Neocloud Brief · neocloud stocks · what is a neocloud · neocloud ETF
DRAM — the memory ETF: owning the HBM supercycle in one basket
The first pure-play memory ETF — 24 holdings, ~+156–161% since its April 2026 launch, and the fastest-growing ETF launch on record. The question isn’t whether the theme is real; it’s whether you can live with the cycle.
Updated 18 Sep 2026: the holdings count on this page was corrected from 17 to 24 after checking the fund’s own page, and a top-ten holdings table was added. See the corrections log.
01The 30-second read
Fast read
DRAM is the world’s first memory-focused ETF — a 17-stock basket of the memory and storage supply chain (Micron, Samsung, SK Hynix, SanDisk, and more), launched April 2, 2026. It’s been a monster: ~+156–161% since launch, ~$26–27B in assets, and the fastest-growing ETF launch on record.
The thesis: every AI GPU is useless without HBM, and memory pricing is in a scarcity supercycle. But this is a cycle trade, not a hold-forever trade — returns are tied to memory pricing, which is notoriously violent, and the 0.65% fee applies on the way down too. You’re not buying a balance sheet; you’re buying the whole cycle in one ticker.
02What DRAM actually is
DRAM (Roundhill Memory ETF) is a diversified fund that owns a focused basket of memory-chip makers supplying the AI HBM and storage boom. Unlike the other three tickers in the Neocloud Brief, it’s not an operating company — no revenue, no debt, no capex plan. It’s a single-symbol way to own the memory supply chain: the HBM leaders (Micron, SK Hynix, Samsung), storage names (SanDisk), and the equipment and materials sub-suppliers around them.
Launched April 2, 2026, the fund became the first pure-play memory ETF and immediately caught fire: the fastest-growing ETF launch on record, pulling in $21B+ of inflows in its first months and reaching ~$25.9B in AUM (per the fund page, Sep 2026). It trades at a 0.65% expense ratio with 24 holdings.
Top holdings — as reported by the fund, 09/20/2026. Weights combine share positions with exposure held via total return swaps.
| Holding | Weight |
|---|---|
| Micron Technology | 25.81% |
| Samsung Electronics | 24.15% |
| SK hynix | 23.57% |
| CXMT (China) | 5.15% |
| SanDisk | 4.97% |
| Seagate Technology | 4.53% |
| Western Digital | 3.60% |
| Kioxia Holdings | 3.33% |
| Nanya Technology | 2.57% |
| Winbond Electronics | 1.17% |
The top three names are roughly three-quarters of the fund. The remaining 14 holdings are smaller suppliers. Note that a Chinese memory maker, CXMT, is now a top-five position — relevant if export-control headlines start moving the sector. A leveraged version also exists: the Roundhill T-REX 2X Long DRAM Daily Target ETF (RAM), which aims for twice the daily return of DRAM.
03The bull case
1 · HBM is the bottleneck of the buildout
- Every GPU cluster needs high-bandwidth memory in proportion to compute, and the memory makers are at effective scarcity
- Structural, and independent of any one neocloud’s balance sheet
2 · The returns speak for themselves
- ~+156–161% since April 2 launch; momentum continued through the fall (~+9% recent quarter)
- 52-week range $26.14–$81.34 shows how violently the trend has moved — in both directions
3 · Diversified where it counts
- 24 holdings across the supply chain — if Micron stumbles, Samsung and SK hynix carry the basket
- The least company-specific way to play AI memory
4 · Massive demand flows
- ~$26–27B AUM in months — institutional adoption locked in early
- Supports liquidity and tight spreads for later entrants
04The bear case
1 · Memory pricing is violent
- DRAM/HBM prices have historically swung by multiples within a couple of years
- +156% since launch is the happy side of that volatility — it works in reverse (2017–18, 2021)
2 · Concentration in a theme
- 24 holdings diversifies names, not ideas — every holding is memory/storage
- In a cycle downturn, all 17 decline together
3 · A fee on a cyclical asset
- 0.65% applies whether the theme is booming or busted
- ETF investors tend to buy hot themes near the top — when the fee hurts most
4 · No fundamental floor
- No balance sheet, backlog, or pricing-power story of its own — it goes wherever memory prices go
- Quarterly swings in memory pricing (and Micron/Samsung/SK Hynix earnings) move it directly
05The numbers
| Metric | Value | Note |
|---|---|---|
| Fund | Roundhill Memory ETF | First pure-play memory ETF |
| Launched | Apr 2, 2026 | |
| Return since inception | +161.51% | Per the fund page, as of Jun 30, 2026 |
| AUM | ~$25.86B | Per the fund page, Sep 2026 |
| Holdings | 24 | Micron, Samsung, SK hynix, CXMT, SanDisk, Seagate, Western Digital, Kioxia |
| Expense ratio | 0.65% | |
| 52-week range | $26.14–$81.34 | Cyclical moves are large |
| Flows (first months) | $21B+ | Per ETF.com reporting |
| Consensus | N/A | No single analyst target for an ETF |
06What could change the story
Bull triggers
- HBM supply stays tight — allocation and pricing power at Micron, SK Hynix, Samsung
- Memory pricing keeps ratcheting up into AI capacity additions
- More AI capex announcements flow through to memory demand
- Upstream suppliers (equipment, materials) showing strength confirms buildout breadth
Bear triggers
- Memory pricing rolls over — watch quarterly DRAM/HBM contract prices and spot trends
- Supply catches up: new HBM capacity faster than AI demand
- A major supplier guides down on pricing, not just volume
- Thematic crowding unwinds — hot launch money leaves quickly
07What’s next
EARNINGS
Micron and SK Hynix reports. Micron is the bellwether — guidance on pricing and HBM mix moves the whole basket.
Memory pricing checks. DRAM/HBM contract prices and spot trends are the cycle’s vital sign.
Supply vs. demand. Whether new HBM capacity lands faster than AI demand decides if the supercycle extends or rolls.
08FAQ
Is DRAM a good way to play AI?
It’s the direct way to play memory: HBM is a hard bottleneck for every GPU cluster. But it’s a cycle trade — returns track memory pricing, which moves in supercycles, not straight lines.
Is DRAM profitable?
Profitability doesn’t apply to an ETF — it owns profitable companies (Micron, Samsung, SK Hynix are deeply profitable this cycle). The relevant question is whether the cycle keeps expanding.
Why is DRAM up so much since launch?
HBM scarcity + AI capex + the first pure-play memory ETF pulling in $21B+ of inflows in months. ~+156–161% since April 2 bundles the pricing supercycle and fund adoption.
Is DRAM a buy?
There’s no analyst target for an ETF — the decision is about the memory cycle, not a company. This page is research, not a recommendation.
Sources & method
- Fund facts and holdings from the official Roundhill Memory ETF page (roundhillinvestments.com/etf/dram, read 2026-09-18): 24 holdings, AUM $25.86B, expense ratio 0.65%, Cboe BZX listing, closing price $59.61, +161.51% since inception as of 6/30/26, top holdings as of 09/20/2026
- Fund facts and performance as reported by watcher sources as of Sep 17–18, 2026 (TipRanks, StockAnalysis, Tickeron, EBC Financial Group / ETF.com); hub brief snapshot Sep 18, 2026
- AUM figures vary by date: ~$26.1B in the hub snapshot; ~$27.4B reported mid-August 2026 — both shown, approximate
- Returns since launch are approximate and point-in-time; 52-week range per TipRanks ($26.14 low, $81.34 high)