AMD has taken over MEXT, a memory tiering technology that allows flash storage to appear as DRAM to applications, directly addressing memory bottlenecks in AI workloads. The move integrates a key infrastructure piece into AMD's data center processor roadmap.
Memory bandwidth and capacity constraints are acute pain points for LLM serving and fine-tuning at scale. MEXT integration positions AMD to compete on cost-efficiency metrics that matter to cloud operators running transformer inference.