I've had the p2p endpoint enabled on my witness node running on a Mini PC. And the CPU usage has been meager even for that hardware. And I'm also running a VSC node and some apps on it as well.
I'm planning to make a public seed node with a domain, hopefully sometime soon. After the experience gained (and the continuing hived optimizations), I'm much more confident this hardware can handle it well. And I keep learning more docker features, it has some super cool abilities to limit the hardware usage of containers - this has been one of my main concerns in regards to potential DOS attacks.
It seems to me that if we learn and demonstrate how the network can run on meager hardware, this will contribute vastly more to the resilience. It will become accessible (both financially and in terms of required skills) to many more people to host nodes in their locations. And it will cost them very little, both as initial cost and ongoing electricity/maintenance (or necessary upgrades).