The memory chip industry is witnessing an unprecedented shift: buyers are now proactively seeking long-term contracts with suppliers. This role reversal stems from the structural surge in memory demand driven by the AI computing arms race. Cloud service providers are realizing that DRAM and NAND are no longer readily available commodities but critical components of AI infrastructure. To secure essential capacity for the coming years, they are compelled to offer advance payments and long-term commitments—known as Long-Term Agreements (LTAs). This change in business models is central to the latest research report by JPMorgan semiconductor analyst Jay Kwon. As LTAs account for a growing share of memory manufacturers' shipments, the traditional price-to-book (P/B) valuation method, used for decades, has become outdated. A shift to the price-to-earnings (P/E) framework is now essential. Based on this logic, JPMorgan raised its price target for Samsung Electronics to 480,000 Korean won (applying an 8x P/E multiple to projected 2026-2027 EPS), significantly increased SK Hynix's target from 1.8 million to 3 million won, and doubled Kioxia's target from 38,000 to 80,000 Japanese yen, making it the highest in the market. The underlying driver of this revaluation is a supply-demand gap that is difficult to reverse. Estimates indicate that even under aggressive capacity expansion assumptions, AI memory supply from 2026 to 2030 will fall short of cloud providers' demand—a deficit equivalent to approximately 450,000 wafers per month. With buyers having no alternatives, suppliers have gained genuine pricing power for the first time. The memory industry is transitioning from a "build-and-wait" model to a "pay-first, produce-later" approach, resembling the foundry industry's "make-to-order" model. The profit visibility brought by LTAs is key to shedding the "highly cyclical" label. Recent EPS upgrades are just the beginning; the fundamental transformation driven by these long-term contracts will ultimately determine the stock prices of memory giants. A 450,000-wafer-per-month deficit has completely shifted bargaining power. Historically, every price increase in the memory industry has been short-lived, as expanded production inevitably pushed prices back down. This time, however, constraints arise from both supply and demand. On the supply side, major memory manufacturers, having just emerged from a severe downturn, are wary of expanding capacity. Without guaranteed orders, no one is willing to commit significant capital amid uncertain marginal returns. Kioxia's capital expenditure as a percentage of revenue is only 5% this year, compared to an average of over 20% in the past five years. This restraint is not coincidental but a collective industry choice. Samsung and SK Hynix are also keeping their capital expenditure ratios in the mid-single digits over the next two years. On the demand side, growth is accelerating. Generative AI is exponentially increasing server memory throughput requirements, while the proliferation of agentic AI is driving explosive demand for enterprise SSDs. Kioxia's average selling price (ASP) surged over 100% quarter-on-quarter, while overall DRAM prices have tripled over the past year. The supply-demand squeeze has left cloud providers with little leverage in negotiations. Micron and others have already disclosed LTA progress, with one U.S. NAND manufacturer signing five LTA contracts covering over one-third of its bit demand for fiscal 2027. The industry expects Samsung and SK Hynix to soon announce the largest LTA deals in semiconductor history. Valuation System Overhaul: From "Cyclical Stocks" to "Foundry-Like," LTAs Reshape Pricing Power For over a decade, the capital market has anchored memory chip valuations to the price-to-book (P/B) ratio. The reasons are straightforward: highly homogeneous products, weak bargaining power, significant capital expenditure, and earnings heavily influenced by macroeconomic cycles. In the AI era, this logic is being completely overturned. JPMorgan's report presents a forward-looking view: the memory industry is undergoing a structural shift "from cyclical to secular." The key catalyst is the widespread adoption of Long-Term Agreements (LTAs). As AI servers drive explosive demand for HBM (High Bandwidth Memory) and enterprise SSDs, cloud service providers (CSPs) are proactively signing 3- to 5-year LTAs with memory manufacturers to ensure supply chain security. JPMorgan notes multiple LTA cases in the industry and anticipates that Samsung and SK Hynix will soon disclose the largest LTA contracts in semiconductor history. Some LTAs not only lock in prices (or set price floors) but also include advance payment mechanisms. This "make-to-order" model increasingly resembles the foundry model of companies like TSMC. The enhanced visibility and certainty of profits are prompting Wall Street to shift memory stock valuations from traditional P/B to price-to-earnings (P/E) ratios. JPMorgan asserts that LTAs are the core factor driving industry re-rating, applying an 8x P/E multiple to projected 2026-2027 earnings (historically 6x during downturns). Divergence Among Giants: SK Hynix Leads, Samsung's Financial Firepower, Kioxia Breaks Out While the industry is trending upward, the competitive landscape is quietly shifting. Samsung: After a catch-up rally, special dividends are the final piece