Samir Rajadnya, Microsoft | theCUBE + NYSE Wired: AI Data Pipeline Forum 2026
summary
Microsoft's pathfinding team scouts new memory and storage tech and claims composable memory is the lever that turns scarce DRAM into usable capacity by dynamically reallocating memory across servers and racks, cutting stranded memory and improving Azure efficiency. The talk names optical interconnects as the enabler for rack-to-rack low-power, low-latency memory access, flags KV cache and disaggregated serving as immediate scaling pain points, and pushes vertical integration (workload→system→software→silicon) as a pragmatic way enterprises can optimize without hyperscaler budgets.
AI-written summary · about
chapters (7)
- 0:00OCP and Microsoft: Advancing AI Infrastructure
- 2:27Data Center Architecture at Azure Scale
- 4:59Composable Architecture and Dynamic Memory Allocation
- 7:32Interconnects Enable Flexible AI Systems
- 10:48Disaggregated Memory for Efficient AI Inference
- 14:52Workload-Driven Architecture and Vertical Integration
- 18:00OCP Collaboration and Future Computing Architectures
