Cloud Providers Lock Up Memory Supply Through 2028 as Nvidia AI Server Demand Deepens DRAM and NAND Shortages

Major cloud providers have already booked all available 2027 memory supply under long-term agreements and are moving early to secure 2028 volumes, supply-chain sources said, as AI data-center expansion tied to Nvidia's next hardware cycle tightens the market. With Nvidia's Vera Rubin accelerators due to ship in the second half of 2026 and HBM4 entering mass production, shortages in DRAM and NAND are expected to intensify through 2027 and into 2028, leaving OEMs and module makers with little room for additional orders.

The scramble is reinforcing memory pricing and diverting capacity toward AI servers, crowding out standard and PC memory that analysts said account for roughly 60% to 70% of the DRAM market. Sources said some AI server buyers are already cutting planned memory configurations to 64GB or 96GB from 128GB to contain costs, underscoring how tighter supply is beginning to slow upgrade cycles.