SE: Compared to the scale up and scale out of racks in a data center, can clusters help reduce power consumption, or do they use more power and are therefore a part of the AI data center’s power problem? But the software allocation and utilization of those pieces, or the slicing of that hardware cluster into software or virtual groups, is a different problem. To solve the [power] problem we just discussed, you need to know and see the history of previous workloads at run time. Emmer: Clusters don’t solve the power problem by themselves. SE: When it comes to interconnects, I/Os, and data flow, are there different challenges between rack scale up, scale out, and compute clusters, or are they similar?