Better Artificial Intelligence Stock: Advanced Micro Devices vs. SK Hynix

SK Hynix is favored over AMD based on its lower valuation, leading roughly 50% HBM market share, and stronger profitability: FY2025 revenue rose 46.8% to KRW97.2 trillion, with a 44.2% net margin and KRW18.2 trillion in free cash flow. AMD delivered FY2025 revenue growth of 34.3% to $34.6 billion and $4.3 billion in net income, but trades at a substantially higher 36.0x forward P/E versus SK Hynix's 5.4x. The article highlights SK Hynix's reported Q2 revenue of KRW79.3 trillion, up 51% from Q1, as evidence of accelerating AI-memory demand.

Analysis

The relevant distinction is not “compute versus memory,” but where AI-system bottlenecks sit in the next hardware refresh cycle. HBM content per accelerator and increasingly complex memory packaging make SK Hynix’s earnings unusually leveraged to accelerator unit growth and HBM pricing, while AMD must still convert design wins into sustained deployment volumes against NVDA’s software and networking lock-in. This favors SKHY over AMD for the next 1-3 quarters if hyperscaler capex remains robust, but the Hynix multiple discount is partly compensation for memory’s historically violent profit-cycle reversals.

The key second-order beneficiary is TSM: both AMD GPU ramps and HBM-linked advanced packaging demand increase pressure on leading-edge wafer and CoWoS capacity. The more important competitive risk to SK Hynix is not conventional DRAM weakness initially, but Samsung and Micron (MU) qualifying comparable HBM generations at major accelerator customers; that would shift the market from supply-constrained pricing to customer-driven allocation. Conversely, AMD’s upside requires evidence that accelerator revenue is incremental rather than replacing CPU spend, and that gross margin holds as it uses pricing to gain share.

The article’s valuation comparison should not be accepted at face value: cross-border accounting, ADR liquidity, cycle-peak forward earnings, and inconsistent reported quarterly figures can materially distort a simple P/E comparison. Near term, a newly accessible ADR can attract passive and retail flows, but this is not a durable catalyst without verified HBM contract visibility and capacity guidance. The contrarian setup is that SKHY may be cheap on peak margins rather than structurally mispriced; memory pricing or inventory normalization can compress earnings faster than a headline multiple suggests.

Key Decisions for Investors

• Initiate a 1-3 month relative-value position: long SKHY / short AMD, dollar-neutral, only after confirming SKHY ADR average daily liquidity and borrow availability. Target 10-15% relative outperformance; stop if AMD reports accelerator revenue/gross-margin upside while SK Hynix signals HBM pricing or shipment growth below plan.
• Add TSM on pullbacks as the higher-quality picks-and-shovels expression of simultaneous AMD and AI-memory demand. Use a 6-12 month horizon; thesis fails if leading-edge utilization or advanced-packaging capacity guidance is cut, not merely on one quarter of customer digestion.
• Do not short NVDA solely as a hedge to long SKHY: NVDA captures system-level software, networking, and platform economics that memory suppliers do not. If a semiconductor hedge is required, AMD is the cleaner relative short because its valuation depends more heavily on unproven AI share capture.
• Set an event watch on Samsung and MU HBM qualification announcements, HBM contract-price commentary, and hyperscaler capex guidance over the next two earnings cycles. A credible multi-supplier HBM ramp is the trigger to reduce SKHY exposure; absent that evidence, HBM tightness can support estimates despite broader commodity-memory volatility.