Why is SVM considered memory efficient in some cases?

SVM is memory efficient at prediction time because only support vectors are stored.

Unlike models storing entire datasets, SVM stores only support vectors and their weights. This sparsity reduces memory footprint during inference.