Micron-linked swap positions, Kioxia, Sandisk, Seagate Technology and Western Digital. As of May 27, SK Hynix was the largest holding at 25.94%, followed by Samsung at 18.38%.
Yes. DRAM is more concentrated than a broad semiconductor ETF, with 18 holdings and 99.33% of assets in its top 10 positions. That structure increases upside when memory prices rise, but it also increases downside if the cycle turns.
The ETF could fall if memory prices peak, hyperscaler server orders slow, HBM delivery disappoints, or Micron margins compress. The clearest warning sign would be forward pricing that is falling or weaker margin guidance despite continued ETF inflows.
More inflows would confirm popularity, not durability. The next decisive test is Q2 2026 memory pricing, followed by earnings updates from Micron, SK Hynix, and Samsung. The signal to watch is whether DRAM and NAND price increases flow through to sustained margins and free cash flow, rather than one-quarter earnings spikes.
If prices keep rising while server demand absorbs capacity, DRAM’s record will look less like thematic excess and more like early recognition of a new AI supply constraint. If margins flatten while ETF assets keep growing, the signal changes: flows would be chasing a cycle that has already delivered its easiest gains.
DRAM’s record will matter less than the next margin print, because the ETF has already proved investors can find the bottleneck; the market still has to prove the bottleneck can outlast the cycle.
