Guidance Miss Becomes a Bloodbath: DRAM’s 15% Implosion is At The Center

The Roundhill Memory ETF (CBOE:DRAM) closed Friday, June 5, 2026 at $55.79, down 15% on the day from a prior close of $65.70. Over the two-day window from June 3 through June 5, the fund dropped 20%, going from $69.71 to that Friday close. If you had $10,000 in DRAM at Wednesday’s open, you had roughly $8,000 by Friday’s bell. That is the headline. The interesting part is why a single ETF could move that hard while the VIX sat at 15.40, well inside its normal range.

An ETF that is essentially three stocks in a trench coat

DRAM is the only pure-play memory ETF on the market, and its fact sheet makes clear what that means. As of May 11, 2026, the top three holdings were Samsung Electronics at 24.99%, SK hynix at 24.22%, and Micron at 23.83%. That is 73.04% of the fund in three names, all making the same product for the same customers. The remaining sleeve is more of the same logic: Kioxia at 4.87%, SanDisk at 4.66%, Western Digital at 4.64%, Seagate at 4.49%, and a few smaller Taiwanese names. The expense ratio is 0.65%. There is nowhere to hide inside this wrapper. When the memory complex breaks, the ETF breaks with it, and there is no software ballast or industrial diversifier to soften the move.

On Friday, Micron Technology (NASDAQ:MU | MU Price Prediction) fell 13%, from $996 to $864. SanDisk (NASDAQ:SNDK) dropped 11% to $1,559, and Western Digital (NASDAQ:WDC) fell 11% to $512. SK Hynix dropped roughly 10% in the Seoul session. With Samsung, SK hynix and Micron all giving up roughly a tenth of their value in a single trading day, a 15% print on the ETF was close to arithmetic, not bad luck.

The two-step selloff: a guidance miss, then a hot payroll print

This was a one-two punch built around the AI capex story and the rate story arriving on the same week. On Wednesday, June 3, Broadcom guided fiscal Q3 AI semiconductor revenue to $16.0 billion versus expectations closer to $17.2 billion, and CEO Hock Tan said on the call that Google may use multiple chip suppliers. Broadcom (NASDAQ:AVGO) shed 14% on the week, ending Friday at $386. AVGO is not a DRAM holding, but the read-through was direct. If the single most vocal customer of HBM-adjacent infrastructure is signaling that hyperscaler capex might lean on lower bills of materials, the pricing power that has carried memory pricing through 2026 starts to wobble.

Then on Friday morning, May nonfarm payrolls printed at 172,000 against an 80,000 expectation, and the 2-year Treasury yield jumped to 4.16%. The 10-year sits at 4.47%, in the 93rd percentile of the past 12 months. A surprise that hot revives the conversation about whether the Fed needs to hike rather than cut, and high-multiple semiconductor stocks priced for a long, smooth AI ramp do not enjoy that conversation. The reddit timeline matched the chart, with one of the highest-upvoted posts of the week titled “Blew my account, truly done” on r/wallstreetbets.

What was actually being priced before Wednesday

The thing to keep in mind is how big the run had been. DRAM is up 101% over the past year, even after Friday. Micron is up 203% year to date and 715% over the past year. SanDisk has done the more absurd thing, up 557% year to date. The bull case rested on a story Sanjay Mehrotra has been telling for four quarters: HBM3e sold out, order books stretching into 2027, and structural earnings power that the cyclical memory playbook never accommodated. Micron’s last quarter delivered $13.643 billion in revenue, up 56.6% year over year, with GAAP gross margin at 56.0% versus 38.4% a year earlier, and guided fiscal Q2 to $18.70 billion in revenue and 68% non-GAAP gross margin. Those numbers come from a market that believes HBM is a contracted, semi-custom product more than it is a commodity.

That is also why Friday hurt as much as it did. The minute anyone credible suggests that AI capex is fragmenting across suppliers, or that HBM ASPs might soften, the multiple compression on a stock that printed those margins is mechanical. You only need to believe the marginal buyer was paying for “sold out through 2027.”

What to watch, and why June 24 is the only date that matters

The cleanest test of all of this is on the calendar. Micron reports fiscal Q3 results on June 24, 2026, after the close. That report will say something concrete about HBM contract pricing into the second half, about whether hyperscaler order patterns have changed in the wake of Broadcom’s comments, and about what 2027 looks like when management can no longer wave at “sold out” as the answer. Prediction markets are already pricing the binary. Polymarket has Micron 71% likely to touch $840 in the week of June 8 but also 57% to 58% likely to close above $980 to $1,000 by end of June, which is the market saying it expects the earnings report to settle this.

Two other signals matter alongside the report. The first is any concrete update on Samsung’s HBM3e qualification with NVIDIA, which would change the supplier-share math that Hock Tan hinted at. The second is SK hynix’s commentary on Q3 customer delivery schedules, the cleanest tell on whether multi-year contracts are being renegotiated or extended. Samsung and SK hynix together are 49.21% of DRAM, so anything they say about volumes will move this ETF more than any U.S. macro datapoint.

The honest read here is that DRAM did exactly what a 73%-concentrated thematic fund is built to do. It captured the upside of the HBM trade on the way up, and it captured the downside in a single session when two of the conditions underneath the trade got questioned at the same time. The fund’s expense ratio buys you exposure to a thesis, not diversification away from it. If you believe Mehrotra on June 24 about pricing power and contract structure, the Friday move is a reset inside an ongoing story. If the report comes in soft on HBM ASPs, the Friday move is the start of the multiple-compression conversation that bears have been waiting on since the cycle began. Either way, the ETF is the wrong place to hedge that view and the right place to express it cleanly. The signal is on June 24, and the rest is noise around it.