Bryan Kelly later explains that "Another challenge is they're remotely accessed. When a workload wants to access its keys that are stored in HSM, it has to do a TLS connection to the HSM, ask it to perform an operation with the key, and then return the data. That creates latency… so, it becomes impractical for some workloads." This would mean that all cloud VMs benefit from these HSMs; however, for larger-scale HPC workloads that are latency-sensitive, Microsoft must employ an entirely different security approach. Additionally, open-source projects like Caliptra RoT 2.0 are also integrated across the server vertical, providing an extra layer of security protection. Finally, CPUs and GPUs have their own trusted execution environments (TEEs), which are the final point of security for VMs and shared hardware. By the time an attack vector is loaded onto the system, it must pass several layers of protection and key checks before any execution happens. At the scale most CSPs operate today, it is necessary to ensure maximum security of sensitive user data. Keep in mind that hyperscalers like Azure work with governments, so all of this is a good selling point for their systems.
