KEY POINTS
- The Roundhill Memory ETF (DRAM) lost 20% from June 3 to June 5 after Broadcom's fiscal Q3 AI revenue guidance of $16B missed expectations of $17.2B, exposing the fund's 73% concentration in three stocks.
- DRAM had been the fastest ETF launch in history, amassing $6.5 billion in AUM in its first 27 trading days — a pace that exceeded even the iShares Bitcoin Trust.
- Watch whether the fund stabilizes above $55 or continues to bleed as investors reassess concentration risk in thematic ETFs built around a single supply chain.
The Roundhill Memory ETF lost 20% of its value in two trading days. From a close of $69.71 on June 3 to $55.79 on June 5, the fastest ETF launch in history became the most vivid illustration of concentration risk in the 2026 thematic ETF boom. The catalyst was Broadcom's fiscal Q2 earnings report, where CEO Hock Tan guided fiscal Q3 AI semiconductor revenue to $16.0 billion against consensus expectations closer to $17.2 billion and noted that Google might diversify its chip suppliers.
DRAM recovered partially on Monday, closing at $60.52 as the Intel-Google deal rallied the broader chip sector. But the two-day drawdown exposed a structural vulnerability that every investor in the fund needs to understand.
The Concentration Problem
DRAM launched on April 2, 2026, and within 27 trading days had attracted $6.5 billion in assets under management, surpassing the pace set by the iShares Bitcoin Trust (IBIT) in January 2024. By late May, AUM had approached $9.7 billion. The fund's appeal was simple: it offered pure-play exposure to the memory chip companies — Samsung Electronics, SK Hynix, and Micron Technology — that were printing record margins on the back of insatiable AI demand for high-bandwidth memory.
The problem is that those three names accounted for 73.04% of the fund as of May 11. Samsung at 24.99%, SK Hynix at 24.22%, and Micron at 23.83%. Three companies making the same product for the same customers, all subject to the same demand cycle. When Broadcom's guidance miss raised questions about the trajectory of AI chip demand, it did not matter that Broadcom is not a memory company. The signal traveled instantly: if AI chip buyers are diversifying or slowing orders, memory demand follows.
This is the textbook case of what happens when a thematic ETF becomes a leveraged bet on a single supply chain. The fund performed spectacularly on the way up — DRAM gained 151% from its April launch through early June, making it the best-performing ETF of the year by a wide margin. But the same concentration that drove the rally amplified the crash. A diversified semiconductor ETF like SMH, which fell 8% over the same period, absorbed the same shock with a fraction of the damage.
The Broadcom Trigger
Broadcom's actual results were not bad by any conventional measure. Q2 revenue hit a record $22.2 billion, up 48% year-over-year. AI semiconductor revenue of $10.8 billion grew 143% year-over-year. Earnings per share of $2.44 beat consensus by 5%. On paper, those are numbers any company would celebrate.
The disappointment was entirely about forward guidance. Tan's fiscal Q3 AI revenue projection of $16.0 billion implied a deceleration from the pace Wall Street had modeled. More importantly, his comment about Google exploring multiple chip suppliers was interpreted as a signal that the hyperscaler spending cycle — the engine driving the entire AI supply chain — might be broadening in ways that reduce pricing power for incumbent suppliers.
For DRAM's holdings, the implications were immediate. Samsung, SK Hynix, and Micron have been the primary beneficiaries of the AI memory supercycle because high-bandwidth memory (HBM) chips command margins three to four times higher than commodity DRAM. If the AI build-out decelerates or pricing power erodes, those premium margins compress, and the stocks rerate accordingly.
Lessons for Thematic ETF Investors
The DRAM episode is instructive beyond its specific holdings. The thematic ETF industry has exploded in 2026, with over 300 new launches in the first four months alone. Only five thematic ETFs are beating the S&P 500 this year. Three hundred and eighty-eight are not. The average thematic ETF underperforms because extreme concentration amplifies both gains and losses, and the timing of launch typically coincides with peak enthusiasm rather than peak opportunity.
DRAM is not an outlier; it is the archetype. The Corgi Lithography & Semiconductor Photonics ETF (EUV), another thematic fund launched in May, has attracted over $200 million in assets by targeting an even narrower slice of the semiconductor supply chain. The VistaShares AI Supercycle ETF (AIS) is up 119% year-to-date but carries similar concentration risk in a handful of AI infrastructure names.
For traders considering these products, the lesson is straightforward: thematic ETFs are trading instruments, not portfolio building blocks. The entry point matters enormously, and position sizing should reflect the reality that a 20% drawdown in two days is not a tail risk — it is a feature of the product design.
What to Watch
DRAM's recovery to $60.52 on Monday leaves it roughly 13% below its pre-Broadcom high. The fund needs two things to recover further. First, a stabilization narrative from the memory chip companies themselves. Micron reports earnings later this month, and its HBM guidance will be the single most important data point for the fund. Second, the broader semiconductor sector needs to hold the gains from Monday's Intel rally through Wednesday's CPI print.
If both conditions hold, DRAM likely stabilizes in the $58 to $65 range. If either fails, the fund faces the risk of becoming an exit trade as momentum investors who chased the launch-month gains decide the risk-reward no longer justifies the concentration.
