The demand for AI-driven memory is entering a phase of exponential expansion, while supply growth lags far behind. Nomura Securities posits that the current memory supercycle is fundamentally distinct from any in history. Structural supply-demand imbalances, long-term agreements (LTAs) securing earnings visibility, and the self-reinforcing investment cycle of AI are fundamentally reshaping the valuation logic of the memory industry. On May 15, Nomura significantly raised its price targets for Samsung Electronics and SK Hynix, from 340,000 KRW and 2.34 million KRW to 590,000 KRW and 4.00 million KRW respectively, maintaining a Buy rating for both, implying an upside potential exceeding 118%. Nomura argues that the current approximate 12-month forward P/E ratio of 6x for both companies severely undervalues the sustainability and stability of their earnings. Their risk premium should converge towards that of Taiwan Semiconductor Manufacturing (TSM), rather than continuing to maintain the historically high discount associated with cyclical industries. The core of this revaluation logic lies in the explosive growth of AI inference workloads, which is driving demand for KV cache memory to expand at a multiplicative rate. Nomura forecasts that memory demand could grow by thousands of times over the next five years, while industry supply is expected to increase by only about 5 to 6 times in the same period. This structural widening of the supply-demand gap, combined with binding long-term supply agreements, suggests that the high profitability of memory manufacturers will possess unprecedented sustainability.
**AI Inference Surge Propels Memory Demand into Exponential Growth**
Nomura notes that the memory industry has entered a phase of structural growth since the release of ChatGPT in December 2022. As AI applications shift from training to inference, and with the rapid adoption of Retrieval-Augmented Generation (RAG) and Agentic AI, memory demand is undergoing a qualitative leap. Nomura's central thesis is that the demand for KV cache memory in AI inference scenarios is a multiplicative function of several variables: user base, usage duration, task complexity, inference token consumption, and more. Taking token consumption as an example, a simple Q&A might require only about 30 output tokens, while generating a one-hour video could need approximately 100 million output tokens. With the rise of Agentic AI, per-user token consumption is climbing exponentially. Nomura estimates that memory demand could increase by thousands of times over the next five years. In contrast, constrained by capacity expansion cycles, industry supply is expected to grow only about 5 to 6 times (a CAGR of approximately 30%) in the same period. This indicates that structural undersupply will persist long-term and cannot be fundamentally reversed through software optimization or architectural adjustments. From the perspective of data center capital expenditure, Nomura expects global data center capex to grow from 1.16 trillion USD in 2025 to 6.13 trillion USD in 2030. The proportion of memory within data center capex is projected to rise from 9% in 2025 to 23% in 2030.
**LTAs Lock in Earnings, Eroding the "Cyclical Stock" Thesis**
Nomura believes that traditional investor concerns about the memory industry—specifically whether current LTAs are truly different from historically failed LTA structures—are being disproven by reality. The structure of current LTAs is significantly different from the past: 3 to 5-year minimum contract terms, prepayment arrangements, and capital expenditure support commitments have made contract cancellations substantially more difficult. More importantly, Nomura argues that customer demand for HBM and high-performance memory already exceeds the industry's medium-to-long-term supply capacity. This means the stable earnings outlook for memory manufacturers is increasingly supported by the structural demand environment of undersupply, rather than being dependent on the LTAs themselves. Based on this, Nomura expects Samsung Electronics and SK Hynix to achieve approximately 30% annual revenue and profit growth over the next 3 to 5 years, with profits in 2026 projected to leap 7 to 8 times. This growth trajectory is highly similar to the approximately 30% CAGR expected for Taiwan Semiconductor Manufacturing (TSM) from 2025 to 2028, which currently trades at a 12-month forward P/E ratio of about 20x. Nomura explicitly states that the risk premium implied by Samsung and Hynix's current forward P/E of about 6x (approximately 19% for Samsung, 24% for Hynix) is significantly higher than Taiwan Semiconductor Manufacturing (TSM)'s level of about 6%, and this gap is no longer justified. As market confidence strengthens, and with the potential for South Korea's inclusion in the MSCI Developed Markets Index in June 2026, Nomura expects the valuation discount for the two companies to gradually narrow.
**Samsung: Price Target Raised to 590,000 KRW, Implying 118% Upside**
Nomura raised its price target for Samsung Electronics from 340,000 KRW to 590,000 KRW. The valuation method was upgraded from a P/B ratio of 3.0x to 5.0x, based on a 12-month forward book value per share of 117,669 KRW. The target discount rate was lowered from 17% to 10%, though this remains well above Taiwan Semiconductor Manufacturing (TSM)'s approximate 6%. Regarding earnings forecasts, Nomura expects Samsung's operating profit for 2026-2028 to be 30.7 trillion, 43.2 trillion, and 51.1 trillion KRW respectively, representing year-on-year growth rates of 604%, 41%, and 18%. The memory business is the core driver, with operating margins for DRAM and NAND expected to remain around 71% and 63% respectively. HBM profitability is projected to gradually converge towards the level of commodity DRAM by 2027. Nomura also notes that profit improvement in Samsung's foundry business will be constrained: over 30% of 2026 revenue is expected to be allocated for bonuses, and operating costs at US fabs are about 50% higher than in South Korea, indicating persistent fixed-cost pressures. For shareholder returns, Nomura anticipates Samsung will allocate 50% of its free cash flow for returns, with free cash flow yields projected at 13% and 20% for 2026 and 2027 respectively.
**SK Hynix: Price Target Raised to 4.00 Million KRW, Premium for Pure-Play Status**
Nomura raised its price target for SK Hynix from 2.34 million KRW to 4.00 million KRW. The target P/B ratio was increased from 3.5x to 6.0x, based on a 12-month forward book value per share of 673,248 KRW, implying an upside potential of about 120%. Nomura forecasts SK Hynix's operating profit for 2026-2028 to be 28.1 trillion, 39.4 trillion, and 48.0 trillion KRW respectively, with year-on-year growth rates of 496%, 40%, and 22%. Hynix's ROE is significantly higher than Samsung's, projected at 100%, 73%, and 54% for 2026-2028, primarily benefiting from its higher financial leverage and pure-play memory business model. In the HBM segment, Nomura expects Hynix's HBM ASP to gradually increase from about 12.9 USD/GB in 2026 to about 20.9 USD/GB in 2027. The operating margin for HBM is projected to rise from 63% in 2026 to 72% in 2027, converging towards the level of commodity DRAM. Nomura also points out that as Hynix's cash balance is expected to reach approximately 100 trillion KRW in the first half of 2026, the company may broaden its investor base through an active shareholder return policy and a potential ADR listing in the United States.
Comments