Variance is the amount of change in the target estimation with a different training data.

Ideally, this estimate should not vary a lot. But some ML algorithms have high variance than others. In such cases, they are sensitive to the specifics of the training data.