Why would power will become cheaper? Also manufacturing refinement? Sure it happens but not to any meaningful degree these days it’s more about bribing a new process to the same cost/efficiency as your older manufacturing processes.
As far as economy of scales goes it’s also not a given.
Large cloud providers are much more adapt at reusing old hardware for their PAAS and weaker instances.
I’ve had a chance to talk to a few reps from Dell once and while cloud buy a lot of servers they also upgrade them much less often than large enterprises that run their own DC this is because Amazon can always continue to make money on 3-5 year old hardware by shifting it off to cheaper tiers or by running things like SQS or Lambdas on it in the background.
On the other hand if you have a 3-5 year old server and you need more horse power if you manage your own DC or rack space the hosting costs make it economical to keep underperforming hardware.
Not to mention other costs such as support are quite different, if an Amazon server dies they don’t care. Traditional self hosted environments will require extended support because they can’t have the same redundancy as Amazon.
If you delegate a 5 year old server to some back office applications while you use your new ones for client facing apps the old server is still a mission critical box even if it doesn’t directly run revenue generating services.
Basically Amazon buys to grow, on-prem orgs buy to both grow and upgrade.
If you have an organization the size of Amazon that is completely self hosted they will end up buying much more hardware over a decade than Amazon.
As far as economy of scales goes it’s also not a given.
Large cloud providers are much more adapt at reusing old hardware for their PAAS and weaker instances.
I’ve had a chance to talk to a few reps from Dell once and while cloud buy a lot of servers they also upgrade them much less often than large enterprises that run their own DC this is because Amazon can always continue to make money on 3-5 year old hardware by shifting it off to cheaper tiers or by running things like SQS or Lambdas on it in the background.
On the other hand if you have a 3-5 year old server and you need more horse power if you manage your own DC or rack space the hosting costs make it economical to keep underperforming hardware.
Not to mention other costs such as support are quite different, if an Amazon server dies they don’t care. Traditional self hosted environments will require extended support because they can’t have the same redundancy as Amazon.
If you delegate a 5 year old server to some back office applications while you use your new ones for client facing apps the old server is still a mission critical box even if it doesn’t directly run revenue generating services.
Basically Amazon buys to grow, on-prem orgs buy to both grow and upgrade.
If you have an organization the size of Amazon that is completely self hosted they will end up buying much more hardware over a decade than Amazon.