The second. Because everything in Haskell is immutable, it can take a _lot_ of liberties in terms of letting structures share parts of one another, in the case where one version is derived from the other.

Conversely, in the case where something like a list is modified in entirety (e.g. with a `map` function), if the compiler can determine that the original is no longer needed, it can run the map operation in place - much like you might do on an array in C - avoiding the need for a second copy of the structure in-memory.