I think most of the CPU efficiencies have been made for general purpose servers and some companies may have transitioned to custom cpus and servers a couple years back. I think what will drive the transition to new cloud/server hardware is drastic transitions in usage, such as the focus on MachineLearning. So whenever there are large leaps made with that hardware, the cloud/server providers will make the transition too. I would not be surprised to see a more rapid cadence than once every 18 months when it comes to ML now.