> It will not be possible to repair the computers if they fail
This is the part that concerns me the most. Pretty much all DCs have multiple 24/7 staff to deal with hardware failures and equipment swaps... telling a client "you can't access your hardware for 5 years" wouldn't go over too well.
Well, you clearly wouldn't lease dedicated hardware with this model. You just redistribute load amongst data pods when there is a failure, and pull the whole pod from service for refurbishment once it passes your total failure threshold.
Think about it in a cloud model where you aren’t renting a server you know by name. Imagine running something like object storage or FaaS where the cloud provider handles everything behind the scenes and can failover at any time without you seeing more than perhaps a few failed requests.
In that model, hardware failing is just a factor in total overhead cost. If the hardware doesn’t fail immediately it might be cheaper to leave a dead node in the rack than to pay a human to touch it, especially if they’ve already recouped a significant percentage of the purchase cost by the time it fails. Over the life of a server the cost of cooling is enough that a substantial savings will push that breakeven point earlier.
This is the part that concerns me the most. Pretty much all DCs have multiple 24/7 staff to deal with hardware failures and equipment swaps... telling a client "you can't access your hardware for 5 years" wouldn't go over too well.