Better Artificial Intelligence Stock: Arm Holdings vs. SK Hynix

SK Hynix is presented as the preferred AI-semiconductor investment, supported by 46.8% FY2025 revenue growth to KRW97.2 trillion, a 44.2% net margin, and more than 50% share of the high-bandwidth-memory market. Its valuation is materially lower than Arm's, at 5.4x forward P/E and 9.7x sales versus Arm's 125.0x and 57.3x, respectively. Arm posted FY2026 revenue of $4.9 billion, up 22.8%, and is expanding from IP licensing into fabless data-center CPU products, but faces China exposure, customer concentration, and a substantially richer valuation.

Analysis

The relevant relative-value question is not low-multiple memory versus high-multiple IP, but whether HBM supply remains qualification-constrained through 2027. SK Hynix’s earnings power is unusually levered to the HBM3E/HBM4 mix and premium pricing, while MU is the most credible incremental supply response; a successful Micron HBM ramp or Samsung qualification at leading AI-accelerator customers would compress Hynix’s memory premium before it necessarily weakens industry DRAM pricing. The reported quarterly versus annual revenue figures in the source are internally inconsistent, so position sizing should rely on company filings and channel checks rather than the article’s stated financials.

ARM’s strategic move toward complete data-center chip designs creates a second-order conflict with the ecosystem that pays its royalties. Higher-value design wins could raise revenue per server chip over 6-18 months, but they also risk pushing hyperscalers and chip partners toward RISC-V or internally differentiated architectures; the near-term economic benefit is uncertain because reference-design activity can carry materially lower incremental margins than licensing. At ARM’s valuation, even sustained royalty growth may not be enough without evidence that data-center penetration translates into higher royalty rates and not merely unit mix.

The likely immediate catalyst is limited because this is promotional comparison content rather than new fundamental information. Over 1-3 months, watch AI-accelerator shipment guidance, HBM lead times and pricing, and any customer qualification disclosures from MU and Samsung; over 6-18 months, HBM4 yield leadership and capacity discipline determine whether SK Hynix retains scarcity rents. A broad AI capex pause would hurt both, but ARM’s duration-heavy multiple is more exposed to a modest reduction in long-term growth assumptions.

Key Decisions for Investors

• Initiate a 3-6 month relative-value position: long SKHY and short ARM in dollar-neutral sizing. The thesis is HBM scarcity versus ARM multiple vulnerability; target a 15-25% relative return, with a stop if ARM data-center royalty guidance is raised materially while SK Hynix indicates HBM pricing or mix deterioration.
• Use MU as the principal disconfirmation watch item rather than an outright short: reassess SKHY exposure if Micron demonstrates sustained HBM shipment growth, competitive yields, and customer qualification beyond initial volume. That development could narrow SK Hynix’s premium within one to two earnings cycles.
• Avoid treating SKHY’s US listing as a pure substitute for its Korean ordinary shares without checking ADR liquidity, conversion terms, borrow availability, and KRW exposure. If liquidity is inadequate, express the long leg through the Korean listing or a Korea semiconductor vehicle and hedge currency separately.
• For ARM holders, reduce exposure into the next results unless management quantifies data-center design revenue, royalty-rate uplift, and partner economics. A guide that shows licensing growth without corresponding royalty acceleration would be a multiple-compression catalyst despite continued headline AI growth.