Korea Storage Research: Samsung's long-term contract "limits the decline but does not limit the rise," and spot prices continue to rebound ahead of the peak season in the fourth quarter.

The demand for AI servers continues to grow, persistently supporting the upward trend in storage prices, while the pricing power of leading storage manufacturers has significantly increased. According to a research report published by Bank of America Merrill Lynch on August 1, Samsung Electronics has incorporated 60% to 70% of its memory sales into long-term agreement (LTA) systems, with contract terms clearly favoring the supply side: the extent of price reductions is restricted, while the potential for price increases is virtually uncapped. Against the backdrop of rising demand for high-end storage driven by AI computing infrastructure and limited supply expansion, Samsung locks in demand from major customers through long-term agreements while retaining the flexibility for price increases. The report forecasts that the spot prices of DRAM and NAND will continue to rebound before the peak season in the fourth quarter. Data from TrendForce shows that the contract price of DRAM rose by approximately 10% month-on-month in July, with a quarterly increase of 30% to 50%; the price of server DRAM continues to reach historic highs. The report suggests that the growth in demand for AI servers, customer inventory replenishment, and pre-stock for new terminal products will continue to support rising storage prices. Samsung is increasing the proportion of long-term agreements to lock in long-term demand from AI customers. Currently, approximately 60% to 70% of Samsung Electronics' memory sales are completed through LTAs, with the contract structure clearly beneficial to the supply side. According to the Bank of America Merrill Lynch research, Samsung's LTA terms restrict the extent of price reductions, with single-quarter decreases typically not exceeding 5%; however, the potential for price increases can reach 10% to 20% or even higher, with no clear upper limit set. Notably, the LTAs signed by Samsung with major U.S. technology companies primarily adopt a five-year rolling model, allowing renewal for the next period just before or after the first year's contract expires, thus establishing a long-term binding relationship. The report believes that this model enhances the revenue certainty of Samsung's storage business while maintaining its flexibility for price increases during periods of tight supply and demand. As demand for AI servers continues to grow, storage manufacturers are locking in demand through long-term agreements while enhancing their pricing control capabilities. Both DRAM and NAND prices are rising, driven by AI and inventory replenishment demands in August. The spot market for storage has recently remained strong. According to data from DRAMeXchange, as of the report's publication, the spot price for 16Gb DDR5 reached $51, an increase of 733% compared to the same period last year; the spot price for 16Gb DDR4 reached $85.2, up 896% year-on-year; and the price for 8Gb DDR4 reached $42.1, up 722% year-on-year. In terms of NAND, the spot price for a 1Tb wafer stood at $26.4, up 3% week-on-week and 415% year-on-year. The report identifies three main factors that will support further increases in storage prices in August: firstly, an increase in orders from downstream customers, with inventory replenishment demand strengthening; secondly, despite ongoing rising storage costs, multiple OEMs still plan to advance new product launches in September and the fourth quarter, driving procurement demand; and thirdly, the level of terminal inventory has significantly decreased, initiating a channel replenishment cycle. In addition, the supply in the spot market remains tight. Due to the time required for storage manufacturers to release production capacity, new market supply is unable to quickly match the growing demand for AI servers, high-end PCs, and smart terminals. For server DRAM, the price of a 64GB DDR5 memory module has surpassed $1480, while the price for a DDR4 memory module has reached $1300, both hitting historic highs. Client SSD prices have doubled since the end of 2025, while the overall price increase for 2025 was only about 35% to 40%. Major cloud service providers continue to ramp up investment in AI, providing long-term support for storage demand. The main driver of rising storage prices still comes from the wave of investment in AI infrastructure. Bank of America Merrill Lynch data indicates that the combined capital expenditure of the five major hyperscale cloud service providers, including Amazon, Microsoft, Alphabet, Meta, and Oracle, is expected to reach $730 billion by 2026, reflecting a year-on-year growth of about 100%; from 2027 to 2028, the relevant capital expenditures are anticipated to exceed $1 trillion annually. Meanwhile, AWS, Azure, and Google Cloud are expected to maintain revenue growth of 35% to 45% over the next few years, continuously supporting investments in AI computing infrastructure. Although some cloud service providers may face pressure on free cash flow during the 2026 to 2027 period, Bank of America Merrill Lynch believes this reflects the long-term commitment of technology giants to AI infrastructure development, which will also continue to drive demand in the storage, advanced packaging, and server supply chains. As the AI capital expenditure cycle continues to progress, this current upward cycle in storage prices is supported by three factors: AI computing power, long-term supply agreements, and a prolonged tight supply and demand situation. This article is sourced from "Wall Street Jiewen," authored by Li Jia, edited by GMTEight: Chen Qiuda.