Today, let's talk about DRAM.
DRAM is not an ordinary semiconductor ETF. It is a highly concentrated “AI memory supercycle ETF.”
When U.S. stocks go through a broad pullback, DRAM behaves less like the market and more like a high-leverage chip stock.
Last Friday, DRAM closed at 55.79, down 15.08% in a single day. Its intraday range was 55.38-61.17, and its 52-week range is 26.14-70.15. That tells you this is not the volatility profile of a typical ETF. It behaves more like a high-beta thematic stock.
Last night, because the underlying names rebounded, its price also bounced back to around 60.
1. What exactly is it?
DRAM is the Memory ETF launched by Roundhill. It only listed in early April 2026 and focuses on the global memory supply chain, covering core AI infrastructure bottlenecks such as HBM, DRAM, and NAND.
Roundhill itself defines it as a pure-play exposure to the memory/storage bottleneck in the AI revolution.
Its biggest feature is extreme concentration.
According to public holdings data, SK Hynix is about 27.3%, Samsung Electronics is about 20.18%, Micron-related swaps add up to more than 24%, and the rest includes Kioxia, SanDisk, Seagate, WDC, and others. The top ten holdings account for about 98.37%.
So buying DRAM is not really buying a diversified ETF. It is essentially buying: SK Hynix + Samsung + Micron + the Japanese and U.S. memory supply chain.
It is like an enhanced portfolio of the three global memory giants.
2. Macro: AI is moving from a GPU shortage to a “memory” shortage
Previously, the market paid the most attention to GPUs.
But as AI models get larger and training and inference become more intensive, the real bottleneck keeps spreading outward: GPUs do the compute, HBM feeds the data, DRAM/NAND handle storage and movement, optical communications transport the data, and power supplies the energy.
Without enough HBM and high-speed storage, GPUs end up waiting for data, which lowers compute utilization.
The macro logic for DRAM is very clear: the faster AI data centers expand, the stronger memory demand becomes.
That is also why memory stocks have exploded this year.
The Roundhill Memory ETF once traded at about 8.2 times expected 2026 earnings, while the software ETF IGV traded at 27.7 times. This shows that while the market is chasing the memory cycle, it also believes memory stocks, after their earnings surge, are not valued as aggressively as software.
That is the most attractive part of DRAM: it benefits from AI infrastructure demand, but it is not carrying the most expensive AI software valuations in the traditional sense.
3. Industry logic: the memory supercycle is real, but cyclical stocks remain cyclical
The core formula behind this memory rally is: strong AI demand + tight HBM supply + rising DRAM/NAND prices + capacity discipline from the three giants.
Samsung and SK Hynix together control about 70% of the global DRAM market. Both companies are avoiding aggressive capacity expansion in an effort to reduce the risk of future oversupply. The article also notes that 2026 memory demand is expected to grow 35%, while supply is expected to grow about 23%.
Barron's also mentioned that DRAM and NAND prices are rising sharply. UBS expects certain DRAM contract prices to rise 62% quarter over quarter in Q1 2026, while NAND prices are expected to rise about 40%. At the same time, Micron expects demand to continue exceeding supply in 2026.
This shows the memory supercycle is not pure speculation.
But this is also where the problem lies: memory is always a cyclical industry.
Today there is a shortage; tomorrow capex may expand.
Today prices are surging; tomorrow end customers may suppress demand.
Today the three giants have pricing power; tomorrow new capacity from China, Japan, and the U.S. may gradually enter the market.
DRAM's biggest risk is not that AI demand suddenly disappears. It is that the market has already priced in the idea that “supply will remain tight for the next several years.”
Once that perfect narrative cracks, the stock price becomes very sensitive.
4. Flows: a crowded trade for both retail and institutions
DRAM has not been listed for long, but it has already become popular.
MarketWatch previously reported that DRAM had nearly doubled after listing, with assets of about $6 billion, and that institutions had begun preparing a 2x leveraged version.
The Wall Street Journal also noted that the ETF attracted more than $250 million in retail net inflows in roughly six weeks after listing, with an inflow pace that even exceeded some long-running popular products.
This means it is no longer a niche investment tool. It has become a crowded trade within the AI hardware rally.
