A lot of speculation here. But at least fun to speculate about, given people (i.e. hyperscalers) still need CPUs to feed their accelerators. The post describes working x86 implementations with 16 and ...