Roundhill Memory ETF
Roundhill Investments
The first US-listed memory pure-play — an actively managed (no tracked index) basket of semiconductor-memory makers spanning HBM, DRAM, NAND/SSD, NOR and HDD, screening for companies with ≥50% of revenue/profit from memory. Adviser-run modified-market-cap weighting with a 25% single-name cap, rebalanced at least quarterly, at a 0.65% expense ratio. Holds direct equity plus total-return swaps (to reach the Korea-listed names that have no US ADRs) backed by a large T-bill collateral sleeve, so raw line-items split a company across legs and sum to >100% gross. On a consolidated economic basis it is effectively ~6 operating names plus cash, dominated by the Samsung / SK Hynix / Micron memory oligopoly (top-3 ~73%, of which the Korea duopoly is ~half).
- Concentrated — Top disclosed positions are 50%+ of the disclosed book.
Top holdings
DRAM, NAND/SSD, HBM; broadest memory line. Korea-listed (no US ADR), held via direct equity + total-return swap
HBM #1 (~50-55% share), DRAM, NAND. Korea-listed, direct equity + swap legs combined
Only US DRAM/HBM/NAND maker; held primarily via total-return swap legs
NAND flash, SSDs, memory cards (post-WD spinoff)
NAND flash / SSD; Tokyo-listed ex-Toshiba memory unit
HDDs / storage after the SanDisk spinoff
Recent moves
Among the fastest asset-gathering ETF launches on record — ~$1B in ~10 trading days, ~$6B in 5 weeks, peaking around ~$20B notional mid-Jun 2026; net assets stand at ~$17.5B (Jun 29). Volatility has spiked: -6.5% on Jun 26 on Samsung/SK Hynix supply-glut fears. A 2x-leveraged sibling (RAM, T-REX/Roundhill) launched Jun 24, 2026, adding reflexive flow risk to the underlying.
Our take
Bull: the cleanest single-ticker bet on the AI memory super-cycle — HBM/DRAM/NAND reported essentially sold out and contracted through 2026, record memory-maker margins (SK Hynix ~72% operating margin), and HBM's ~3x wafer intensity tightening the whole memory market into 2027. Bear: memory is the most cyclical semi product, prone to a brutal ~3-4-year cost-per-bit boom-bust, with China/CXMT commodity-DRAM glut risk flagged for late 2027; the fund is ~73% in three names and ~half in the Korea duopoly accessed via swaps, layering Korea-macro/FX beta and swap-amplified downside; and it is a 3-month-old, crowded, fast-money vehicle with no track record (Morningstar notes nearly every holding trades above fair value). A high-conviction, high-volatility sleeve, not a diversified core holding.
The bull case is real and verifiable, which is what makes this dangerous.
HBM's ~3x wafer intensity is structurally cannibalizing commodity-DRAM capacity, memory is reportedly sold out and contracted into 2027, and the big three are posting record margins (SK Hynix operating margin reported ~70%+ in 2026). But DRAM strips out every diversifying buffer that SMH/SOXX/AIQ carry and hands you undiluted memory beta — the most boom-bust product in the industry, prone to 3-4 year cost-per-bit glut cycles. The generic 'leveraged-ETF volatility-decay' warning is the WRONG critique for DRAM itself (it's unlevered); the right critiques are extreme concentration, Korea macro/FX + swap counterparty exposure, and a live reflexivity risk from the RAM sibling. Our read: a coherent thesis packaged into a fragile, launch-timed, fast-money vehicle whose 3-month track record is a bull-market artifact, not evidence of edge.
Thesis
DRAM is the first US-listed pure-play on semiconductor memory — an actively managed, ~73%-concentrated bet on the Samsung / SK Hynix / Micron oligopoly riding the AI-driven HBM/DRAM super-cycle. The strategy's underlying logic is genuinely sound (memory is the tightest, highest-margin corner of the AI-compute chain right now), but it wraps the single most cyclical product in all of semis into a three-month-old, crowded vehicle that just round-tripped ~40% from its June peak despite record earnings — the textbook signature of a cycle-top trade rather than a validated one.
Actively managed (no tracked index): the adviser screens for companies deriving ≥50% of revenue/profit from memory (HBM, DRAM, NAND, NOR, HDD, embedded), then applies a modified-market-cap weighting with a 25% single-name cap, rebalanced at least quarterly, at a 0.65% expense ratio (verified on the issuer page, Jul 2026). Critically, DRAM is NOT a leveraged or daily-reset fund. It holds direct equity plus total-return swaps — but the swaps exist to REACH the Korea-listed names (Samsung, SK Hynix) that have no clean US ADR and to keep the concentrated book RIC-diversification-compliant, backed by a large T-bill collateral sleeve; they provide access, not leverage. That distinction matters because there IS a leveraged 2x daily-reset sibling — RAM (Roundhill / T-REX), launched Jun 24 2026 — and the two are routinely conflated.
Assessment
- Cleanest single-ticker expression of a thesis that actually checks out on primary data — HBM cannibalizing DRAM wafers, memory contracted into 2027, record oligopoly margins. It isn't a marketing label bolted onto a generic tech basket; the exposure genuinely IS memory.