The nature of crowded trades is simple: when they rise, capital rushes in together; when they fall, everyone runs at the same time.
So the 15% single-day crash on June 5 was not an accident. Behind it was a simultaneous cooldown in memory stocks, the Korean stock market, AI small caps, and thematic ETFs.
5. Technicals: short-term sentiment is damaged, but the trend is not fully dead
From a technical standpoint, DRAM closed last Friday at 55.79, which is already about a 20% pullback from its 52-week high of 70.15.
That kind of move looks extreme for an ordinary ETF, but for a highly concentrated thematic ETF like DRAM, it is actually fairly normal.
There are three key zones:
First resistance zone: 65-70
This is the previous high area. If DRAM can regain this range, it would show that capital is willing to continue trading the memory supercycle.
First support zone: 52-55
This is near the current price and also the lower edge of the previous strong platform. If DRAM can stabilize here, it may enter high-level consolidation rather than a trend reversal.
Key support zone: 45-48
If U.S. stocks continue to correct and the Korean market continues to de-rate, it would not be strange for DRAM to fall into this range. This is where medium-term risk-reward would start to reappear.
Extreme drawdown zone: 38-42
If the AI hardware sector broadly cools down, or Samsung, Hynix, and Micron keep selling off, DRAM could return to this area.
For an ETF that rose from $26 to $70, that is not unreasonable.
6. Price scenarios
Bullish case: a renewed challenge of $70-80
Conditions:
The U.S. market stops falling;
CPI/PPI no longer pushes rates higher;
Micron, SK Hynix, and Samsung continue to deliver strong results;
HBM and DRAM contract prices continue rising;
AI data center capex is not questioned by the market.
In this scenario, DRAM has a chance to challenge its previous high again. In an extremely strong case, it could look toward 75-80 or even higher.
Base case: high-level consolidation between $50 and $65
This is what I see as the most probable short-term outcome.
The logic is that the memory supercycle has not ended, but the previous rally was too fast; the U.S. stock pullback cooled down high-beta thematic ETFs; capital is unwilling to chase immediately, but it is also unwilling to completely abandon the AI memory theme.
In this case, DRAM is likely to fluctuate between $50 and $65 while waiting for the next round of earnings and price-cycle confirmation.
Bearish case: $40-48
Triggers include:
U.S. stocks continue to de-rate;
The Korean market continues to fall sharply;
SK Hynix and Samsung break trend support;
The market starts questioning HBM pricing power;
Expectations for memory price increases slow;
Thematic ETF flows turn from net inflows to net outflows.
In this scenario, it would not be strange for DRAM to return to $40-48.
That would not mean the industry logic is broken. It would mean the crowded trade is unwinding.
7. Bottom line
DRAM is a very interesting ETF.
Its advantage is that it is pure enough, concentrated enough, and closely tied to the AI memory bottleneck.
Its weakness is also that it is too pure, too concentrated, and too easily tied to the memory cycle and the Korean stock market.
Long term, I remain constructive.
The AI era's demand for HBM, DRAM, and NAND is not a short-term story. It is infrastructure demand.
Medium term, watch memory prices and the earnings of the three giants.
As long as HBM remains in shortage and Samsung, Hynix, and Micron still have pricing power, the DRAM thesis remains intact.
Short term, it is not suitable for chasing highs, but it is worth watching after pullbacks.
Around 55 is already much more comfortable than 70, but it still cannot be called cheap. If it continues pulling back to 45-50, the risk-reward becomes more attractive.
DRAM is the “memory ETF” of the AI era.
What it buys is the pricing power of the memory supercycle.
The logic is strong. But the biggest risk right now is not weak demand. It is that the previous rally was too large, capital is too crowded, and the market is starting to reassess the pricing power of the three giants.
So the best strategy for this ETF is not to chase. It is: observe after big drops, scale in, and do not go all in.
Because the real money in memory cycles is often not made when sentiment is hottest, but when the market starts to doubt the cycle while fundamentals have not yet broken.
Original author: Metamental | 方外之域
X account: @yijiangren
Original post: https://x.com/yijiangren/status/2064323636958466077?s=20
Risk note: This is a collected community viewpoint and does not constitute investment advice. DYOR.