Most of the algorithms we use in machine learning are chosen specifically because they parallelize well. If we could increase clock speeds arbitrarily high, a huge selection of algorithms we passed on would suddenly become viable.