It’s both, but I think you’re really underestimating the scope of the problem.
First, consider the number of variables the AI needs to account for. Chess has a maximum of 32 pieces, of exactly six different varieties, and a fixed grid of 64 possible positions, with only two players to account for. In Civilization IV, on the biggest map, you have a grid of 10240 squares, populated by ~75 unique units, whose total numbers can number in the hundreds, controlled by up to twelve different players. Plus, the units have a huge number of different abilities and statistics (compared to chess, where the only variable is really how the unit moves, plus two special cases for en passant and castling) which affect both how those units can be employed, and how they need to be reacted to. Plus, there’s at least a dozen different possible terrain modifiers that can be applied to each square, sometimes multiple modifiers per square. So, yeah, it’s a massively more complicated AI problem than you find in chess.
On top of which, people have been designing AI for chess since 1950, and that’s on top of literally centuries of thought and philosophy on how best to play the game. Just on the computer side, that’s sixty years of people working on the same problem using exactly the same parameters, the forefront of which is being done as pure AI research by leading figures in the field working to no particular deadline. By contrast, the only people working on the AI for a Civilization game are employees of Take-Two Interactive, they’re working on a game that’s been around for less than twenty years, and the game fundamentally changes every three or four years, which requires massive revamping of the AI. And they’ve got to get it done before Christmas if they want to get paid.