The global memory chip industry is undergoing a historic paradigm shift: it is transforming from a traditional cyclical commodity with sharp booms and busts into a highly strategic AI infrastructure resource with strong predictability. The most critical impact lies in the complete overhaul of the valuation framework—a transition from the traditional Price-to-Book (P/B) ratio to the Price-to-Earnings (P/E) ratio. On June 1st, Goldman Sachs stated in its latest in-depth research report on the global semiconductor memory industry that the current memory chip upcycle differs from past ones. AI-driven demand sustainability, constrained supply growth, and structural changes in Long-Term Agreements (LTAs) are driving the memory industry's transition from a highly cyclical commodity track to an AI infrastructure track with predictable earnings. The report indicates that four disruptive changes are occurring in the industry's fundamentals and valuation logic:
First, the supply-demand gap has been comprehensively revised upward. The tightness of supply in the DRAM, NAND, and HBM markets in 2027 will exceed that of 2026, and the shortage is expected to persist into 2028.
Second, a historic shift in the valuation framework is underway, with the industry benchmark formally moving from Price-to-Book (P/B) to Price-to-Earnings (P/E). This shift has driven substantial upward revisions in the price targets for the "top three" memory giants (implying approximately 53% upside for SK Hynix and 60% for Samsung Electronics).
Third, the pricing logic for HBM is being reassessed. The average price of HBM in 2027 is projected to achieve a "catch-up rally" of up to 44% relative to standard DRAM. The global Total Addressable Market (TAM) for HBM in 2027 has been revised upward by 54% to $116 billion.
Finally, the mid-to-long-term operating profit forecasts for the top three companies have been comprehensively raised, with high-profit margins expected to persist throughout the forecast period.
Coincidentally, prior to Goldman Sachs, major Wall Street investment banks Morgan Stanley and J.P. Morgan pointed to the same conclusion in their latest research reports: memory giants like Samsung Electronics and SK Hynix are at a historic inflection point for a valuation paradigm shift. Morgan Stanley and J.P. Morgan believe that the large-scale implementation of Long-Term Agreements (LTAs) is expected to prompt the market to reprice these companies from "strongly cyclical products" to "technology infrastructure with stable cash flow attributes." Currently, the forward P/E ratio for major memory players is only about 7.3x, representing a 50%–80% valuation discount compared to TSMC. Three Structural Anomalies: Why Will This Cycle Remain Elevated Long-Term? The current cycle has completely diverged from the historical trajectory driven solely by cloud data centers in 2017-2018. Goldman Sachs believes the underlying logic of the fundamentals is being reshaped by three structural forces: Demand Side: AI Servers Take Absolute Dominance The cyclical drag from consumer electronics has been completely marginalized. Data shows that in 2025, servers will account for approximately 50% of total industry DRAM demand and 40% of NAND demand; by 2028, these proportions are expected to further climb to 61% and 43%, respectively. The global server memory market size in 2025 (approximately $449 billion) is already 7.4 times that of 2017. As Large Language Models (LLMs) evolve towards Agentic AI, token consumption by 2030 is projected to reach over 24 times current processing capacity. Memory bandwidth and capacity have become core bottlenecks constraining AI development. Supply Side: HBM Capacity "Cannibalization" Intensifies The expansion of traditional memory production is being hard-constrained by physical conditions. HBM production requires 3-4 times the wafer capacity of standard DRAM. As HBM iterates towards HBM4 and HBM4E, the wafer consumption ratio per unit product continues to rise. Between 2026 and 2030, approximately 30% of the combined available cleanroom capacity of the three major manufacturers (about 1.39 million to 1.54 million wafers per month) will be mandatorily allocated for HBM production. This will cause the supply Compound Annual Growth Rate (CAGR) for traditional DRAM to shrink significantly from 19% in 2017-2018 to 15%. Business Model: Long-Term Agreements (LTAs) Reshape Earnings Certainty Memory manufacturers and hyperscale cloud providers are systematically dampening cyclical volatility through LTAs. Clear financial evidence has emerged within the industry: SanDisk's Q3 2026 financial report disclosed that its new business model agreements include $42 billion in Remaining Performance Obligations (RPO) and $400 million in prepayments, coupled with penalty clauses for breaches. The history of the silicon wafer industry indicates that the widespread adoption of LTAs can grant oligopolistic industries extremely high-profit stability. This has become the core foundation supporting the memory sector's entitlement to higher valuation multiples. Supply-Demand Gap Analysis: A More Severe Shortage Looms in 2027 Data indicates that the supply-demand gaps for the three major categories in 2027 are not narrowing but instead worsening compared to 2026:
DRAM: The supply-demand gaps for 2026, 2027, and 2028 are projected to be -5.0%, -5.9%, and -3.9%, respectively. Driven by strong server DRAM demand, the average price growth rate for traditional DRAM in 2026 is forecasted to be as high as 326% year-over-year, with operating profit margins expected to remain at historically extreme levels around 80%.
