In its Asia Tech report 'Memory – A Small Wrinkle' released on August 6, Morgan Stanley continues to be optimistic about leading Korean memory chipmakers, maintaining its price targets for SK Hynix and Samsung Electronics. The price target for SK Hynix is set at 2.6 million Korean won, and for Samsung Electronics common stock at 381,000 Korean won. Based on the benchmark stock price used in the research report, these imply potential upside of approximately 74% and 65%, respectively.

Such a high implied upside does not mean Morgan Stanley believes memory prices will continue to rise at their previous pace. On the contrary, the report already observes narrowing price increases, rising channel inventory, and new production capacity gradually entering the market. It judges that the memory industry will transition into the late stage of the cycle in the fourth quarter of 2026.

Morgan Stanley remains bullish, primarily betting on three key themes: continued upward revisions in AI capital expenditure, improved profit visibility due to long-term supply agreements, and the recent pullback in memory stock valuations, which already reflects some concerns about a cyclical peak.

Earnings forecasts also reflect this caution. Morgan Stanley raised its 2026 EPS forecast for SK Hynix by 13%, mainly due to including a one-time investment gain of KRW 63.27 trillion in Q2. However, its 2026 operating profit forecast was lowered by 7%. The 2026 EPS forecast for Samsung Electronics was cut by 10%, mainly reflecting weaker consumer businesses such as smartphones.

The adjustments to earnings forecasts for both companies for 2027-2028 are relatively minor. Based on the revised forecasts in the report, SK Hynix's 2027 EPS is expected to grow by about 25%, and Samsung Electronics' by about 49%, roughly corresponding to the report's stated range of 25% to 50%.

Prices Still Rising, But the Memory Cycle Begins to Decelerate

Product price increases have been the most direct driver behind the recent memory stock rally. AI server demand has boosted demand for HBM, server DRAM, and enterprise SSDs, while capacity expansion has struggled to keep pace, driving significant price increases for DRAM and NAND.

Citing TrendForce data, Morgan Stanley notes that the sequential increase in overall DRAM and NAND contract prices peaked in Q1 2026 at 96% and 88%, respectively. It is expected to be 61% and 58% in Q2, but forecasts fall to 16% and 13% in Q3, and further to 6% and 3% in Q4.

The report's latest channel checks indicate early Q3 DRAM contract price settlements are up about 15% sequentially, slightly below the previous expectation of 20%. NAND contract prices are up about 20%. PC DRAM contract prices are expected to rise 15% to 20% sequentially in Q3, a significant slowdown from the 45% to 50% increase in Q2.

Therefore, a more accurate description is not that 'memory prices have peaked,' but that prices are still rising, albeit at a decelerating pace. Morgan Stanley expects the industry to gradually transition into the late stage of the cycle by Q4 2026. At that point, the operating leverage from price increases will weaken, and the potential for further earnings beats will also decrease.

It is important to note that the slowdown in price increases may not be entirely due to improved supply. The report points out that the slowdown in price increases for some consumer DRAM is partly because buyers are nearing their cost tolerance limits. Meanwhile, demand from AI-related customers remains robust, and suppliers are shifting some consumer-grade capacity to products like enterprise SSDs.

Sequential Change in Memory Contract Prices. Sequential increases in DRAM and NAND contract prices are declining quarter-over-quarter from their peaks.

AI Computing Power Still in Short Supply, Cloud Capex Continues to Be Revised Upwards

The first theme supporting Morgan Stanley's continued bullish view is that AI data center computing power demand is still not fully satisfied.

The report cites earnings reports and management commentary from four major US cloud providers. It states that Alphabet and Microsoft's cloud computing demand still exceeds their internal available capacity; Amazon expects capacity to remain insufficient in 2026, with a significant portion of 2027 capacity already booked; Meta anticipates industry computing power supply will remain tight for the foreseeable future.

This information does not prove that all AI investments will ultimately achieve expected returns, but it at least indicates that major cloud providers have not significantly scaled back infrastructure construction. Therefore, Morgan Stanley's cloud capital expenditure tracking model has raised its 2027 year-on-year growth forecast from 14% a month ago to 29%.

This is crucial for memory manufacturers. AI servers require not only GPUs but also HBM, server DRAM, and enterprise SSDs. As long as cloud providers continue to build data centers, memory demand will not be entirely dictated by the traditional consumer electronics cycle of PCs and smartphones.

However, 'structural AI demand' and 'cyclical price adjustments' can coexist. Morgan Stanley's core judgment is precisely that: AI demand may prolong the profit cycle for memory companies but will not eliminate memory price and inventory cycles entirely.

