Yea it's more like the opposite, in theory you only need 1 hidden layer to get the same capacity in the network as with more hidden layers, but in practice, since auto encoders, many hidden layers gradually decreasing in size are easier to get to from local energy minimums to globally low energy minimums