AI Computing Boom Pushes Memory into Seller's Market! BofA: Samsung's Long-Term Contracts Feature 'Limited Downside, Unlimited Upside,' Spot Prices to Keep Rebounding Before Q4

According to Bank of America Merrill Lynch's memory industry research report released last Saturday (August 1), sustained demand growth from AI servers is pushing DRAM and NAND prices into a new upward cycle, with major manufacturers regaining significant pricing power on the supply side.

Samsung, in particular, has incorporated 60% to 70% of its sales into long-term agreement (LTA) frameworks. The contract terms exhibit a distinctly 'supplier-friendly' nature, capping quarterly price reductions at 5% while allowing price hikes of 10% to 20%, or even without an upper limit. For large U.S. tech clients, Samsung employs a five-year rolling renewal model.

Per the latest data from TrendForce, DRAM contract prices rose approximately 10% month-on-month last month and surged 30% to 50% quarter-on-quarter.

In DRAMeXchange quotations, the spot price for 16Gb DDR5 reached $51, up 733% year-on-year; 16Gb DDR4 hit $85.2, up 896% annually; and 1Tb NAND wafers stood at $26.4, up 415% year-over-year. Meanwhile, the contract price for 64GB DDR5 server modules surpassed $1,480, setting a new historical high.

Bank of America anticipates that spot prices will continue to rebound ahead of the peak season in the fourth quarter this year, driven by three factors: customer inventory restocking, OEM preparations for new product launches, and slow ramp-up of production capacity.

The underlying driver is the massive capital expenditure wave from cloud giants. BofA estimates that Amazon, Microsoft, Alphabet, Meta, and Oracle will collectively spend around $730 billion in capital expenditures by 2026, representing a roughly 100% year-on-year increase. From 2027 to 2028, this figure could exceed $1 trillion annually.

Memory is transitioning from a 'cyclical commodity' to a 'must-have infrastructure component for AI.' Original equipment manufacturers are prioritizing new production capacity toward HBM and server DRAM, passively compressing supply for general-purpose memory.

Analysts point out that Samsung’s strategy of locking in volume via LTAs and securing profits through asymmetric terms signals a structural shift in the memory industry—from 'cyclical volatility' to 'price rigidity.' Short-term risks include cloud vendors’ strained free cash flows and the renegotiation of LTAs in 2028. However, this upcycle, fueled by AI compute demand, long-term contract structures, and tight supply-demand fundamentals, is far from over.