This relies on two assumptions, that compute won't get cheaper and there won't be large algorithmic improvements, both of which keep getting proven wrong.