1 instance on aws...sure hope our availability zone stays up next year.
When hardware was expensive the idea was to make everything as tiny and efficient as possible. Now that hardware is cheap the idea is to make everything as scalable as possible so as to take advantage of all that extra hardware you can now afford.
The next phase of this evolution is to go back to making everything tiny and efficient again as we reach the theoretical limits of horizontal scalability. A good way to "force" such practices is to make your developers use single (or just limited) hardware for everything.
I doubt your boss is coming from this angle but I've actually witnessed this put into practice by forcing a small development team to run their code on Raspberry Pis. It really brings home the idea that, "No, really: Your code is slow and inefficient."
It makes people re-think their architectures and often forces them to divide tasks up into smaller pieces that can be written efficiently. Also, nothing says, "ditch your ridiculously complicated and slow build process" than forcing it to run on embedded hardware.
If your build forces a multi-core embedded system with a gig of RAM to thrash and crash you need to re-think things a bit =)