The best explanation I've found for how Haskell's evaluation works is here: http://chimera.labs.oreilly.com/books/1230000000929/ch02.html#sec_par-eval-whnf

On Monday, December 23, 2013, Eduardo Sato wrote:
Hello, guys.
Recently I came across the definition of the function 'forever' on hoogle. I am intrigued that it works.
The recursive definition does make sense to me in a mathematical way, but I can't figure out how it works under the hood in terms of thunks.
To tell you the truth, I don't know how laziness works in general in haskell.
Can someone help me understand how it works in this example, and give some pointers to materials on the subject?
The "tying the knot" article on the wiki is pretty mind bending too. 
-- | @'forever' act@ repeats the action infinitely. 
forever     :: (Monad m) => m a -> m b 
{-# INLINE forever #-}
forever a   = let a' = a >> a' in a'
--
Eduardo Sato