the article is incorrect for memory, google cloud allows you to create custom configurations. e.g. 1 cpu and 100 gb of memory. That itself allows us to cut the cost 10% just because predefined instances doesn't match our memory and cpu footprint.