Stockfish evaluates the position of the entire board, e.g. mobility of pieces, how much opposing space they control, etc. But still, the evaluation function is hardcoded and weaker than AlphaZero.
By that I just mean Alpha Go is assuming less about board state than StockFish and other conventional chess AIs do. Evidently the two engines have different assessments which is why Stockfish will greedily lap up a sacrifice by Alpha Zero onto to then lose because of it.