Demand for AI servers continues to grow, persistently supporting upward memory prices, giving top memory manufacturers significantly greater pricing power.

According to a Bank of America Merrill Lynch research report released on August 1, Samsung Electronics has placed 60% to 70% of its memory sales under long-term agreement (LTA) contracts, with terms clearly favoring the supply side: price decreases are capped, while upward price potential is essentially unlimited. Against a backdrop of rising high-end memory demand driven by AI computing buildouts and constrained supply expansion, Samsung is using LTAs to lock in major customer demand while retaining pricing flexibility.

The report predicts that spot prices for both DRAM and NAND will continue to rebound ahead of the Q4 peak season. Data from TrendForce shows that DRAM contract prices rose approximately 10% month-over-month in July, with quarterly gains ranging from 30% to 50%; server DRAM prices continue to hit record highs. The report believes that growing AI server demand, customer inventory replenishment, and new product launches by end-device makers will continue to support upward memory pricing.

Expanding LTA Ratio to Lock in AI Customer Demand

Samsung Electronics now conducts approximately 60% to 70% of its memory sales through LTAs, with a contract structure clearly advantageous to the supplier.

According to the Bank of America Merrill Lynch survey, Samsung's LTA terms limit price declines, with quarterly drops typically not exceeding 5%; however, upward price adjustments can reach 10% to 20% or even higher, with no stated ceiling. Notably, Samsung's LTAs with major US technology companies primarily use a five-year rolling model, where the next contract cycle can be renewed as the first year ends, creating a long-term binding relationship.

The report believes this model enhances revenue certainty for Samsung's memory business while preserving its ability to raise prices during periods of tight supply and demand. As AI server demand continues to grow, memory manufacturers are locking in demand through long-term agreements while strengthening their control over pricing.

DRAM and NAND Prices Rise Together; AI and Restocking Demand Drive August Rally

The memory spot market has remained robust recently.

According to DRAMeXchange data, as of the report's release, the spot price of 16Gb DDR5 reached $51, up 733% year-over-year; the spot price of 16Gb DDR4 reached $85.2, up 896% year-over-year; and the price of 8Gb DDR4 reached $42.1, up 722% year-over-year. On the NAND side, the spot price of 1Tb wafers stood at $26.4, up 3% week-over-week and 415% year-over-year.

The report identifies three main factors supporting continued memory price increases in August: First, increasing orders from downstream customers and strengthening restocking demand; second, despite rising memory costs, multiple OEM manufacturers are still planning new product launches in September and Q4, driving procurement demand; and third, inventory levels at end-device makers have dropped significantly, initiating a channel replenishment cycle.

Additionally, supply remains tight in the spot market. As memory manufacturers require time to ramp up production capacity, new market supply is struggling to quickly match the growth in demand from AI servers, high-end PCs, and smart devices. For server DRAM, the contract price for 64GB DDR5 memory modules has already exceeded $1,480, and DDR4 module contract prices have reached $1,300, both setting new all-time highs. Client SSD prices have doubled since the end of 2025, compared to a full-year increase of only about 35% to 40% in 2025.

Hyperscalers Continue to Boost AI Spending, Providing Long-Term Support for Memory Demand

The core driver of rising memory prices remains the wave of investment in AI infrastructure.

Bank of America Merrill Lynch data shows that the combined capital expenditures of the five largest hyperscale cloud providers—Amazon, Microsoft, Alphabet, Meta, and Oracle—are projected to reach $730 billion in 2026, an increase of approximately 100% year-over-year; from 2027 to 2028, related capital expenditures are expected to surpass $1 trillion annually.

Meanwhile, AWS, Azure, and Google Cloud are expected to maintain 35% to 45% revenue growth in the coming years, continuing to support AI computing infrastructure investment. Although some cloud providers may face temporary free cash flow pressure in 2026 or 2027, Bank of America Merrill Lynch believes this reflects the tech giants' long-term commitment to AI infrastructure construction, which will continue to drive demand for memory, advanced packaging, and the server supply chain.

As the AI capital expenditure cycle progresses, this current memory upcycle is supported by three factors: AI computing power, long-term supply agreements, and persistently tight supply and demand.