A four-node AHV cluster has 512 GB of memory per host available to user VMs. Running VMs are assigned 1,900 GB of memory in total, and none of them uses memory overcommit. A new VM is requested with 256 GB of memory. CPU, storage, and network capacity are ample. Which power-on outcome should you predict?

By default AHV allocates a VM's full configured memory from the physical memory of the host it runs on, and a VM must fit on a single host. In this cluster, total memory for user VMs is 2,048 GB and 1,900 GB is already assigned, leaving about 148 GB free across all four hosts combined, so no single host can supply 256 GB and the power-on is refused. Memory overcommit is an opt-in, per-VM setting that lets AHV reclaim unused guest memory; Prism Central does not switch it on automatically to make a placement succeed. AHV does not oversubscribe memory by default, so the idea that the VM starts with degraded performance is wrong. vCPUs, on the other hand, can be oversubscribed: AHV schedules vCPUs onto physical cores and does not refuse a power-on because the vCPU total exceeds the core count, so a vCPU conflict is not the cause. If HA Guarantee mode were enabled, even more memory would be held back for failover, making the shortfall worse. Exam caveat: on AHV, memory is the hard placement constraint and CPU is the elastic one. Operational check: compare per-host free memory in Prism with the VM's requested memory, then free memory, add hosts, or right-size the VM before retrying.