DRAM Emerges as Core AI Data-Center Bottleneck and Major Spending Driver

DRAM temporarily stores data needed for AI model training and inference, and its speed, capacity and bandwidth directly affect GPU computing efficiency. As models and data-center clusters grow, memory has shifted from a supporting component to core infrastructure and could determine the cost and pace of cloud providers' expansion.

Gavin Baker, an investment manager who was an early SpaceX investor, said DRAM has become the most critical underlying bottleneck to AI performance. He estimates that hyperscale cloud providers will spend 30% to 40% of their capital expenditure directly on DRAM by 2027, suggesting memory could become one of the largest data-center expense categories.