The general thought is that computers still have problems with calculating long-term plans and positional considerations correctly.
There are many examples of modern computers getting "confused" in closed positions where long-term planning is worth more than brute force thinking. Many have a contempt factor which forces them to make a poor move rather than accept a draw against a weaker opponent. Nakamura has exploited this algorithm in many famous games.
For positional considerations, the reason for this weakness is the rarity of possibilities. Except for some lines in the Ruy Lopez, Alekhine, and the Caro-Kann, capturing toward the center is the norm. There are exceptions where capturing away from the center is better, mostly for attacking purposes. Programming in the exceptions would cause an increase in the size of the program and slow down the speed. However the computer does search far enough ahead the this is becoming less of a concern.