In the AWS case it probably would have to be transparent to users if they designed their own cpu. I can’t imagine they’d get very far requiring recompilation or

In the AWS case it probably would have to be transparent to users if they designed their own cpu. I can’t imagine they’d get very far requiring recompilation or with a performance impacting translation layer. My main point is that cloud services infrastructure seems to be moving away from open source software hypervisors (eg Xen) on commodity hardware towards opaque custom virtualization on custom hardware. If they wanted to assure their own security against bugs like spectre they would be more likely to design their own CPUs than to use an open source one.