Two Ways to Own the AI Memory Trade: Marvell and SK hynix Both Drop 8% on the Same Headline

Marvell Technology, Inc. (Nasdaq: MRVL) and SK hynix Inc. (Nasdaq: SKHY) led the semiconductor selloff Monday morning after Anthropic CEO Dario Amodei published a 3,800-word essay calling for a slower pace of AI capability development. Both stocks fell roughly 8% in premarket trading. Yet the two companies enter this drawdown from opposite ends of the valuation spectrum — and that difference matters far more than the shared headline.

Key Points

• Marvell (MRVL) fell 8.06% premarket to $217.07, erasing Friday’s 4.03% gain that closed the stock at $236.10
• SK hynix (SKHY) dropped 7.66% premarket to $175.50 after closing Friday at $190.07
• Marvell trades at 78.2x trailing earnings; SK hynix trades at 11.2x
• Both carry elevated volatility — MRVL beta of 2.25, SKHY beta of 2.39
• Analysts initiated coverage on both last week: Piper Sandler at Overweight on MRVL ($270 target) and JP Morgan at Overweight on SKHY ($245 target)
• Neither company has revised guidance. The selloff reflects sentiment, not disclosed fundamentals.

The Catalyst Was a Question, Not a Number

Amodei argued Saturday that the industry “must slow the pace at which we improve the capabilities of A.I. models.” OpenAI CEO Sam Altman publicly agreed. Consequently, investors began repricing a risk that had not previously been on the table: what happens to semiconductor demand if the largest AI labs voluntarily decelerate?

Importantly, Bernstein analyst Madison Rezaei characterized the framework as a security-and-governance proposal rather than a capital spending signal. “At this point, it’s not a call for a lowering of capex or stopping model training,” Rezaei wrote. “However, many investors have started to question what happens if training slows.”

That distinction is the entire trade. No customer cut an order. No company cut guidance. The market simply repriced the option on a slowdown — and it charged the highest-multiple names the most.

Marvell: Growth Priced for Perfection

Marvell has been one of the year’s defining AI infrastructure winners. The Wilmington, Delaware-based company designs custom silicon, interconnect products, and optical DSPs that move data inside AI data centers — exactly the plumbing that scales with training cluster buildouts.

The stock has returned 178% year to date, against 11.9% for the S&P 500. Over the trailing twelve months, it has gained more than 255%. Marvell now carries a $212.2 billion market capitalization on $9.45 billion in trailing revenue.

However, that performance leaves little cushion. At 78.2x trailing earnings and $3.02 in trailing EPS, Marvell requires the AI capex cycle to continue compounding. Analysts still see upside — the average twelve-month target sits at $285.30, with Piper Sandler initiating Overweight at $270 on September 10 — but the dispersion is striking. Targets range from $143 to $400, a spread that reveals how little consensus exists on where this business settles.

SK hynix: Scale at a Discount

SK hynix presents a fundamentally different profile. The Icheon-based memory manufacturer supplies DRAM, NAND, and — critically — high-bandwidth memory, the constrained component that AI accelerators cannot function without. Its HBM4 and HBM4E roadmaps have made it a structural beneficiary of the same buildout driving Marvell.

Nevertheless, the market values it at 11.2x trailing earnings on $16.97 in trailing EPS. The company posts an 85.6% profit margin and a 92.7% return on equity, with total debt-to-equity of just 8%. Analysts carry an average target of $248.43 against Monday’s premarket $175.50, and JP Morgan initiated coverage at Overweight with a $245 target on September 10.

The gap between an 11x memory manufacturer and a 78x fabless designer — both selling into the same AI data center — is the most interesting thing on the tape this morning.

What the Divergence Implies

Two stocks falling the same amount on the same headline does not mean they carry the same risk. A multiple compression event treats a 78x stock and an 11x stock very differently, and the premarket move suggests investors sold the theme rather than distinguishing between the businesses.

Investors should also note the context. The Bank of Korea has flagged concentration risk in the semiconductor market and the possibility of demand shifting as new technologies emerge. Meanwhile, Marvell’s longer-term revenue framework and its cost structure face scrutiny from BofA, which has outlined specific risks to the growth case.

What to Watch

Three variables should govern how this resolves:

• Whether any hyperscaler or AI lab actually alters capital expenditure guidance. Until one does, this remains a sentiment move.
• HBM pricing and allocation through year-end, which determines whether SK hynix’s discount is a value opportunity or an accurate read on cyclicality.
• Marvell’s next earnings report, estimated for December 1, and whether custom silicon design wins convert into the revenue trajectory the multiple already assumes.