Hm maybe there is something worthwhile here. I implemented a Python version with the same semantics as itertools.product, except only consuming the iterators when absolutely necessary; still only consuming each iterator once, but building its pools incrementally. See https://gist.github.com/sweeneyde/fb6734d7b9f7d17e132c28af9ecb6270 from itertools import count it = LazyProductObject(count(0), count(0), "ab", repeat=3) assert next(it) == (0, 0, "a", 0, 0, "a", 0, 0, "a") assert next(it) == (0, 0, "a", 0, 0, "a", 0, 0, "b") assert next(it) == (0, 0, "a", 0, 0, "a", 0, 1, "a") assert next(it) == (0, 0, "a", 0, 0, "a", 0, 1, "b") assert next(it) == (0, 0, "a", 0, 0, "a", 0, 2, "a") assert next(it) == (0, 0, "a", 0, 0, "a", 0, 2, "b") It looks like it could be made to have a similar tight loop as the current implementation if we decided to re-write it in C.