According to a Bank of America Merrill Lynch survey, Samsung Electronics has placed 60%–70% of its memory sales under long-term agreements with terms clearly favoring suppliers: price reductions are capped (no more than 5% per quarter), while upside pricing flexibility is essentially unlimited. Against the backdrop of surging AI computing demand and constrained capacity expansion, Samsung is using these long-term contracts to lock in major customers while preserving pricing power. Spot prices for DRAM and NAND have continued to rebound ahead of the fourth-quarter peak season, supported by both AI-related capital expenditures and inventory restocking cycles.
Sustained growth in AI server demand continues to support upward pressure on memory prices, significantly enhancing leading memory manufacturers’ pricing power.
According to a research report released by Bank of America Merrill Lynch on August 1, $Samsung Electronics (005930.KR)$ has already placed 60% to 70% of its memory sales under long-term agreements (LTAs), with contract terms clearly favoring suppliers: price reductions are capped, while upside potential is largely uncapped. Against the backdrop of surging demand for high-end memory driven by AI infrastructure build-outs and constrained supply-side capacity expansion, Samsung has locked in demand from major customers through long-term agreements while retaining pricing flexibility on the upside.
The report forecasts that DRAM and NAND spot prices will continue to rebound ahead of the fourth-quarter peak season. According to TrendForce data, DRAM contract prices rose approximately 10% month-over-month in July, with quarterly increases reaching 30% to 50%; server DRAM prices have continued to hit record highs. The report notes that growing AI server demand, customer inventory restocking, and new product launches at the end-user level will further underpin upward momentum in memory prices.
Samsung Expands Long-Term Agreement Share to Lock in Long-Term AI Customer Demand
Samsung Electronics currently executes approximately 60% to 70% of its memory sales through LTAs, with contract structures clearly advantageous to suppliers.
Per Bank of America Merrill Lynch’s research, Samsung’s LTA terms cap price reductions—typically limiting quarterly declines to no more than 5%—while allowing price increases of 10% to 20% or even higher, with no explicit upper limit set.
Notably, Samsung’s LTAs with major U.S. technology companies predominantly follow a five-year rolling structure, enabling renewal of the next cycle shortly before or after the initial year expires, thereby establishing long-term binding relationships.
The report suggests this model enhances revenue certainty for Samsung’s memory business while preserving its ability to raise prices during periods of tight supply-demand balance. As AI server demand continues to grow, memory manufacturers are increasingly using long-term agreements to secure customer demand and strengthen pricing control.
DRAM and NAND Prices Rise in Tandem, Driven by AI Demand and Inventory Restocking in August
The memory spot market has remained robust recently.
According to DRAMeXchange data, as of the report's release date, the spot price for 16Gb DDR5 reached USD 51, up 733% year-over-year (YoY); the spot price for 16Gb DDR4 reached USD 85.2, up 896% YoY; and the spot price for 8Gb DDR4 reached USD 42.1, up 722% YoY. On the NAND side, the spot price for 1Tb wafers was reported at USD 26.4, up 3% week-over-week (WoW) and 415% YoY.
The report attributes the continued increase in memory prices in August primarily to three factors: First, downstream customer orders have increased, and inventory replenishment demand is strengthening. Second, despite persistently rising memory costs, several OEMs still plan to launch new products in September and the fourth quarter, driving procurement demand. Third, end-market inventory levels have declined significantly, and channel restocking cycles are now commencing.
Furthermore, supply in the spot market remains tight. As memory manufacturers require time to ramp up capacity, newly available supply cannot quickly match the growing demand from AI servers, high-end PCs, and smart devices.
In the server DRAM segment, the contract price for 64GB DDR5 memory modules has surpassed USD 1,480, while DDR4 module contract prices have reached USD 1,300—both setting new historical highs. Client SSD prices have doubled since the end of 2025, compared with a full-year increase of only approximately 35% to 40% in 2025.
Hyperscale cloud providers continue to intensify investments in AI, providing long-term support for memory demand.
The core driver behind rising memory prices remains the wave of AI infrastructure investment.
According to Bank of America Merrill Lynch data, the combined capital expenditures of the five major hyperscale cloud providers—Amazon, Microsoft, Alphabet, Meta, and Oracle—are projected to reach USD 730 billion in 2026, an increase of approximately 100% YoY. From 2027 to 2028, annual capital expenditures are expected to exceed USD 1 trillion.
Meanwhile, AWS, Azure, and Google Cloud are expected to sustain revenue growth of 35% to 45% over the coming years, continuing to underpin investment in AI computing infrastructure.
Although some cloud providers may face temporary pressures between 2026 and 2027,free cash flowBank of America Merrill Lynch believes this reflects tech giants' long-term commitment to AI infrastructure investment and will continue to drive demand across the storage, advanced packaging, and server supply chains.
As the AI-driven capital expenditure cycle continues, the current memory upcycle is jointly supported by three key factors: AI computing demand, long-term supply agreements, and persistently tight supply-demand dynamics.
