Microsoft and Google Negotiate 3-Year SK Hynix DRAM Deals Amid AI Shortage
- Microsoft and Google are negotiating unprecedented long-term supply agreements (LTAs) with SK Hynix to secure DRAM, the volatile memory used in servers and AI infrastructure.
- The proposed contracts would span three years, starting in 2026.
- The terms being discussed are described as unprecedented in the history of the memory industry.
Microsoft and Google are negotiating unprecedented long-term supply agreements (LTAs) with SK Hynix to secure DRAM, the volatile memory used in servers and AI infrastructure. These negotiations come as the industry faces severe supply shortages driven by the expansion of artificial intelligence investments.
The proposed contracts would span three years, starting in 2026. According to industry sources, the scale of these agreements is rumored to reach tens of trillions of won. The move represents a strategic shift for big tech firms, moving away from opportunistic, short-term purchasing toward a model focused on securing volume.
Unprecedented Contract Terms
The terms being discussed are described as unprecedented in the history of the memory industry. To ensure a steady supply of components, Microsoft and Google are proposing a prepayment scheme where they would provide upfront payments covering between 10% and 30% of the total contract value.
the negotiations include provisions for a minimum price guarantee. This price floor is designed to protect the supplier from significant drops in DRAM unit prices during the three-year contract period, providing stability in a market traditionally characterized by cyclical price volatility.
SK Hynix is reportedly in the final stages of coordinating a specific agreement with Microsoft for server-grade DDR5 memory, a high-performance standard for modern data centers.
The Impact of the AI Supercycle
The urgency to secure these agreements is driven by a critical bottleneck in AI infrastructure. While High Bandwidth Memory (HBM) has captured significant attention, the demand for conventional server DRAM and NAND has also increased significantly as AI shifts from the training phase to inference and distributed architectures.

This demand has led to extreme price fluctuations. For example, DDR4 prices have soared nearly tenfold within a single year. Because priority is often given to HBM and AI-specific memory, the supply of standard server DRAM has tightened, creating a seller-dominated market.
The financial impact of this trend is evident in the performance of memory manufacturers. SK Hynix closed 2025 with record revenues of 97.1467 trillion won and a record operating profit of 47.2063 trillion won. Both SK Hynix and Samsung Electronics are forecast to post record-high earnings for the first quarter of 2026.
Shift in Market Dynamics
Historically, major memory buyers preferred shorter contracts to take advantage of price drops during the semiconductor industry’s natural cycles. However, the explosion of AI infrastructure has disrupted this balance, making memory chips strategic reserve resources rather than mere commodities.
By committing to multi-year deals with upfront payments and price floors, Microsoft and Google are attempting to insulate their supply chains from quarterly market fluctuations and ensure they have the hardware necessary to maintain their AI trajectories.
While these developments have been widely reported by South Korean press and industry sources, neither Microsoft nor Google has officially confirmed the agreements as signed contracts.
