The first-ever memory ETF launches, a signal to sell short?

Author: Deep Tide TechFlow

When a trade gets so crowded that it needs a standalone ETF just to accommodate retail investors, smart money has usually already been selling.

On April 2, Roundhill Investments officially launched the world’s first pure-play DRAM semiconductor ETF, ticker $DRAM, using the name of memory modules directly. Its closing price on the first day was $27.76, and after-hours it climbed another 5% to $29.15.

It looks lively. But a few hours later, BTIG issued a cold, hard research note: the listing of the DRAM ETF is precisely a reverse sell signal for memory stocks.

Don’t rush to call it sensational. This Wall Street rule has been verified over and over again.

An ETF’s “holdings sheet”: three giants gobble up three-quarters

First, let’s see what $DRAM actually bought.

This ETF currently holds just 9 stocks—extremely concentrated. Micron, Samsung Electronics, and SK hynix each make up roughly 25% on average, and together they account for nearly three-quarters of the fund’s entire portfolio. The remaining scraps are allocated to Kioxia, SanDisk, Western Digital, Seagate, and other storage companies.

The fee is 0.65%, not exactly cheap. For now, there’s no options trading. To meet RIC (Regulated Investment Company) diversification requirements, the fund had to “make up” the compliance requirement through Total Return Swaps—basically because the holdings are too concentrated, so it had to get through the review using derivatives.

Roundhill CEO Dave Mazza’s rationale is very direct: “Memory is becoming the core of the AI ecosystem.” That checks out. HBM (high-bandwidth memory) really is one of today’s most scarce bottlenecks in AI infrastructure. SK hynix’s HBM market share is over 60%, Micron’s HBM production capacity has already been sold out through the end of 2026, and Samsung is also racing to catch up.

The product logic isn’t the problem—the timing is.

Roundhill’s “kiss of death”: a history of precise reverse indicators

BTIG pulled out the history of Roundhill’s own products, and the picture is pretty grim.

The most classic case is the Roundhill MEME ETF. This fund that tracks retail-favorite meme stocks first launched in December 2021, right at the absolute peak of the Meme stock bubble. After that, the UBS MEME Index crashed by about 80%, and the fund was forced to liquidate in November 2023. Even more striking, it relaunched again in October 2025, just as meme stocks rebounded 100% off the lows. So what happened? After the relaunch, the index fell another ~40%.

Two launches, two times it precisely topped. If you use Roundhill’s product launch date as a reverse indicator to short, the returns would likely be higher than if you bought it.

This isn’t just a Roundhill issue. BTIG points to a broader rule: the launch of theme ETFs often marks the “consensus peak” of a trade.

In October 2021, ProShares launched the first U.S. Bitcoin futures ETF ($BITO). First-day trading volume topped $1 billion, and the entire market cheered. One month later, Bitcoin topped at $69,000 and then crashed 77%.

In November 2017, ProShares launched the EMTY ETF shorting physical retail. As it turned out, the physical retail index rebounded by 50% over the following 9 months.

In January 2008, VanEck launched a Coal ETF (KOL). After that, coal stocks endured a 12-year bear market, plunging 99%. KOL liquidated at the lows in December 2020, and after the liquidation, coal stocks surged 660%.

ETF launch = topping out. ETF liquidation = bottoming out. This pattern keeps showing up, and the underlying logic is simple: when a theme gets hot enough that an ETF issuer thinks “retail will buy it,” the market’s advance is often already nearing its end. ETF issuers are always merchants chasing momentum—they sell packaged Beta, with nothing to do with Alpha.

After a 350% surge—who’s swimming naked?

The warning signals at the data level are already very clear.

Goldman’s TMT memory exposure index surged 350% over the past year, hitting as high as a 400% gain at the February peak—then only after that did the DRAM ETF arrive. Micron’s stock price once deviated from the 200-day moving average by more than 150%. That deviation is beyond what occurred even in the 2000 tech bubble—an extreme level never seen in Micron’s history. BTIG noted that if Micron merely reverted to the 200-day moving average, it would imply a drop of about 30% from current levels.

