The High Cost of AI Memory

Summary

Shawn Kim explains how AI demand is driving a structural shift in memory chip prices, which have historically fallen but are now rising sharply. This 'chipflation' is creating supply bottlenecks, impacting PCs, smartphones, and data center costs, though the direct CPI impact is small.

• Memory prices have risen more than sixfold in the past year, breaking decades of price declines.
• AI data center buildouts are the main driver, with servers accounting for 59% of DRAM demand by 2028.
• A two-tier market has emerged: large AI buyers get priority access, while traditional buyers face shortages.
• PC memory demand faces a 15% shortfall in 2027, smartphones 12%, leading to price increases or spec cuts.
• The memory market is expected to grow from $220B in 2025 to $890B in 2026, adding $600B in incremental revenue.
• Direct impact on CPI is modest (~0.1%), but producer prices, corporate margins, and cloud costs are affected.
• HBM usage per AI chip has jumped 7x, and total data center HBM use has grown orders of magnitude.
• Supply relief is slow due to long lead times for new memory capacity.