Here's a similar one: What's the best possible time complexity of a program that plays perfect chess, assuming we could write such a program?
Answer: O(1), since it could be written as one big lookup table from game state to optimal move. To get a nontrivial complexity class, you have to have some parameter that can vary, so e.g. the question "what's the time complexity of a program that plays perfect chess on an n-by-n board?" (assuming a suitable generalization of the chess rules is given) is much harder to answer.