We're currently replacing 5 year old multi million $ broadcast production infrastructure. It still serves it's original function, our requirements haven't really changed significantly and there's still a low failure rate of hardware. Some low level systems will switch to cloud, but our base PAM and MAM will remain in-house, due to the large amounts of data storage and datarates required to edit hi-res video - Proxy workflows are not always desirable. However there is talk of, and a general industry trend, for these systems to move to cloud solutions - Something I strongly disagree with and a personal current bugbear.
As you mention improvements in code can also significantly improve efficiency, point in hand I'm currently holding off a major release software upgrade of our array used for video transcoding because the hardware is end of life and due the the impact of the upgrade, it may as well be carried out in conjunction with the hardware refresh. This is despite the fact the current software version will run more efficiently on the end of life hardware, especially when running framerate conversion.
I completely understand the scalability of cloud solutions and the real-estate costs of hosting in-house. But I would still argue that the environmental costs of refreshing hardware at such a rate outweighs the energy savings, especially when the generation of electricity via renewable sources is increasing at the rate it is.. Again, like the motor industry, it primarily comes down to economic not environmental reasons.