This will depend on the problem; I expect big algorithmic performance improvements in AI since the algorithms are still new, inefficent, and constantly being improved. But maybe not for sorting, fast fourier transforms, or other well-studied basic algorithms?