for valuation completion. Samsung's stock has surged 143% this year, but its lag in HBM technology remains a weakness. Its recent improvements in HBM validation are more a result of industry tailwinds than a re-establishment of technological moats. The financial logic is clearer. Blended ASPs for DRAM and NAND are expected to rise 293% and 234% year-over-year by the end of 2026, with operating profit forecasts revised upward by 2% to 11%. However, due to rising labor costs from strikes and tax pressures, EPS expectations have only been adjusted by 1% to 5%. A larger catalyst lies in cash distribution. From 2024 to 2026, Samsung's available funds for shareholder returns are nearly 160 trillion won, with over 115 trillion won available for special returns after regular dividends. The market favors direct dividends over share buybacks, with expectations for special dividends after Q3 2026 already priced in. If management fails to deliver, the valuation transformation will lack its final piece. SK Hynix: Up over 200% this year, but a 31% EPS CAGR indicates further upside. SK Hynix has been the standout performer in this rally, with its price target jumping from 1.8 million to 3 million won. This is supported by a significant improvement in profit quality—projected EPS CAGR of 31% from 2026 to 2028. ASP negotiations for HBM are accelerating, and the increasing proportion of LTAs not only stabilizes pricing but also gives Hynix a strong position in negotiations with cloud providers. Even with new capacity plans, overall DRAM bit supply growth from 2027 to 2028 is expected to remain below 20%, sustaining a high-margin supply-demand imbalance. The cash return path is equally clear: from 2025 to 2027, free cash flow of 240 trillion won will be fully distributed to investors (14 trillion in 2025, 66 trillion in 2026, and 160 trillion in 2027), with total shareholder return yields of 5% and 12% in 2026 and 2027, respectively. Kioxia: After a 326% surge, a shareholder return roadmap at the investor day will determine if the valuation ceiling can be broken. Kioxia's upward trajectory is straightforward: Q4 ASP surged over 100% quarter-on-quarter, operating profit quadrupled, and capital expenditure fell to 5% of revenue. High-margin SSDs and storage products now account for about 60% of total sales, up from 50%, with marginal contribution rates expected to remain at 85% to 90% until fiscal 2027. LTAs are already in place—Kioxia secured mid-term contracts covering 2027 to 2028 in Q4. Combined with restrained capacity expansion, 17 trillion yen in free cash flow will be released over the next three years. The current 7x P/E multiple yields an 80,000-yen target price, still a 10% discount to the memory industry's 15-year historical average of 8x. This discount can only be eliminated with one trigger: a clear, impactful shareholder return plan at the investor day. The unresolved expectation of divestment by the Bain Capital-led consortium is the biggest external variable suppressing valuation premiums. Generating profits is only the first step; how profits are distributed to shareholders will determine whether the 326% surge can continue. Boundaries of the New Framework: Can the Industry Truly Escape "Cycles"? This logic is not without risks. The greatest risk lies at the start of the chain: if LTA negotiations lack transparency or if cloud providers' AI capital expenditure significantly declines, the P/E valuation framework built on "profit certainty" would lose its foundation. Specifically: First, the ultimate test of AI monetization. JPMorgan pointedly notes that investors remain concerned about a "disconnect between AI capital expenditure competition and semiconductor ecosystem profitability." If cloud providers' AI models fail to deliver commensurate commercial returns or if underlying technology breakthroughs reduce AI server memory consumption, the currently inflated demand expectations could shrink. Second, the "contractual integrity" of LTAs remains untested. Historically, during the 2017 DRAM shortage, manufacturers signed forward purchase agreements, but these were easily abandoned in subsequent downturns as demand slowed and inventories rose. This time, whether memory manufacturers can strengthen contract enforcement through mechanisms like third-party financial derivatives will determine the viability of the "escape from cyclicality" thesis. Finally, structural shifts in global capacity cannot be ignored. Beyond traditional giants' restrained expansion on new nodes like BiCS 8/10, global supply chain diversification is accelerating. The gradual rise and release of emerging market capacities (such as domestic NAND production in China) are diversifying the global memory supply structure. In the foreseeable future, this regional capacity increase will test the existing global supply-demand balance, adding more variables to the industry's competitive landscape. JPMorgan believes that AI has indeed provided the memory industry with new "armor," and the adoption of LTAs has widened the industry's moat like never before. However, cycles may be extended or smoothed but are unlikely to be eliminated entirely. Amid the fervent capital frenzy, only those who maintain technological leadership and exercise restraint in capacity expansion will truly enjoy the long-term compounding benefits of transitioning from "cyclical stocks" to "growth stocks."