SK Hynix pitches rack-level pooled memory for AI data centers

· Tech · South Korea

SK Hynix presented its Memory-Centric AI Rack System, or MARS, at the Open Compute Project Global Summit. The architecture separates memory from individual servers and pools it at the rack level over Compute Express Link fabrics. SK Hynix says this addresses the capacity and cost limits of conventional server designs, where memory is tied to each processor. The proposal marks a shift from processor-centric to memory-centric data center design as AI workloads put growing pressure on memory capacity. CXL pooling lets multiple servers draw on shared memory rather than each carrying its own, which could change how operators plan capacity and spending. The concept is at the industry-consortium stage, not yet a shipping product standard.

Why it matters：Memory pooling at rack scale could reshape data center architecture and procurement, giving memory makers more influence over system design as AI clusters grow.

Key points

  • MARS pools memory at the rack level over CXL fabrics.
  • SK Hynix says it tackles capacity and cost limits of per-server memory.
  • It was presented at the Open Compute Project Global Summit.

Sources (original reporting)

More stories from this day