Microsoft bets on composable memory as AI memory supply tightens

Microsoft is exploring composable memory that pools capacity across servers and assigns it to workloads on demand. Principal architect Samir Rajadnya said the approach aims to use scarce memory more efficiently across Azure's AI infrastructure.

• Composable memory is pooled across servers and assigned on demand
• Microsoft is scouting system-level technologies for Azure
• Interconnects — copper or optical — are key as memory leaves the chassis
• The work will be showcased at the Open Compute Project Global Summit