So how might datacentre automation work? Specially designed servers that can be racked and de-racked by a robot are designed and deployed. These would be similar to blade servers to avoid wiring issues. Robots would take servers from a store and place these into the racks. When a server fails, or the server has a component that fails, the monitoring system would give the robot the row, rack and server positions. The robot would then remove this server and deliver to a workshop for fixing or recycling. Meanwhile the robot would deploy a replacement from stock.
A similar approach (although slightly more complicated) should be possible for drives swaps of storage arrays. Again, some customisation of the design of the IT hardware would make the task easier for the robot.
This should be possible with existing technology levels, e.g. using floor guides (like Amazon's Kiva), and other optical cues.
It might also be possible to combine this idea with my thoughts on datacentres and lasers.