The frenzy across the memory sector is well documented. EWY (iShares Korea ETF) surged about 140% over the past year, but if you break it down, 84 percentage points of that gain came from just two stocks—Samsung and SK hynix. This “Korea ETF” has effectively become a stand-in for a memory ETF. Samsung is about 27%, SK hynix about 20%, and together they make up nearly half.

And that’s exactly the demand $DRAM is aiming to capture. Over the past year, EWY pulled in $8.3 billion. The main—and really only—reason many investors bought the Korea ETF was to bet on memory. Roundhill has precisely targeted this demand gap.

But “precisely capturing demand” and “precisely stepping on the top” are often impossible to tell apart until after the fact.

The “other side” of a supercycle

To be fair, the bull case logic is strong enough.

Bank of America defines 2026 as a “supercycle similar to the 1990s,” predicting global DRAM revenue growth of 51% and NAND growth of 45%. Goldman forecasts the 2026 HBM market size at $54.6 billion, up 58% year over year. WSTS expects the global semiconductor market to grow by more than 25% in 2026, approaching $975 billion.

Micron’s fiscal 2025 data center revenue jumped 137% to $20.7 billion. HBM production capacity has already been fully sold out for 2026, and capital expenditures are planned at $2.0 billion (up 45% year over year). SK hynix maintains a market share of over 50% in the HBM3E space, and it is the preferred supplier for customized chips by Nvidia and Google.

All of these reflect real industry trends; they’re not tied to hype. AI demand for memory is structural—each GPU generation increases the amount of HBM required by multiples. H100 needs 80GB; by the GB300 NVL72 architecture, it already requires 17.3TB.

So the core contradiction is clear: the memory industry is undoubtedly a good business—but is the “good price” for that business still there?

An analogy: when BITO launched in October 2021, the long-term outlook for Bitcoin was correct. After spot Bitcoin ETFs were approved in 2024, BTC indeed made new highs. But if you bought on the day BITO launched, you had to first endure a 77% drawdown, then wait three years to break even.

Industry trends can be right, but trades can still be wrong. Timing is everything.

Deep Tide’s view: it’s not exactly a death knell—but the alarm bells are absolutely ringing

Our take: the launch of the DRAM ETF has no inherent link to the memory industry topping and collapsing. But it also absolutely cannot be treated as a “go in comfortably” signal. It’s more like an extremely precise mood thermometer. When an industry gets so hot that it needs to issue a dedicated ETF just to feed retail investors’ appetite, it at least indicates three things:

First, the Easy Money phase is over. Of the 350% rise in memory stocks over the past year, the vast majority was valuation expansion—not earnings catching up. Next, memory stocks will need to prove that today’s prices are justified with real growth in fundamentals. There’s extremely little room for error.

Second, the “theme ETF trap” deserves especially high vigilance. Roundhill’s track record is the best teaching material. When an investment theme is packaged into a zero-barrier retail product, it often means institutions are already trimming their positions and retail is the one picking up the tab. Calling it a conspiracy is too heavy; this is the natural ecosystem of capital markets. The incentives of product issuers determine that they will always chase what’s hot and will never try to predict turning points.

Third, the real risk lies in pricing; the industry fundamentals themselves don’t offer much to worry about. Micron being 150% off the 200-day moving average is more exaggerated than during the tech bubble era. Even if AI doubles its demand for memory, a single 30% technical pullback is still well within a reasonable range.

History won’t simply repeat itself, but it always rhymes. After BITO launched, Bitcoin crashed 77%; the MEME ETF topped precisely twice; can $DRAM break this spell?

We can only be certain of one thing: when everyone believes the same trade “can’t lose,” that’s when it’s the most dangerous.