DRAM Drops 14% as Broadcom's Guidance Resets the AI Chip Trade

The Catalyst: Broadcom Set the Bar Too High for Itself

This is a Broadcom move, not a DRAM-specific one. After the close on June 3, Broadcom reported $22.2 billion in record quarterly revenue with AI chip revenue of $10.8 billion, up 143% year over year — by any normal standard, a blowout. The market didn't care. The problem was the guide: Broadcom projected $16 billion in Q3 AI semiconductor revenue against roughly $17.2 billion expected, and pointedly held its full-year AI target unchanged rather than raising it the way a momentum-priced stock needs to.

Two comments on the call made it worse. CEO Hock Tan acknowledged Google would likely source custom chips from multiple suppliers and warned that the surge in AI silicon was pressuring gross margins. Broadcom shares fell roughly 13-14% on June 4 — its worst single session in over a year. When the most-watched custom-silicon name says 'growth is fine but not accelerating,' every adjacent AI semiconductor trade gets remarked at once.

Why It Hit DRAM So Hard

The Roundhill Memory ETF is concentrated by design, and that concentration cut both ways here. SK Hynix, Samsung and Micron together make up roughly 73% of the basket, so when the memory complex sells off, DRAM doesn't diversify the risk — it amplifies it. On June 5, SK Hynix dropped more than 8% and Samsung fell more than 5% in Seoul as the Broadcom-linked selloff swept Korean chip names. On the U.S. side, Micron extended a sharp slide into Friday after tumbling 7.7% to around $996 the prior session.

The leverage on the underlying names matched the perp action almost exactly: Korean leveraged memory ETFs plunged about 14% as Samsung and SK Hynix tumbled. DRAM's 14.44% move on Hyperliquid isn't an outlier or a basis dislocation — it's the perp tracking a top-heavy basket through a coordinated two-day repricing across both Korean and U.S. memory holdings.

The Setup: First Real Test of a Parabolic Run

Context matters here. This ETF has gained roughly 125% since launching on April 2, and the perp hit a 52-week high of $70.15 on June 2 — three days before this print. So the move off the top isn't a thesis-breaker; it's the first genuine drawdown in a trade that went nearly vertical, meeting an exogenous catalyst with zero cushion. Price action tells that story: from a $65.70 prior close, DRAM traded a wide $56.86–$61.16 range before settling near $57.25.

Nothing in the Broadcom report changed the structural memory bull case — HBM demand, tight DRAM supply, the data-center buildout. What changed is positioning. A basket priced for continuous upside revisions just got a reminder that even 143% growth can disappoint when expectations run ahead of it. Whether $57 holds or the basket retests its pre-parabola base depends less on memory fundamentals and more on whether the broader AI-chip complex is done resetting after Broadcom.