Do you have any specifics in mind? I think compute-as-a-service is already here. Public cloud vendors such AWS and Azure are first and foremost vendors of infrastructure-as-a-service, which is inherently built on top of virtual computing, virtual networking and virtual storage. So in a sense, they already offer some form of compute-as-a-service. For example, AWS offers instance types P5*, which can include a few H100 or H200 GPUs. Going forward, they could have some form of whole computer types such as whole rack with Hopper or Blackwell GPUs.