I imagine you get that 96 if you stick together 4x Xeon CPUs with 24 cores each onto one server.

https://www.intel.com/content/www/us/en/products/processors/...

You only need 2 processors, since each cores gives you 2 vCPUs. "For the n1 series of machine types, a virtual CPU is implemented as a single hardware hyper-thread" --https://cloud.google.com/compute/docs/machine-types

so the bare metal might still have 4, such that it could host 2 of these VMs at once I guess.

There aren't quad socket Skylake yet :). You can also see that the largest public single package is 28 cores (56 threads). I'll sadly have to let you figure out how many because silly.

Disclosure: I work on Google Cloud.

Quad socket 24 core would fit. Skylake has up to 28cores and 8 socket support.

Likely quad socket 24 core with 768GB of memory to accommodate 2 regular instances or 1 memory + 1 compute instance.