Cloud Provider Capital Expenditure Growth Forecasts. The chart shows the 2027 cloud capex growth forecast being revised up from 14% to 29%.

Long-Term Agreements Improve Order Visibility, But Cannot Lock in Future Profits

The second source of support comes from Long-Term Agreements (LTAs). Compared to traditional quarterly purchasing, multi-year contracts can help suppliers lock in some demand in advance and plan capacity and capital expenditures accordingly.

SK Hynix's official Q2 earnings announcement shows the company has completed LTA negotiations with about 10 key customers and continues discussions with others. Morgan Stanley further summarizes the company's earnings call, stating these agreements typically last about five years, but terms and pricing mechanisms vary by customer and product. Some agreements include deposits, and prices also adjust with market fluctuations.

SK Hynix did not disclose the specific proportion of capacity or revenue covered by LTAs, stating only that it aims to keep agreements at an 'appropriate level' to improve downside protection while reserving capacity for new demand. Therefore, it is not appropriate to describe its LTAs as having already locked in most future revenue.

The specific terms for Samsung Electronics mainly come from Morgan Stanley's summary of its Q2 earnings call. According to the report, Samsung plans to cover 60% to 70% of its capacity with long-term agreements, has reached agreements with five global large-scale data center customers, and has another five in final negotiations. The agreements use a rolling five-year structure, require prepayments from customers, and set price floors for some mainstream products.

The report's table listing potential clients like AWS, Microsoft, Google, Meta, and Oracle is clearly marked as based on media reports and should not be written as officially confirmed counterparties by Samsung.

The significance of LTAs lies mainly in improving demand visibility, supporting capacity investment, and reducing some price volatility, not in completely locking in future profits. The coverage, duration, price adjustment mechanisms, and default protections vary across contracts. If AI construction slows, product specifications change, or market prices adjust significantly, the protection these agreements can provide has limits.

Comparison of LTA Among Major Memory Manufacturers. The chart compares the LTA coverage, duration, and pricing arrangements of Samsung, SK Hynix, Micron, SanDisk, and Kioxia.

High Price Target Bets on Extended Cycle; New Supply to Determine Upside

The market has already begun pricing in a slowdown in memory earnings growth. Morgan Stanley notes that the forward 12-month P/E ratio for DRAM-related stocks typically leads forward 12-month EPS by about two months. Recent valuations have declined significantly, reflecting investors pricing in lower earnings growth.

This also explains why memory companies' current strong earnings may not lead to continued synchronous stock price increases. Investor focus is shifting from 2026 profits to the sustainability in 2027-2028: whether AI capex can continue growing, whether LTAs can withstand a downturn, and whether prices can hold after new capacity comes online.

High profits themselves attract supply. Morgan Stanley points out that current DRAM gross margins, nearing 90%, are at historically abnormal highs. If high returns incentivize leading manufacturers to accelerate expansion or attract new suppliers, current margins may revert toward their long-term averages.

Here, it's important to distinguish between two metrics: the nearly 90% figure discussed in the report refers to DRAM gross margins, while SK Hynix's Q2 76% figure is the company's overall operating margin—they are not directly comparable. SK Hynix officially disclosed Q2 2026 revenue of KRW 79.3187 trillion and operating profit of KRW 60.5426 trillion, resulting in an operating margin of 76%.

Capacity expansion by Chinese manufacturers is also listed as a long-term risk. The report states that new supply from CXMT (ChangXin Memory Technologies) and YMTC (Yangtze Memory Technologies) could weaken the tightness of some products. Among them, CXMT's roadmap includes supplying HBM to the Chinese market as early as 2027. This is Morgan Stanley's description of the company's roadmap and does not equate to Chinese HBM achieving large-scale mass production or being able to immediately replace high-end Korean products.

Furthermore, the report expects new capacity from major manufacturers to start production from 2027-2028. If AI demand continues to grow rapidly, this new supply may not cause immediate oversupply in the short term. However, if cloud provider capex slows, the release of supply could accelerate price declines.

Therefore, the key assumptions behind the 2.6 million won price target are not that memory prices will rise indefinitely, but that AI demand can extend the profit cycle, LTAs can reduce volatility, and recent valuation declines have already priced in a significant portion of cyclical risk. What truly needs verification going forward is whether these long-term agreements can endure a price downturn and whether memory companies can find new drivers for EPS growth beyond 2028.

DRAM Valuation and Earnings Expectations. This chart shows that valuations typically change ahead of earnings expectations. It should not be interpreted as indicating a stable, mechanical causal relationship between the two.