Author: Rita

The third quarter DRAM contract price increase fell from an expected 20% to around 15%, while memory module inventory rose to 12.5 weeks, and signs of weakened demand in China's consumer electronics emerged. Morgan Stanley acknowledged in its latest storage industry research report released in August that the second derivative of pricing is peaking, and minor cyclical fluctuations have appeared. However, Morgan Stanley's core judgment is that this is merely a normal cyclical slowdown, not a turning point. AI capital expenditures are still accelerating, and the capex growth rate of the four major hyperscale companies has been raised from the previously forecasted 14% to 29%. Long-term agreements are fundamentally changing the profitability visibility and capital return logic of the memory industry. Morgan Stanley maintains an overweight rating for both Samsung and SK Hynix, with target prices of 381,000 KRW and 2,600,000 KRW, implying an upside of 74% to 84%.

Second derivative of pricing peaks, AI demand changes cycle

The early trading of third quarter DRAM contract prices increased by about 15%, lower than Morgan Stanley's previous expectation of 20%. NAND contract prices rose by about 20%. The reason for the narrowing price increase comes from both ends. On the demand side, China's consumer electronics demand has weakened, and PC and smartphone customers have reached their cost tolerance limit, leading to a decline in purchasing urgency. On the supply side, module manufacturers and distributors' inventory has rebounded from a low level to 12.5 weeks, and purchasing strategies are normalizing. The spot price of DDR3 has fallen below contract prices, which is an early signal of weak consumer demand.

However, Morgan Stanley believes these signals do not constitute a trend reversal. AI-related demand remains strong, enterprise-level SSDs are in high demand, and cloud service providers are still actively signing long-term DRAM agreements, targeting coverage of at least 90% of demand. HBM pricing is still under negotiation, with suppliers aiming to flatten HBM profit margins to levels comparable to commodity DRAM, which implies an annual price increase of 50% to 100%. DRAM has become one of the key bottlenecks in AI construction, but the market seems not to believe in the sustainability of the cycle, with current memory stocks trading at about 3 times forward P/E, hardly factoring in growth premiums. Morgan Stanley's judgment is that if investors start to reprice the durability of earnings and capital return prospects, multiple expansions will become a greater driver of returns.

LTA changes industry structure, capital return accelerates

Morgan Stanley views the progress of long-term agreements as the most underestimated structural change in this memory cycle. Samsung explicitly disclosed in its second quarter financial report that it aims to place 60% to 70% of its capacity under long-term agreements, having signed contracts with the world's top five data center customers, with five more customers in the final negotiation stage. The contracts adopt a rolling five-year structure, requiring customers to make advance payments, with about 25% of expected payments already received. Samsung has also set a minimum price floor for mass-market products to limit volatility risk.

SK Hynix has completed LTA negotiations with about 10 customers, with agreements usually lasting around five years, and pricing structures customized to absorb volatility. Micron has finalized 16 customer agreements covering approximately 20% of DRAM capacity and one-third of NAND capacity, spanning from 2026 to 2030, adopting a take-or-pay and non-cancelable structure, with minimum revenue estimated at around $10 billion. SanDisk has signed 10 new business model agreements with a minimum contract revenue of approximately $9.39 billion.

Morgan Stanley believes that if the market fully values LTAs, the trading multiples of memory stocks should be higher. Investors are currently reluctant to assign significant value until these agreements prove their resilience during the industry's downturn cycle. However, LTAs have moved from the intention phase to specific disclosure, with every major supplier quantifying the percentage of output or revenue to be included in contracts, and multiple companies disclosing margin and advance payments. As LTAs validate their downside protection in the next cycle, the durability of profitability in the memory industry may face repricing.

Samsung and SK Hynix valuations are attractive

Morgan Stanley maintains an overweight rating on both Samsung and SK Hynix. Samsung's target price is 381,000 KRW, based on a residual income valuation model, implying an anticipated price-to-book ratio of about 2 times in 2027, consistent with the peak levels of the industry's commodity cycle. Morgan Stanley expects Samsung's HBM to improve in the second half of 2026 and may gain market share, with positive trends in commodity pricing, allocating 50% of free cash flow for capital returns. The current stock price corresponds to an anticipated P/E ratio of about 5 times for 2026, and about 3 times for 2027.

SK Hynix's target price is 2,600,000 KRW, implying an upside of about 74%, with an anticipated P/E ratio of about 5.8 times in 2027. AI inference demand will support the commodity cycle to remain strong from 2026 to 2027, with a solid HBM leadership position. The earnings per share forecast for 2026 has been raised by 13%, mainly reflecting second quarter asset disposal gains, with forecasts for 2027 to 2028 remaining mostly unchanged. The target price for SK Hynix corresponds to an anticipated price-to-book ratio of about 2.6 times in 2027.

Morgan Stanley's sensitivity analysis shows that the market has not yet assigned any meaningful P/E premium to memory profitability and free cash flow supported by LTAs. The current valuation is at a historical low, with Samsung and SK Hynix's NTM price-to-book ratios at approximately 1.4 times and 1.3 times, far below the peak of over 2 times during the commodity cycle. If the ROIC story of AI capital expenditure continues to validate, and LTAs demonstrate their value in the next down cycle, memory stocks could face a reshaping of the valuation framework.

Disclaimer

This article is a compilation and interpretation of a third-party brokerage research report (Morgan Stanley, August 6, 2026) by Trend Research, combined with publicly available market information. The ratings, target prices, profit forecasts, and related judgments cited in the article are the opinions of the brokerage's analysts and only represent their institution's position, not the opinions of Trend Research, nor do they constitute any investment advice.

Markets have risks, decisions should be independent. This article should not be used as the basis for buying or selling any securities.