- The active + swap structure solves a real access problem: Samsung (005930.KS) and SK Hynix (000660.KS), the two dominant memory makers, have no clean US listing, so a US investor cannot easily own ~half the global memory market without a wrapper like this.
- Concentrated exposure to a genuine oligopoly — three players controlling HBM/DRAM/NAND is a structurally better pricing dynamic than the fragmented memory industry of past cycles, which is the strongest argument that 'this cycle is different.'
- 0.65% is reasonable for an active thematic — cheaper than ARK's 0.75-0.88% suite — even if it buys what is effectively a three-name basket.
- 'Volatility decay' is misapplied to DRAM — it is unlevered; the swaps are for access, not leverage. Fund commentary that warns of daily-reset compounding drag on DRAM is describing its sibling RAM. The real DRAM risks (cyclicality, concentration, swap counterparty/financing cost) get obscured by that confusion.
- Memory is the most cyclical product in semis. A pure-play deliberately removes the design/equipment/foundry diversification that lets SMH or SOXX ride out a memory glut — you own the amplitude of the cycle with none of the shock absorbers.
- ~73% in three names and roughly half in the Korea duopoly accessed via swaps stacks Korea macro/FX beta and counterparty/financing risk on top of already-high memory beta. On a consolidated basis it's ~6 operating names plus cash.
- Reflexivity from the RAM 2x sibling: a daily-reset leveraged fund must buy strength and sell weakness in the SAME underlying names every day, mechanically amplifying DRAM's own volatility — a structural feedback loop that did not exist before Jun 24 2026.
- Crowded fast-money: fastest asset-gather on record, and it pulled in ~$8.8B of inflows DURING the ~40% July drawdown. Money chasing a falling knife is a fragility signal, not a floor.
- Three months old with no full-cycle drawdown observed until this month; Morningstar flagged nearly every holding trading above fair value at launch, and there is no evidence the active screen beats a static Samsung/Hynix/Micron basket.
- 2026-07Update, 31 Jul 2026: the inflow is larger than first counted — ~$9.4B since the June peak, more than $10B over the trailing 30 days and ~$2.5B in July's final five sessions, into a -44% drawdown.
Record
The record is a launch-timing artifact, not a demonstrated edge. DRAM peaked around +180% since its Apr 2 2026 inception (issuer +179.84% Jun 22; independent ~+166% Jun 29), then fell ~40% in about three weeks — from a $80.72 peak (Jun 22) to a $48.64 intraday low, closing near the low-$50s on Jul 20 2026 (52-week range $26.14-$81.34). With 73% in three names the fund IS the memory-price cycle; the round trip was driven almost entirely by memory beta, not stock selection. The July selloff is the tell — it accelerated AFTER Samsung reported RECORD results (Jul 7), meaning the market is trading memory as an event-driven cycle top and refusing to reward even excellent prints. Inception landed on the frothiest leg of the cycle, so the headline triple-digit gain describes when the fund launched, not a repeatable process.
- 2026-07Update, 31 Jul 2026: the drawdown ran deeper — the low is the 29 Jul close of $44.85 ($44.40 intraday), -44.4% from the 22 Jun peak close of $80.72 and -45.4% high-to-low. July alone was -31.8%.
Risks & fit
- Memory glut / cost-per-bit reversal into 2027 — the industry's recurring boom-bust, with China/CXMT commodity-DRAM ramp specifically flagged for late 2027.
- Single-name concentration: a stumble at Micron, Samsung, or SK Hynix moves ~a quarter of the fund each.
- Korea FX/macro/policy exposure plus swap counterparty risk and embedded financing cost on the ~half of the book reached synthetically.
- Reflexive drawdown amplification from the RAM 2x sibling's forced daily rebalancing into the same underlying names.
- Crowding/liquidity: a fast-money unwind could overwhelm a book that is really ~6 operating names, exaggerating both the drawdown and any tracking gap to NAV.
- No track record and no evidence of active alpha — a full memory cycle has not yet run through this fund.
Our 'crowded cycle-top' read would be wrong if memory contract prices keep climbing through 2027 with HBM staying sold out, the big three sustain record margins without a capacity glut, AND DRAM's active screen demonstrably beats a static three-name memory basket over a full cycle — in which case the concentration is justified conviction, not fragility. Conversely, the read is confirmed if cost-per-bit rolls over and inventories build into 2027 while the fund stays ~73% concentrated: the record 3-month gain then proves to be launch-timing, and the July drawdown the first installment of a normal memory bust.
A tactical, high-conviction satellite for an investor who already owns a diversified core, holds a specific view on the memory cycle, and can tolerate 40%+ drawdowns and actively monitor a cyclical position. By design it concentrates rather than diversifies, so it behaves as a single-theme sleeve — the opposite of a set-and-forget holding. An investor wanting memory exposure with a buffer gets it more diluted through SOXX / SMH / AIQ; RAM is the identical bet dialed to 2x with genuine daily-reset decay layered on top.
0.65% is fair for an active thematic ETF and undercuts the ARK suite, but two caveats: you're paying active fees for what is functionally a three-name concentrated basket, and the total-return swaps used to hold the Korea names carry embedded financing costs that sit on top of the headline expense ratio and are far less visible.