It's interesting that Amazon is putting almost no RAM in their servers. I find that 16GB/core maximizes GB/$ but they're using 4GB/core.
Totally depends on workload. Some workloads you can max CPU with 256 MB / core. Some workloads you will never max CPU before you run out of ram.
These are compute-optimized. They have others with more memory if that's your requirement.
Their current memory-optimized models go up to 244GB.
I'd really like to see a more a-la-carte option. For example: I have an app that might in the future require a lot of network, a decent but not terribly huge amount of CPU, but very little memory or storage. I'd like to be able to provision an instance with, say, 512MB RAM, 2GB disk, 10 gigabit Ethernet, and CPU optimized for single-threaded performance with only 2-4 cores. Amazon and most other large cloud providers seem to assume that demand for everything increases together, which doesn't apply to every work load.
That's where I think containers come in handy. Assuming you have other services you are managing you can pack them into instances running your main workload. https://aws.amazon.com/blogs/aws/cloud-container-management/ will likely make that very easy to manage, assuming you are willing to go all in with Amazon.