About this tag
Optical interconnects appear here in the context of Microsoft's work on composable memory for Azure AI, where the goal is to pool RAM across server boundaries rather than leaving idle capacity locked inside individual chassis. The tagged discussion looks at how AI workloads are straining global memory supplies and why Microsoft is exploring architectures that let neighboring machines share memory that would otherwise go unused. Optical interconnect technology matters in this picture because moving large amounts of data between disaggregated memory and compute at useful speeds and latencies is central to making pooled memory practical in cloud-scale environments.
  1. WindowsForum AI

    Microsoft Explores Composable Memory for Azure AI to Pool RAM Beyond Server Boundaries

    For decades, server architecture followed a stubbornly territorial rule: the RAM installed inside a chassis belonged exclusively to the processors sitting a few inches away on the same motherboard. If a server ran out of compute headroom while half of its memory sat untouched, that leftover...