Morgan Stanley analyst Shawn Kim reports that new memory capacity takes years to build and qualify, creating a two-tier market where large AI and cloud buyers secure priority access through long-term agreements and prepayments. AI memory use is rapidly increasing, with newer AI chips using 7.2 times more HBM than earlier generations, and full systems using 65 times more. Servers are projected to account for 59% of DRAM demand by 2028, up from 37% in 2023, and enterprise SSDs will reach 65% of NAND demand, up from 18%.
Memory supply cannot keep pace with AI demand, forcing a two-tier market where large AI buyers secure capacity via prepayments, leaving traditional buyers to compete for residual supply.