Samsung Electronics and SK Hynix: AI Agent-Driven Memory Demand Paradigm Shift and Price Supercycle

The memory industry is undergoing a structural demand transformation powered by Agentic AI, coupled with an unprecedented, AI inference-supported supply constraint, driving a potent price cycle. Samsung Electronics and SK Hynix are positioned not only for significant profit growth from volume and price increases but also for a strategic upgrade from cyclical commodity suppliers to key bottleneck components in AI infrastructure.

The Paradigm Shift: Agentic AI Transfers Bottleneck to Memory

Agentic AI workloads are fundamentally reshaping computing architecture, shifting the primary bottleneck from GPU compute to CPU orchestration and memory bandwidth/capacity. The investment implication is a structural, exponential increase in system-level memory demand that exceeds linear market expectations. Evidence shows AI agents, unlike single-response LLMs, execute multi-step workflows where CPU and memory I/O latency dominates over GPU compute. The need to manage long contexts and KV caches is accelerating a shift from HBM-centric configurations towards tiered architectures combining HBM, DRAM, and NVMe SSDs. Real-market demand is materializing, with Intel and AMD confirming high-core-count server CPUs are sold out, and deployments of standalone CPUs without GPU pairings, such as NVIDIA's Vera CPU for Meta, are scaling.

Unprecedented Price Momentum and Tight Supply

Channel checks point to a dramatic acceleration in memory pricing for 2Q26, far surpassing previous forecasts, indicative of extreme supply tightness and inventory scarcity across the chain. The investment takeaway is that leading suppliers wield exceptional pricing power. Server DRAM DDR5 contract prices are poised to start at a +50% QoQ increase, based on spot transactions and a lack of inventory at both hyperscalers and producers. DDR4 initial hikes are likely in the +30-40% range, with 8Gb chips trading at $18 versus $12 in January. Enterprise SSD pricing could double again in 2Q26. This pricing strength is occurring mid-cycle, with further upside expected.

Major Earnings Revisions Reflect New Reality

Consensus earnings forecasts for memory leaders are undergoing substantial upward revisions, reflecting the strength of the AI-driven cycle and pricing power. This implies significant near-term earnings upside and potential for further consensus catch-up. SK Hynix's 2026-27 EPS estimates have been raised by 24-32%, with its DRAM business projected to achieve an 84% operating margin in 2026. For Samsung Electronics, 2026 smartphone shipment forecasts have been revised from -6% YoY to +5% YoY, driven by record Galaxy S26 pre-orders and share gains in mid/low-end models secured by memory supply access.

Key Risks and Disagreements

The primary risk remains the historical cyclicality of the memory industry. Extremely high prices and margins could eventually trigger capacity expansion, leading to a downturn. Demand could disappoint if macroeconomic conditions worsen or Agentic AI adoption slows. Technologically, competition in the HBM space is intensifying, and capacity additions from Chinese players could erode pricing power over time.

Valuation and Investment Implications

Our revised price targets are SK Hynix at W1,300,000 (implying ~4x 2027e P/E) and Samsung Electronics common shares at W251,000, offering approximately 33% and 37% upside, respectively. The investment preference is for companies with high pricing power in DRAM (Samsung, SK Hynix), HBM, and NAND, as well as semiconductor equipment beneficiaries of rising capex, rather than downstream hardware and consumer-facing companies facing margin pressure.

Appendix: Key Data Summary

Exhibit: DRAM Pricing Forecast Revisions

Exhibit: SK Hynix Key Earnings Revisions