NAND: The supply-demand gaps are projected at -4.4%, -4.6%, and -3.0%, respectively. Enterprise SSD (eSSD) demand is expected to surge by 66% in 2026 and 31% in 2027, supporting NAND operating profit margins in the high range of around 60%.
HBM: The shortage is most critical, with gaps projected at -5.4%, -6.0%, and -4.3%, respectively. Due to the unexpected rise in ASIC-side demand (172% growth in 2026), Goldman Sachs forecasts the HBM market size for 2026, 2027, and 2028 to reach $56 billion, $116 billion, and $168 billion, respectively.
Abandoning P/B for P/E: A Leap in Price Target Revisions for the "Top Three" Based on a qualitative change in earnings visibility, Goldman Sachs has formally anchored its pricing of memory stocks to P/E multiples (using 9x as a benchmark): SK Hynix (Buy): Price target raised to the range of 3.3 million to 3.5 million KRW. Stress tests show that even under an extreme negative scenario of consecutive 30% price declines for two years, its profit margin could still maintain a healthy level of 40%, thoroughly disproving the old logic of "losses at the peak of the cycle." Samsung Electronics (Buy): Price target raised to 480,000 KRW. Operating profit in 2026 is projected to grow over 8 times year-over-year, with Return on Equity (ROE) reaching a historical high of 52%. Its HBM revenue is expected to surge to approximately $44 billion by 2027. However, it is noteworthy that rising memory prices are negatively impacting downstream sectors. The operating profit margin of Samsung's smartphone division is projected to collapse from 11% to a historical low of 2%. Kioxia (Upgraded to Buy): Under the expectation of a "higher for longer" NAND market, based on a P/E multiple of 7.8x for FY3/28E earnings, the 12-month price target is set at 93,000 JPY. Wall Street Consensus Forms: How Morgan Stanley and J.P. Morgan View the Valuation Framework Shift This shift in valuation paradigm from P/B to P/E is not an isolated judgment by a single institution. The latest analyses from Morgan Stanley and J.P. Morgan also strongly resonate with this view. Morgan Stanley explicitly stated that memory has become the absolute bottleneck for AI infrastructure. Long-Term Agreements (LTAs) are transforming the traditional, strongly cyclical business into one with guaranteed, high-margin, long-term cash flows. If the market continues to price these companies as ordinary cyclical commodities, a severe valuation dislocation will occur. Quantitative calculations show that under a neutral scenario (100% LTA coverage for HBM, 70% coverage for standard memory with a 10x P/E), the implied overall P/E for Samsung and SK Hynix should reach 8.5-8.6x. If the LTA coverage rate increases to 80%, the implied P/E would exceed 10.5x.
J.P. Morgan's logic directly addresses the essence of commercial dynamics: the buyer's fear of supply disruption and the seller's concern about demand cancellation jointly facilitate legally binding long-term agreements. The firm also issued aggressive bullish calls: raising Samsung's price target to 480,000 KRW (corresponding to an 8x P/E), SK Hynix's to 3 million KRW, and doubling Kioxia's target price to 80,000 JPY. It is noteworthy that all three top Wall Street institutions have pointed to TSMC as a reference: after securing long-term agreements with Apple in 2014, TSMC successfully shifted its valuation framework to P/E, maintaining it within a 10-30x range for an extended period.
Memory giants, currently trading at a forward P/E of only about 7.3x, face a historic opportunity to narrow the significant 50%-80% valuation discount relative to TSMC. However, Wall Street simultaneously maintains a final, sobering bottom line: the literal terms of a contract are insufficient to fully immunize against cycles. At the end of the 2017 cycle, forward agreements became worthless within mere months following the collapse in demand. This time, the only irrefutable evidence that can support the new valuation framework is the actual appearance of prepayments and legally locked-in deferred revenue obligations on the balance sheet. Without the protection of genuine cash inflows, all grand narratives about transcending the cycle will remain a mirage.
Comments