Given the hyperscalers' tendency to just leave failed machines in place in the rack because it's not worth the money to even swap it out, for specifically cloud

Given the hyperscalers' tendency to just leave failed machines in place in the rack because it's not worth the money to even swap it out, for specifically cloud type stuff the price point and density may be more relevant than the repairability.